TruthArchive.ai - Tweets Saved By @MilkRoadAI
@MilkRoadAI - Milk Road AI
Dario Amodei is the CEO of Anthropic, just laid out the single biggest financial risk in AI right now. It's not whether the technology works. He's pretty confident it will. The risk is whether the money comes back fast enough to justify what's being spent. Building AI data centers costs tens of billions of dollars and takes one to two years to finish. So right now, he has to decide how much compute to buy for infrastructure that won't even be ready until 2027. He's placing a massive bet today on what revenue will look like two years from now. Anthropic has been growing at roughly 10x per year. They went from about $1 billion in early 2025 to around $9 billion by end of 2025 to $14 billion annualized as of February 2026. That growth is insane but he can't just assume it keeps going at that pace forever. If revenue keeps growing 10x a year it would hit $100 billion by end of 2026 and $1 trillion by end of 2027. If he bought a trillion dollars worth of compute based on that assumption and revenue came in at even $800 billion instead, there is no hedge on earth that saves him from bankruptcy. Being off by just 20% when you've committed that much capital is fatal. If the growth rate slows to 5x instead of 10x, or the timeline shifts by just one year, same result. You're basically done. Even if AI becomes genius level in the lab, turning that into actual revenue takes time. He uses the example of disease. AI might discover cures for everything, but you still have to manufacture the drug, run clinical trials, get regulatory approval and distribute it globally. COVID vaccines took a year and a half to reach everyone even with the entire world in a panic. Polio has had a vaccine for 50 years and still hasn't been fully eradicated. The technology being ready and the revenue actually showing up are two very different timelines. So what does he do? He deliberately under buys. He commits to hundreds of billions in infrastructure, not trillions. He accepts the risk that if demand explodes he won't have enough capacity. But he'd rather leave money on the table than bet the entire company on a growth curve that might be off by a year. He also says some of the other AI companies are just throwing money around without doing the math. Committing $100 billion here, $100 billion there, without actually modeling what happens if revenue comes in below expectations. He calls it YOLOing. For context, Big Tech is expected to spend around $625 billion on AI infrastructure in 2026 alone. AI services are only generating about $25 billion in actual revenue against all of that. That's roughly a 4% return on what's being invested. The gap between what's being spent and what's being earned right now is massive. The CEO of one of the top AI companies on the planet is saying out loud that the financial math might not work for a lot of these players. The technology is real. The demand is probably coming. But if the revenue wave shows up even one or two years late, some of the biggest companies in the world are going to be sitting on historic losses. This is the same dynamic that wiped out telecom companies in the early 2000s. They built the infrastructure for demand that eventually came, but it came too late to save the companies that built it. He's basically saying he's trying not to be one of those companies.
Video Transcript AI Summary
Speaker 0 discusses the uncertainty around how fast AI will translate into revenue, noting that even if technology advances quickly, misjudging the pace can be ruinous due to the way data centers are purchased. They reference a concept from Machines of Loving Grace, suggesting we might see a powerful AI country in the data center by 2026 or 2027, and acknowledge a possible one- or two-year error in that hunch.
They pose a question: if AI can cure all diseases, how long would it take to deliver cures for everyone? They explain that biological discovery, drug manufacturing, and regulatory processes (citing vaccines during COVID) create delay, such as the vaccine rollout taking about a year and a half. They ask how long from the lab-created AI to actual universal cures, noting polio vaccines have existed for fifty years and eradication remains difficult in remote regions, with the Gates Foundation and others trying to overcome this.
The speaker asserts that while economic diffusion may not be as difficult as eliminating polio, there are real limits. They outline their expected acceleration curve: a 10x year-over-year revenue increase. At the start of the year, revenue pace is $10 billion annualized; given the time needed to build and reserve data centers, they ask how much compute to buy for 2027. If revenue grows at 10x annually, it could imply $100 billion in 2026 and $1 trillion by the end of 2027, leading to a potential purchase of about $5 trillion in compute starting in 2027 (a trillion dollars per year for five years).
They caution that if revenue is not a trillion dollars, no force could prevent bankruptcy from such a purchase. Thus, they acknowledge risk: either the growth rate remains 10x, slows to 5x, or revenue fails to reach the projected level. They emphasize the need to balance ambitious compute procurement with financial risk, rather than a reckless “YOLO” approach. They observe that some other companies may be acting without fully understanding the risks or performing thorough financial scrutiny. The core message is to behave responsibly, aligning compute investments with anticipated revenue growth and recognizing the potential consequences of overextension.
Full Transcript
Speaker 0: And so we have this uncertainty, is even if the technology goes as fast as I suspect that it will, we don't know exactly how fast it's going to drive revenue. We know it's coming, but with the way you buy these data centers, if you're off by a couple years, that can be ruinous. It is just like how I wrote, you know, in Machines of Loving Grace, I said, look, I think we might get this powerful AI, this country of genius in the data center. That description you gave comes from the Machines of Loving Grace. I said, we'll get that 2026, maybe 2027 again.
That is my hunch. Wouldn't be surprised if I'm off by a year or two, but that is my hunch. Let's say that happens. That's the starting gun. How long does it take to cure all the diseases?
Right? That's that's one of the ways that like drives a huge amount of of of of economic value. Right? Like, you cure you cure every disease. You know, there's a question of how much of that goes to the pharmaceutical company, to the AI company, but there's an enormous consumer surplus because everyone assuming we can get access for everyone, which I care about greatly, we cure all of these diseases.
How long does it take? You have to do the biological discovery. You to manufacture the new drug, you have to go through the regulatory process. I mean, we saw this with vaccines and COVID. There's just this we got the vaccine out to everyone, but it took a year and a half.
So my question is, how long does it take to get the cure for everything, which AI is the genius that can, in theory, invent out to everyone? How long from when that AI first exists in the lab to when diseases have actually been cured for everyone? Right? In in you know, we've had a polio vaccine for fifty years. We're still trying to eradicate it in the most remote corners of Africa.
And, you know, the Gates Foundation is trying as hard as they can. Others are trying as hard as they can, but, you know, that's difficult. Again, I, you know, I don't expect most of the economic diffusion to be as difficult as that. Right? That's like the most difficult case.
But but there's a there's a real dilemma here. And and where I've settled on it is it will be it will be it will be faster than anything we've seen in the world, but it still has its limits. And and so then when we go to buying data centers, you know, you again again, the curve I'm looking at is, okay. We've had a 10x a year increase every year. So beginning of this year, we're looking at 10,000,000,000 in rate of annualized revenue at the beginning of the year.
We have to decide how much compute to buy. And it takes a year or two to actually build out the data centers, to reserve the data center. So basically, I'm saying in 2027, how much compute do I get? Well, I could assume that the revenue will continue growing 10x a year, so it'll be 100,000,000,000 at the 2026 and 1,000,000,000,000 at the end of twenty twenty seven. And so I could buy a trillion dollars.
Actually, it would be like $5,000,000,000,000 of compute because it would be a trillion dollar a year for five years. Right? I could buy a trillion dollars of compute that starts at the 2027. If my revenue is not a trillion dollars, if it's even 800,000,000,000, there's no force on earth. There's no hedge on earth that could stop me from going bankrupt if I buy that much compute.
So even though a part of my brain wonders if it's gonna keep going 10x, I can't buy a trillion dollars a year of compute in 2027. If I'm just off by a year in that rate of growth or if the growth rate is five x a year instead of 10 x a year, then then, you know, then you go bankrupt. And so you end up in a world where, you know, you're supporting hundreds of billions, not trillions, and you accept some risk that there's so much demand that you can't support the revenue, and you accept still some risk that you got it wrong and it's still slow. And so when I talked about behaving responsibly, what I meant actually was not the absolute amount. That that actually was not you know, I think it is true we're spending somewhat less than some of the other players.
It's actually the other things like, have we been thoughtful about it? Or are we YOLO ing and saying, oh, we're gonna do a $100,000,000,000 here, a $100,000,000,000 there. I kinda get the impression that, you know, some of the other companies have not written down the spreadsheet, that they don't really understand the risks they're taking. They're just kinda doing stuff because it sounds cool.
@MilkRoadAI - Milk Road AI
Stop scrolling, watch this video immediately and save this S&P's Daniel Yergin just dropped a brutal reality check: we need 50% more copper for electrification, but new mines take 17 years to come online. A global shortage is now "inevitable." https://t.co/Tp7SS4bc6t
Video Transcript AI Summary
Dan and Kelly discuss the outlook for copper and what’s baked into current prices.
Kelly notes that short-term factors influencing copper include disruptions in mines in parts of the world, tariffs, and uncertainty about Federal Reserve policy. In the longer term, she says, prices reflect a growing need for copper as the world electrifies. They estimate that by 2040 the world will use 50% more electricity than today, which she equates to “building 650 nuclear power plants every year.” Copper is described as the “metal of electrification.”
She explains that much of the demand growth will come from developing countries, and that with the rise of data centers and AI there is a voracious appetite for electricity that has surprised traditional utilities. She cites that data centers used about 4% of US electricity last year, and by 2030 it will be more like 14%, and none of that happens without copper.
Dan recalls that copper was first discussed as a major story in 2022, noting that while prices have risen since, they haven’t surged like major tech equities. He acknowledges that commodities are highly cyclical and asks how investors can ensure continued upside given potential soft data points or supply coming online.
Kelly responds by emphasizing copper’s link to GDP, describing it as a core economic demand vector. She notes that a key factor is government policy toward mining exploration: it takes an average of 17 years to bring a new copper mine online, so investing in copper is a bet on the future and depends on how governments regulate mining exploration.
Overall, the conversation highlights the thesis that long-term copper demand will be driven by electrification and rising electricity use (especially from data centers and AI), while near-term price dynamics will be influenced by mine disruptions, tariffs, and macropolicy. The lag between discovering, permitting, and developing new copper mines (about 17 years) adds to the structural bullish case.
Full Transcript
Speaker 0: Dan, it's great to have you here. Welcome.
Speaker 1: Well, thank you. Kelly is very back. Good to be back with you.
Speaker 0: You know, I was trying to go back and remember the first time you published on copper warning and and trying to draw attention to the fact that so much was needed and were undersupplied. And not that we can buy it, but I thought to myself, if someone had bought it then, I mean, it has absolutely soared. So not that you're here to give trading advice, but is all of this bullishness now priced in or not?
Speaker 1: Well, I think that some of it is has to do with disruptions in mines that's occurring in some parts of the world. Lack of the the impact of tariffs and uncertainty about what happens and also waiting to see what the Federal Reserve does overall. So I think those are the short term factors. But longer term, I think the prices are, if not at this level, at least pointing to the fact that the world's gonna need a lot more copper, as you said, about 50% more than we have today. And the main reason is that the world is electrifying.
We estimate in our own that in a by the year 2040, the world will use 50% more electricity than it does today. That's like building 650 nuclear power plants every year. Wow. Copper is the metal of electrification.
Speaker 0: That's like building 650 nuclear power plants every year?
Speaker 1: Yes. That's that's what the demand growth is looking like because a lot of the demand growth is obviously in the developing world. But now with the advent of, of data centers, AI, you suddenly have this voracious appetite for electricity that is really, in fact, caught traditional utilities by surprise because they've been used to flat demand. We estimate that, you know, data centers used about 4% of US electricity, last year. By 2030, which is pretty close.
It'll be more like 14%, and none of that happens without copper.
Speaker 0: You I I found it was in 2022 you guys first were talking about this. So over the past three years, you know, the price has risen. It it it hasn't risen like NVIDIA's share price, to be sure.
Speaker 1: And it commodity nothing anything has.
Speaker 0: Correct. In commodities, investors know that these things are are hugely cyclical, so they might say, okay, well, I could get in on this trade now, but how do I make sure that there continues to be so much upside? Right? Because every you turn around and then maybe there's a soft economic data point so you can react to that or you turn around and you know supply is being brought online. Mean, that's traditionally what was the famous bet in the eighties where they said, I think commodities are gonna outperform the S and P and you always lose.
Right?
Speaker 1: Yeah. That's right. Well, I think, you know, I think that's true. I mean, cycles are there, and we have to remember that also copper is called doctor copper because one of the vectors of demand is a traditional core economic demand that we see in everything, and that's very responsive to GDP. So that will have a impact.
What also matters is how governments treat companies who wanna come and explore and develop copper mines. It takes an average, Kelly, seventeen years to bring on a new copper mine. So you are making a bet on the future when you start a new mine.
@MilkRoadAI - Milk Road AI
Robert Friedland is warning that humanity faces a copper crisis. This crisis will fundamentally constrain economic growth. It will limit AI deployment and the energy transition. We consume 30 million tonnes of copper annually. But we only recycle 4 million tonnes of it. This means we're dependent on constant new mining. We need new mining to keep the economy running. To maintain just 3% GDP growth over the next 18 years, we need to build out AI data centers. We need to electrify vehicles and expand renewable energy. To do all this, we need to mine as much copper as humanity has mined in the previous 10,000 years combined. The demand explosion is happening from four directions at once. First, conventional economic growth drives baseline copper consumption. This consumption happens in construction, appliances, and traditional infrastructure. Second, the energy transition requires more copper per vehicle for electric cars. It requires massive amounts for renewable energy infrastructure. Wind turbines and solar panel installations are copper intensive. Third, AI data centers are a new copper monster. A single hyperscale facility can consume 50,000 tonnes of copper. By 2030, data centers could need half a million tonnes annually. This is as AI infrastructure explodes across the globe. Fourth, defense spending is driving demand for electrified military equipment. Advanced telecommunications systems also need copper. Add it all up and global copper demand is projected to jump 50% by 2040. It's climbing from 28 million tonnes today to 42 million tonnes annually. Here's the problem: supply can't keep up. Global copper production will peak in 2030. It will peak at around 33 million tonnes. Then it will stagnate or decline. Remaining ore deposits are lower quality. Extraction costs are rising significantly. Mining takes 17 years from discovery to production. By 2040, the world will face a shortage. It will be roughly 10 million tonnes annually. That's about 25% below what's actually needed. You can't solve a 25% shortage just by raising prices. You physically don't have enough copper. Recycling sounds like it should solve this. But it won't. Yes, recycled copper saves 85% of the energy required for mining. Improving collection could recover an additional 6 million tonnes annually. This could happen by mid century. The problem is you can only recycle copper that's already in circulation. Old wiring, end of life cars, and decommissioned electronics contain recycled copper. To actually close the supply gap, the world would need to double recycling rates to 66%. We'd also need to source 10 million additional tonnes from new mining by 2040. Realistic projections show primary supply will actually decline slightly. Why should you care? This copper constraint becomes a chokepoint for everything you care about. If there's not enough copper, AI infrastructure deployment slows down. This means slower AI progress. It means higher cloud computing costs. You'll pay more for software and services you use. Renewable energy deployment decelerates significantly. You can't source enough copper for grid upgrades. New solar and wind installations become harder to build. Electric vehicle adoption stalls. Production becomes constrained and prices spike. The US government has already classified copper as a critical mineral. Policymakers recognize it's essential to national competitiveness. It's essential to national security. China controls significant portions of global copper mining. China also controls global copper processing. This adds geopolitical risk. Copper prices are already signaling the crisis. The metal hit $11,771 per tonne in December 2025. But that's still not high enough to justify building new mines. Friedland estimates you need $15,000 per tonne. That's the price needed to make new mine projects economically viable. JP Morgan projects $12,075 per tonne averaging in 2026. Goldman Sachs forecasts $10,710 in early 2026. Long term projections are around $15,000 by 2035. What needs to happen? Friedland advocates for deploying advanced exploration technology. This technology finds new deposits faster. Building high efficiency mining operations is essential. Massively scaling up recycling infrastructure is critical. But more fundamentally, governments need to stockpile strategic copper reserves. Accelerating permitting for new mines is necessary. Managing demand by prioritizing critical applications is required.
Video Transcript AI Summary
Copper and aluminum are the primary beneficiaries of the grid spending increase. That $800,000,000,000 is going to buy copper, which is money. The oil market, compared to the metals market, is dwarfed by the demand for metals like copper, aluminum, iron ore, gold, and nickel, which are said to be so thinly traded and critical that there is no chance to get off crude oil. You can’t build electric cars, windmills, solar, or a modern military without these metals. Underwater power cables are expensive, and offshore wind and bringing that electricity green requires copper—copper, copper, copper.
Copper now is described as a trillion-dollar annual market by tomorrow morning. There is no copper inventory to meet this demand. Since Mohenjo Daro, humanity has mined 700,000,000 metric tons of copper. If we put that in a big cube for scale (about 4 thirty-meter sides), approximately 80% of all the copper ever mined is still in human possession. Recycling could recover about 80% of that 700,000,000 tons, but it would require tearing down every building in the United States, Europe, Japan, and China. We can recycle copper from buildings and even from the university in front of us, but the consequence would be living in the dark.
Currently, we consume 30,000,000 tons of copper per year, with only 4,000,000 tons recycled. To maintain 3% GDP growth with no electrification, this speaker claims we must mine the same amount of copper in the next eighteen years as we mined in the last ten thousand years. In the next eighteen years, we would need to mine the same copper volume as mined in the entire previous span of human history, without electrification, without data centers, without solar and wind, and without the greening of the world economy. Since 1900, the energy required to produce copper has increased sixteen-fold, and as ore grades decline, more energy is needed to produce the same metal while water consumption has doubled. Grades are declining globally, and easy copper mines are depleted; Chile is highlighted as a major producer (24% of global copper mine production), yet costs are in the third or fourth quartile. They burn coal in the Chilean grid, and solar is ineffective for mining because the sun only shines a few hours a day; solar is useless without grid-scale storage.
The speaker asserts we are heading for a train wreck in Chile and that we need six giant tier-one mines online every year from now until 2050 to meet copper demand for electrification, data centers, and grid upgrades—40% of the production to come from new mines. All the hype about AI is dismissed as fantasy because we do not have the energy. Nuclear power is proposed as a solution, but what are those plants made of? All the metals mentioned earlier. The country reportedly does not have the capability to weld containment vessels in a traditional nuclear power plant anymore, whereas Korea can build a nuclear power plant.
Full Transcript
Speaker 0: Copper and aluminum are the primary beneficiaries of the grid spending increase. That $800,000,000,000 is going to buy copper, which is money. So how big is the oil market compared to the metals market? If you look at that circle, that's crude oil. You want to get rid of crude oil?
The iron ore, the gold, the copper, the aluminum, the nickel, all the metals are so thinly traded and so critical. You have no chance to get off of crude oil. You can't build electric cars and windmills and solar and have a modern military without these metals. So there's a reason why underwater power cables are so expensive. That's what it looks like when you put up a windmill offshore Nantucket Island and you wanna bring that electricity and be green, it's all copper, copper, copper, copper, copper.
Copper right now, we're expecting that to be $270,000,000,000 a year market by tomorrow morning. And where's this metal going to come from? There's no copper inventory at all. Let me show you how bad this problem is. In human history, if we go back to Mohenjo Daro, we have mined 700,000,000 metric tons of copper.
So we put that in a big cube. You see the Eiffel Tower for scale. It's about four thirty meters by four thirty meters. Approximately 80% of all the copper we've ever mined is still in human possession. Now if you want to get that back, no problem.
We can recycle that. All we have to do is tear down every building in The United States, building in Europe, every building in Japan and all of China, and we can get back about 80% of that 700,000,000 tons. But we'll be living and freezing in the dark. These lights are on here because the copper that's embedded in this building. We can tear it off this university and get the copper back.
Yeah, we can recycle it. How much are we using? We're consuming 30,000,000 tons of copper a year, only 4,000,000 tons of which is recycled. That means to maintain 3% GDP growth. Now listen carefully, with no electrification, this is with burning oil and gas.
To maintain global 3% GDP growth we have to mine the same amount of copper in the next eighteen years as we mined in the last ten thousand years. In the next eighteen years I've got to mine the same amount of copper as we mined in the last ten thousand years without electrification, without data centers, without solar and wind and the greening of the world economy. You people have no idea whatsoever what we're facing. You're dreaming. Since 1900 the energy to produce copper, the energy you need to make copper is 16 fold up.
As the ore grade in the world goes down we need more and more energy to produce the very metal that produces the energy and water consumption has doubled. So grades are declining in the world's minds. The easy copper mines were mined. They're like 100 year old ladies laying in bed waiting to die in Chile. So Chile accounts for a lot.
That's 24% of global copper mine production. But you know their costs are in the third or fourth quartile. They're very expensive. They burn coal in the Chilean grid and solar doesn't work for a mine because the sun only shines five hours a day. Solar is useless unless you have grid scale storage.
So we're heading for a train wreck in Chile. We need six giant tier one minutees to come online every year between now and 2050 when I'm scheduled to be a 100 years old. To meet copper demand that means 40% of the production for new mines for electrification, data centers and grid upgrades. And all this bullshit about AI, everything you're hearing, like it's a fantasy because we don't have the energy. Now maybe we can build nuclear power plants.
What are they made out of? All those metals I told you about. We don't even have the capability in this country to to weld the containment vessels in a traditional nuclear power plant. We used to build them in this country. Korea can build a nuclear power
@MilkRoadAI - Milk Road AI
Andrej Karpathy casually shows how to build real apps just by prompting in under 30 minutes. Save this, I promise you'll come back to it. https://t.co/vTkXkX1SFP
Video Transcript AI Summary
Former Tesla AI director Andre Karpathy discusses software in the era of AI, emphasizing how software is changing at a fundamental level and what this means for students entering the industry.
Key framework: three generations of software
- Software 1.0: the code that programs computers.
- Software 2.0: neural networks, where you tune data sets and run optimizers to create model parameters; the weights program the neural nets rather than hand-written code.
- Software 3.0: prompts as programs that program large language models (LLMs); prompts are written in English, effectively a new programming language.
- He notes that a growing amount of GitHub-like activity in software 2.0 blends English with code, and that the ecosystem around LLMs resembles a newer GitHub-like space (e.g., Hugging Face, Model Atlas). An example: tuning a LoRa on Flux’s image generator creates a “git commit” in this space.
Evolving software stacks in practice
- At Tesla Autopilot, the stack evolved from heavy C++ (software 1.0) to neural nets handling image processing and sensor fusion, with many 1.0 components being migrated to 2.0. The neural network grew in capability and size, and the 1.0 code was deleted as functionality migrated to 2.0.
- We now have three distinct programming paradigms: 1.0 coding, 2.0 weights, and 3.0 prompts. Fluent capability in all three is valuable because tasks may be best solved with code, trained networks, or prompts.
LLMs as a new computer and ecosystem view
- Andrew Ng’s “AI is the new electricity” is cited to frame LLMs as utility-like (CapEx for training, OpEx for API serving, metered usage, low latency, high uptime) and also as fabs-like (large CapEx, rapid tech-tree growth), though software nature means more malleability.
- LLMs are compared to operating systems: CPU-like core, memory in context windows, and orchestration of compute/memory for problem solving. App downloads can be run across various LLM platforms similarly to cross-OS apps.
- The diffusion pattern of LLMs is inverted compared to many technologies: governments and corporations often lag behind consumer adoption, with AI topics sometimes used for everyday tasks like “boiling an egg” rather than high-level strategic aims.
Practical implications for developers and students
- Build fluently across paradigms: code in 1.0, tune 2.0 models, and design 3.0 prompts; decide when to code, train, or prompt depending on task.
- Partially autonomous apps: exemplified by Cursor and Perplexity.
- Cursor: traditional interface plus LLM integration, with under-the-hood embeddings, diffs, and multi-LLM orchestration; GUI support for auditing changes; autonomy slider lets users control how much the AI acts vs. what humans verify.
- Perplexity: similar features, with sources cited and ability to scale autonomy from quick search to deep research.
- Autonomy slider concept: users can limit or increase AI autonomy depending on task complexity; the AI handles context management and multi-call orchestration, while humans verify for correctness and security.
- Education and “keeping AI on the leash”: emphasize concrete prompts, better verification, and development of structured education pipelines with auditable AI-generated content.
Opportunities and caveats in AI-assisted workflows
- Education and governance: separate roles for AI-generated courses and AI-assisted delivery to students, ensuring syllabus adherence and auditability.
- Documentation and access for LLMs: docs should be machine-readable (e.g., markdown), and wording should be actionable (avoid “click” commands; provide equivalent API calls like curl) to facilitate LLM interactions.
- Tools to ingest data for LLMs: services that convert GitHub repos into ingestible formats (e.g., git ingest, DeepWiki) to create ready-to-query knowledge bases.
- Agents vs. augmentation: early emphasis on augmentation (Iron Man-like suits) rather than fully autonomous systems; the autonomy slider enables gradual handover from human supervision to more autonomous tasks while maintaining safety and auditability.
- The future of “native” programming: vibe coding and byte coding illustrate how language-based programming lowers barriers, enabling broad participation in software creation; the takeaway is that natural-language interfaces can act as a gateway to software development, even for non-experts.
Closing synthesis
- We’re at an era where enormous code rewriting is needed, and LLMs function as utilities, fabs, and operating systems, though still early—like the 1960s of OS development.
- The next decade will likely feature a spectrum of partially autonomous products with specialized GUIs and rapid verification loops, guided by an autonomy slider and careful human oversight.
- Karpathy envisions an ongoing collaboration with AI: building partial autonomy products, evolving tooling, and experimenting with how the industry and education adapt to this new programming reality. He invites readers to participate in shaping this future.
Full Transcript
Speaker 0: Please welcome former director of AI, Tesla, Andre Karpathy. Hello. Wow. A lot of people here. Hello.
Okay. Yeah. So I'm excited to be here today to talk to you about software in the era of AI. And I'm told that many of you are students, like bachelor's, master's, PhD, and so on, and you're about to enter the industry. And I think it's actually like an extremely unique and very interesting time to enter the industry right now.
And I think fundamentally the reason for that is that software is changing again. And I say again because I actually gave this talk already, but the problem is that software keeps changing. So I actually have a lot of material to create new talks. And I think it's changing quite fundamentally. I think roughly speaking, software has not changed much on such a fundamental level for seventy years.
And then it's changed, I think, about twice quite rapidly in the last few years. And so there's just a huge amount of work to do, a huge amount of software to write and rewrite. So let's take a look at maybe the realm of software. So if we kind of think of this as like the map of software, this is a really cool tool called map of GitHub. This is kind of like all the software that's written.
These are instructions to the computer for carrying out tasks in the digital space. So if you zoom in here, these are all different kinds of repositories, and this is all the code that has been written. And a few years ago, I kind of observed that software was kind of changing, and there was kind of like a new new type of software around, and I called this software two point o at the time. And the idea here was that software one point o is the code you write for the computer. Software two point o are basically neural networks, and in particular, the weights of a neural network.
And you're not writing this code directly. You are most, are more kind of like tuning the data sets and then you're running an optimizer to create the parameters of this neural net. And I think like at the time, neural nets were kind of seen as like just a different kind of classifier, a decision tree or something like that. And so I think it was kind of like I I think this framing was a lot more appropriate. And now actually what we have is kind of like an equivalent of GitHub in the realm of software two point o.
And I think the Hugging Face is basically equivalent of GitHub in software two point o. And there's also model Atlas and you can visualize all the code written there. In case you're curious, by the way, the giant circle, the point in the middle, these are the parameters of flux, the image generator. And so anytime someone tunes a LoRa on top of a flux model, you basically create a git commit in this space and you create a different kind of image generator. So basically what we have is software one point o is the computer code that programs a computer.
Software two point o are the weights which program neural networks. And here's an example of AlexNet image recognizer neural network. Now so far, all of the neural networks that we've been familiar with until recently were kind of like fixed function computers, image to categories or something like that. And I think what's changed and I think as a quite fundamental change is that neural networks became programmable with large language models. And so I see this as quite new, unique, it's a new kind of a computer.
And so in my mind, it's worth giving it a new designation of software three point zero. And basically, your prompts are now programs that program the LLM. And remarkably, these prompts are written in English. So it's kind of a very interesting programming language. So maybe to summarize the difference, if you're doing sentiment classification, for example, you can imagine writing some amount of Python to basically do sentiment classification, or you can train a neural net, or you can prompt a large language model.
So here, this is a few shot prompt and you can imagine changing it and programming the computer in a slightly different way. So basically we have software one point o, software two point o, and I think we're seeing, maybe you've seen a lot of GitHub code is not just like code anymore, there's a bunch of like English interspersed with code. And so I think kind of there's a growing category of new kind of code. So not only is it a new programming paradigm, it's also remarkable to me that it's in our native language of English. And so when this blew my mind a few, I guess years ago now, I tweeted this and I think it captured the attention of a lot of people and this is my currently pinned tweet, is that remarkably we're now programming computers in English.
Now, when I was at, Tesla, we were working on the autopilot and we were trying to get the car to drive. And I sort of showed this slide at the time where you can imagine that the inputs to the car are on the bottom and they're going through a software stack to produce the steering and acceleration. And I made the observation at the time that there was a ton of c plus plus code around in the autopilot, which was the software one point o code, and then there was some neural nets in there doing image recognition. And I kind of observed that over time as we made the autopilot better, basically the neural network grew in capability and size. And in addition to that, all the C plus plus code was being deleted and kind of like was and a lot of the kind of capabilities and functionality that was originally written in one point zero was migrated to two point zero.
So as an example, a lot of the stitching up of information across images from the different cameras and across time was done by neural network and we were able to delete a lot of code. And so the software two point zero stack quite literally ate through the software stack of the autopilot. So I thought this was really remarkable at the time. And I think we're seeing the same thing again where basically we have a new kind of software and it's eating through the stack. We have three completely different programming paradigms.
And I think if you're entering the industry, it's a very good idea to be fluent in all of them because they all have slight pros and cons and you may want to program some functionality in one point o or two point o or three point o. Are you going to train a neural net? Are you going to just prompt an LLM? Should this be a piece of code that's explicit, etcetera? So we all have to make these decisions and actually potentially fluidly transition between these paradigms.
So what I want to get into now is first I want to, in the first part, talk about LLMs and how to kind of like think of this new paradigm and the ecosystem and what that looks like. Like what are, what is this new computer? What does it look like? And what does the ecosystem look like? I was struck by this quote from Andrew Ng actually many years ago now, think.
And I think Andrew is going to be speaking right after me. But he said at the time AI is the new electricity. And I do think that it kind of captures something very interesting in that LLMs certainly feel like they have properties of utilities right now. So LLM Labs like OpenAI, Gemini, Anthropia, etcetera, they spend CapEx to train the LLMs and this is kind of equivalent to building out a grid. And then there's OpEx to serve that intelligence over APIs to all of us.
And this is done through metered access where we pay per million tokens or something like that. And we have a lot of demands that are very utility like demands out of this API. We demand low latency, high uptime, consistent quality, etcetera. In electricity, you would have a transfer switch. So you can transfer your electricity source from like grid and solar or battery or generator.
In LLMs, we have maybe open router and easily switch between the different types of LLMs that exist. Because the LLMs are software, they don't compete for physical space. So it's okay to have basically like six electricity providers and you can switch between them, right, because they don't compete in such a direct way. And I think what's also a little fascinating and we saw this in the last few days actually, a lot of the LLMs went down and people were kind of like stuck and unable to work. And I think it's kind of fascinating to me that when the state of the art LLMs go down, it's actually kind of like an intelligence brownout in the world.
It's kind of like when the voltage is unreliable in the grid and the planet just gets dumber the more reliance we have on these models, which already is like really dramatic and I think will continue to grow. But LLMs don't only have properties of utilities. I think it's also fair to say that they have some properties of fabs. And the reason for this is that the CapEx required for building LLMs is actually quite large. It's not just like building some power station or something like that, right?
You're investing a huge amount of money and I think the tech tree and for the technology is growing quite rapidly. So we're in a world where we have sort of deep tech trees, research and development, secrets that are centralizing inside the LLM labs. And but I think the analogy muddies a little bit also because as I mentioned, this is software. And software is a bit less defensible because it is so malleable. And so I think it's just an interesting kind of thing to think about potentially.
There's many analogies you can make, like a four nanometer process node maybe is something like a cluster with certain max plots. You can think about when you're use when you're using NVIDIA GPUs and you're only doing the software and you're not doing the hardware, that's kind of like the fabless model. But if you're actually also building your own hardware and you're training on TPUs if you're Google, that's kind of like the Intel model where you own your fab. So I think there's some analogies here that make sense. But actually, I think the analogy that makes the most sense perhaps is that in my mind, LLMs have very strong kind of analogies to operating systems.
In that, this is not just electricity or water. It's not something that comes out of the tap as a commodity. These are now increasingly complex software ecosystems. Right? So they're not just like simple commodities like electricity.
And it's kind of interesting to me that the ecosystem is shaping in a very similar kind of way where you have a few closed source providers like Windows or Mac OS, and then you have an open source alternative like Linux. And I think for LLMs as well, we have a kind of a few competing closed source providers. And then maybe the LAMA ecosystem is currently like maybe a close approximation to something that may grow into something like Linux. Again, think it's still very early because these are just simple LMs, but we're starting to see that these are going to get a lot more complicated. It's not just about the LLM itself, it's about all the tool use and the multi modalities and how all of that works.
So when I sort of had this realization a while back, I tried to sketch it out and it kind of seemed to me like LLMs are kind of like a new operating system, right? So the LLM is a new kind of a computer. It's setting, it's kind of like the CPU equivalent. The context windows are kind of like the memory. And then the LLM is orchestrating memory and compute for problem solving using all of these capabilities here.
And so definitely, if you look at it, it looks very much like operating system from that perspective. A few more analogies, for example, if you want to download an app, say I go to Versus Code and I go to download, you can download Versus Code and you can run it on Windows, Linux or Mac. In the same way as you can take an LLM app like Cursor and you can run it on GPT or Cloud or Gemini series. Right? It's just a drop down.
So it's kind of like similar in that way as well. More analogies that I think strike me is that we're kind of like in this 1960s ish era where LLM compute is still very expensive for this new kind of a computer. And that forces the LLMs to be centralized in the cloud and we're all just sort of thin clients that interact with it over the network. And none of us have full utilization of these computers and therefore it makes sense to use time sharing where we're all just a dimension of the batch when they're running the computer in the cloud. And this is very much what computers used to look like during this time.
The operating systems were in the cloud, everything was streamed around and there was batching. And so the personal computing revolution hasn't happened yet because it's just not economical, it doesn't make sense, but I think some people are trying. And it turns out that Mac minis for example are a very good fit for some of the LLMs because it's all if you're doing batch one inference, this is all super memory bound so this actually works. And I think these are some early indications maybe of personal computing, but this hasn't really happened yet. It's not clear what this looks like.
Maybe some of you get to invent what this is or how it works or what this should be. Maybe one more analogy that I'll mention is whenever I talk to ChatGPT or some LLM directly in text, I feel like I'm talking to an operating system through the terminal. Like it's just, it's text, it's direct access to the operating system and I think a GUI hasn't yet really been invented in like a general way. Like should ChatGPT have a GUI, like different than just the tech bubbles? Certainly some of the apps that we're going to go into in a bit have GUI, but there's no like GUI across all the tasks if that makes sense.
There are some ways in which LLMs are different from kind of operating systems in some fairly unique way and from early computing. And I wrote about this one particular property that strikes me as very different this time around, it's that LLMs like flip, they flip the direction of technology diffusion that is usually present in technology. So for example, with electricity, cryptography, computing, flight, Internet, GPS, lots of new transformative technologies that have not been around. Typically, it is the government and corporations that are the first users because it's new and expensive, etcetera, and it only later diffuses to consumer. But I feel like LMs are kind of like flipped around.
So maybe with early computers, it was all about ballistics and military use, but with LLMs, it's all about how do you boil an egg or something like that. This is certainly like a lot of my use. And so it's really fascinating to me that we have a new magical computer and it's like helping me boil an egg. It's not helping the government do something really crazy like some military ballistics or some special technology. Indeed corporations or governments are lagging behind the adoption of all of us, of all of these technologies.
So it's just backwards. And I think it informs maybe some of the uses of how we want to use this technology or like what are some of the first apps and so on. So in summary so far, LLM Labs, bad LLMs, I think it's accurate language to use. But llms are complicated operating systems. They're circa nineteen sixties in computing and we're redoing computing all over again.
And they're currently available via time sharing and distributed like a utility. What is new and unprecedented is that they're not in the hands of a few governments and corporations, they're in the hands of all of us because we all have a computer and it's all just software and Chachi PT was beamed down to our computers like billions of people like instantly and overnight. And this is insane. And it's kind of insane to me that this is the case and now it is our time to enter the industry and program these computers. This is crazy.
So I think this is quite remarkable. Before we program LLMs, we have to kind of like spend some time to think about what these things are. And I especially like to kind of talk about their psychology. So the way I like to think about LLMs is that they're kind of like people spirits. They are stochastic simulations of people.
And the simulator in this case happens to be an autoregressive transformer. So a transformer is a neural net. It's and it just kind of like goes on the level of tokens. It goes chunk, chunk, chunk, chunk, chunk. And there's an almost equal amount of compute for every single chunk.
And this simulator of course is just, is basically there's some weights involved and we fit it to all of text that we have on the internet and so on. And you end up with this kind of a simulator. And because it is trained on humans, it's got this emergent psychology that is human like. So the first thing you'll notice is of course, LLMs have encyclopedic knowledge and memory and they can remember lots of things, a lot more than any single individual human can because they've read so many things. It actually kind of reminds me of this movie Rain Man which I actually really recommend people watch.
It's an amazing movie. I love this movie. And Dustin Hoffman here is an autistic savant who has almost perfect memory. So he can read like a phone book and remember all of the names and phone numbers. And I kind of feel like LNs are kind of like very similar.
They can remember SHA hashes and lots of different kinds of things very, very easily. So they certainly have superpowers in some set in some respects. But they also have a bunch of, I would say, cognitive deficits. So they hallucinate quite a bit and they kind of make up stuff and don't have a very good sort of internal model of self knowledge, not sufficient at least. And this has gotten better but not perfect.
They display jagged intelligence. So they're going to be superhuman in some problem solving domains. And then they're going to make mistakes that basically no human will make. Like, they will insist that 9.11 is greater than 9.9 or that there are two r's in strawberry. These are some famous examples.
But basically there are rough edges that you can trip on. So that's kind of, I think, also kind of unique. They also kind of suffer from anterograde amnesia. So and I think I'm alluding to the fact that if you have a coworker who joins your organization, this coworker will over time learn your organization and they will understand and gain like a huge amount of context on the organization. And they go home and they sleep and they consolidate knowledge and they develop expertise over time.
LLMs don't natively do this and this is not something that has really been solved in the R and D of LLMs, I think. And so context windows are really kind of like working memory and you have to sort of program the working memory quite directly because they don't just kind of like get smarter by default. And I think a lot of people get tripped up by the analogies in this way. In popular culture, I recommend people watch these two movies, Memento and fifty First Dates. In both of these movies, the protagonists, their weights are fixed and their context windows gets wiped every single morning and it's really problematic to go to work or have relationships when this happens and this happens to LLMs all the time.
I guess one more thing I would point to is security kind of related limitations of the use of LLMs. So for example, LLMs are quite gullible. They are susceptible to prompt injection risks. They might leak your data, etcetera. And so and there's many other considerations security related.
So so basically, long story short, you have to load your, you to load your, you have to simultaneously think through this superhuman thing that has a bunch of cognitive deficits and issues. How do we, and yet they are extremely useful. And so how do we program them? And how do we work around their deficits and enjoy their superhuman powers? So what I want to switch to now is talk about the opportunities of how do we use these models and what are some of the biggest opportunities.
This is not a comprehensive list of some of the things that I thought were interesting for this talk. The first thing I'm kind of excited about is what I would call partial autonomy apps. So for example, let's work with the example of coding. You can certainly go to chat GPT directly and you can start copy pasting code around and copy pasting bug reports and stuff around and getting code and copy pasting everything around. Why would you do that?
Why would you go directly to the operating system? It makes a lot more sense to have an app dedicated for this. And so I think many of you use Cursor, I do as well. And cursor is kind of like the thing you want instead. You don't want to just directly go to the chat GPT.
And I think cursor is a very good example of an early LLM app that has a bunch of properties that I think are useful across all the LLM apps. So in particular, you will notice that we have a traditional interface that allows a human to go in and do all the work manually just as before. But in addition to that, we now have this LLM integration that allows us to go in bigger chunks. And so some of the properties of LLM apps that I think are shared and useful to point out. Number one, the LLMs basically do a ton of the context management.
Number two, they orchestrate multiple calls to LLMs, right? So in the case of Cursor, there's under the hood embedding models for all your files, the actual chat models, models that apply diffs to the code and this is all orchestrated for you. A really big one that I think also maybe not fully appreciated always is application specific GUI and the importance of it. Because you don't just want to talk to the operating system directly in text. Text is very hard to read, interpret, understand.
And also like you don't want to take some of these actions natively in text. So it's much better to just see a diff as like red and green change. You can see what's being added or subtracted. It's much easier to just do command y to accept or command n to reject. I shouldn't have to type it in text, right?
So GUI allows a human to audit the work of these fallible systems and to go faster. I'm going to come back to this point a little bit later as well. And the last kind of feature I want to point out is that there's what I call the autonomy slider. So for example, in cursor, can just do tab completion, you're mostly in charge. You can select a chunk of code and command k to change just that chunk of code.
You can do command l to change the entire file. Or you can do command I which just let it rip, do whatever you want in the entire repo. And that's the sort of full autonomy agent gen tech version. And so you are in charge of the autonomy slider. And depending on the complexity of the task at hand, you can tune the amount of autonomy that you're willing to give up for that task.
Maybe to show one more example of a fairly successful LLM app, Perplexity, it also has very similar features to what I've just pointed out in Cursor. It packages up a lot of the information. It orchestrates multiple LLMs. It's got a GUI that allows you to audit some of its work. So for example, it will cite sources and you can imagine inspecting them.
And it's got an autonomy slider. You can either just do a quick search or you can do research or you can do deep research and come back ten minutes later. So this is all just varying levels of autonomy that you give up to the tool. So I guess my question is, I feel like a lot of software will become partially autonomous. And I'm trying to think through like what does that look like?
And for many of you who maintain products and services, how are you going to make your products and services partially autonomous? Can an LLM see everything that a human can see? Can an LLM act in all the ways that a human could act? And can humans supervise and stay in the loop of this activity? Because again, these are fallible systems that aren't yet perfect.
And what does a diff look like in Photoshop or something like that? Also a lot of the traditional software right now, it has all these switches and all this kind of stuff, it's all designed for human. All of this has to change and become accessible to LLMs. So one thing I want to stress with a lot of these LLM apps that I'm not sure gets as much attention as it should is, we're now kind of like cooperating with AIs and usually they are doing the generation and we as humans are doing the verification. It is in our interest to make this loop go as fast as possible so we're getting a lot of work done.
There are two major ways that I think this can be done. Number one, you can speed up verification a lot. And I think GUI's for example are extremely important to this because a GUI utilizes your computer vision GPU in all of our head. Reading text is effortful and it's not fun, but looking at stuff is fun and it's just kind of like a highway to your brain. So I think GUIs are very useful for auditing systems and visual representations in general.
And number two, would say is we have to keep the AI on the leash. I think a lot of people are getting way overexcited with AI agents and it's not useful to me to get a diff of 1,000 lines of code to my repo. Like I have to I'm still the bottleneck, right? Even though that 1,000 lines come out instantly, I have to make sure that this thing is not introducing bugs, it's just like and that it's doing the correct thing, right? And that there's no security issues and so on.
So I think that, yeah, basically we have to sort of like it's in our interest to make the flow of these two go very, very fast and we have somehow keep the AI on the leash because it gets way too overreactive. It's kind of like this. This is how I feel when I do AI assisted coding. If I'm just byte coding, everything is nice and great. But if I'm actually trying to get work done, it's not so great to have an overreactive agent doing all this kind of stuff.
So this slide is not very good, I'm sorry. But I guess I'm trying to develop like many of you some ways of utilizing these agents in my coding workflow and to do AI assisted coding. And in my own work, I'm always scared to get way too big diffs. I always go in small incremental chunks. I want to make sure that everything is good.
I want to spin this loop very, very fast. And I sort of work on small chunks of single concrete thing. And so I think many of you probably are developing similar ways of working with LLMs. I also saw a number of blog posts that try to develop these best practices for working with LLMs. And here's one that I read recently and I thought was quite good.
And it kind of discussed some techniques. Some of them have to do with how you keep the AI on the leash. And so as an example, if you are prompting, if your prompt is vague, then the AI might not do exactly what you wanted and in that case verification will fail. You're going to ask for something else. If verification fails, then you're going to start spinning.
So it makes a lot more sense to spend a bit more time to be more concrete in your prompts which increases the probability of successful verification and you can move forward. And so I think a lot of us are going to end up finding techniques like this. Think in my own work as well, I'm currently interested in what education looks like in together with kind of like now that we have AI and all of them, what does education look like? And I think a large amount of thought for me goes into how we keep AI on the leash. I don't think it just works to go to trashypt and be like, hey, teach me physics.
I don't think this works because the AI is like gets lost in the woods. And so for me, this is actually two separate apps for example. There's an app for a teacher that creates courses and then there's an app that takes courses and serves them to students. And in both cases, we now have this intermediate artifact of a course that is auditable and we can make sure it's good, we can make sure it's consistent and that the AI is kept on the leash with respect to a certain syllabus, a certain like progression of projects and so on. And so this is one way of keeping the AI on the leash and I think has a much higher likelihood of working.
And the AI is not getting lost in the woods. One more kind of analogy I wanted to sort of allude to is I'm not I'm no stranger to partial autonomy and I've kind of worked on this I think for five years at Tesla. And this is also a partial autonomy product and shares a lot of the features. But for example, right there in the instrument panel is the GUI of the autopilot. So it's showing me what the what the neural network sees and so on.
And we have the autonomy slider where over the course of my tenure there, we did more and more autonomous tasks for the user. And maybe the story that I wanted to tell very briefly is, actually the first time I drove a self driving vehicle was in 2013 I had a friend who worked at Waymo and he offered to give me a drive around Palo Alto. I took this picture using Google Glass at the time. And many of you are so young that you might not even know what that is. But yeah, this was like older age at the time.
And we got into this car and we went for about a thirty minute drive around Palo Alto, highways, streets and so on. And this drive was perfect. There was zero interventions. And this was 2013, which is now twelve years ago. And it kind of struck me because at the time when I had this perfect drive, this perfect demo, I felt like, wow, self driving is imminent because this just worked, this is incredible.
But here we are twelve years later and we are still working on autonomy. We are still working on driving agents. And even now we haven't actually like fully solved the problem. Like you may see Waymo's going around and they look driverless, but there's still a lot of teleoperation and a lot of human in the loop of a lot of this driving. So we still haven't even like declared success, but I think it's definitely like going to succeed at this point, but it just took a long time.
And so I think like like this is software is really tricky, I think in the same way that driving is tricky. And so when I see things like, oh, 2025 is the year of agents, I get very concerned and I kind of feel like, you know, this is the decade of agents and this is going to be quite some time. We need humans in the loop. We need to do this carefully. This is software.
Let's be serious here. One more kind of analogy that I always think through is the Ironman suit. I think this is I always love Ironman. I think it's like so correct in a bunch of ways with respect to technology and how it will play out. And what I love about the Iron Man suit is that it's both an augmentation and Tony Stark can drive it, and it's also an agent.
And in some of the movies, the Iron Man suit is quite autonomous and can fly around and find Tony and all this kind of stuff. So And this is the autonomy slider is we can be we can build augmentations or we can build agents and we kind of want to do a bit of both, but at this stage, I would say working with fallible LLMs and so on, I would say, you know, it's less Ironman robots and more Ironman suits that you want to build. It's less like building flashy demos of autonomous agents and more building partial autonomy products. And these products have custom GUIs and UI UX and we're trying to and this is done so that the generation verification loop of the human is very, very fast. But we are not losing the sight of the fact that it is in principle possible to automate this work.
And there should be an autonomy slider in your product and you should be thinking about how you can slide that autonomy slider and make your product sort of more autonomous over time. But this is kind of how I think there's lots of opportunities in these kinds of products. I wanna now switch gears a little bit and talk about one other dimension that I think is very unique. Not only is there a new type of programming language that allows for autonomy in software, but also, as I mentioned, it's programmed in English, which is this natural interface. And suddenly, everyone is a programmer because everyone speaks natural language like English.
So this is extremely bullish and very interesting to me and also completely unprecedented, I would say. It it used to be the case that you need to spend five to ten years studying something to be able to do something in software. This is not the case anymore. So I don't know if by any chance anyone has heard of vibe coding. This this is the tweet that's kind of like introduced this, but I'm told that this is now like a major meme.
Fun story about this is that I've been on Twitter for like fifteen years or something like that at this point and I still have no clue which tweet will become viral and which tweet like fizzles and no one cares. And I thought that this tweet was gonna be the latter. I don't know, was just like a shower of thoughts, but this became like a total meme and I really just can't tell. But I guess I'll get struck a chord and gave a name to something that everyone was feeling but couldn't quite say in words. So now there's Wikipedia page and everything.
This is like Yeah. This is like a major contribution now or something like that. So so Tom Wolf from Hugging Face shared this beautiful video that I really love. These are kids vibe coding. And I find that this is such a wholesome video.
Like, love this video. Like, how can you look at this video and feel bad about the future? The future is great. I think this will end up being like a gateway drug to software development. I'm not a doomer about the future of the generation, and I think yeah.
I love this video. So I tried bytecoding a little bit as well because it's so fun. So bytecoding is so great when you wanna build something super duper custom that doesn't appear to exist and you just wanna wing it because it's a Saturday or something like that. So I built this iOS app and I don't I can't actually program in Swift, but I was really shocked that I was able to build like a super basic app and I'm not gonna explain it. It's really dumb.
But I kind of like this was just like a day of work and this was running on my phone like later that day and I was like, wow. This is amazing. I didn't have to like read through Swift for like five days or something like that to like get started. I also bot coded this app called MenuGen, and this is live. You can try it in menugen.
App. And I basically had this problem where I show up at a restaurant, I read through the menu and I have no idea what any of the things are and I need pictures. So this doesn't exist. So I was like, hey, I'm gonna bot code it. So this is what it looks like.
You go to menugen.app and you take a picture of a of a menu and then menu gen generates the images. And everyone gets $5 in credits for free when you sign up, and therefore, this is a major cost center in my life. So this is a negative negative revenue app for me right now. I've lost a huge amount of money on MenuGen. Okay.
But the fascinating thing about menu gen for me is that the code of the byte the byte coding part, the code was actually the easy part of byte of byte coding menu gen. And this was extreme slog and took another week. So it was really fascinating that I had the menu gen basically demo working on my laptop in a few hours and then it took me a week because I was trying to make it real. And the reason for this is this was just really annoying. So for example, if you try to add Google login to your web page, I know this is very small, but just a huge amount of instructions of this clerk library telling me how to integrate this.
And this is crazy. Like, it's telling me, go to this URL, click on this drop down, choose this, go to this, and click on that. And it's like telling me what to do. Like, a computer is telling me the actions I should be taking. Like, you do it.
Why am I doing this? What the hell? I had to follow all these instructions. This was crazy. So I think the last part of my talk, therefore, focuses on can we just build for agents?
I don't want to do this work. Can agents do this? Thank you. Okay. So roughly speaking, I think there's a new category of consumer and manipulator of digital information.
It used to be just humans through GUIs or computers through APIs, and now we have a completely new thing. And agents are their computers, but they are human like, kind of. Right? They're people spirits. There's people spirits on the Internet and they need to interact with our software infrastructure.
What can we build for them? It's a new thing. So as an example, can you have robots.txt on your domain and you can instruct or like advise, I suppose, web crawlers on how to behave on your website. In the same way, you can have maybe l n dot t x t file, which is just a simple markdown that's telling LLMs what this domain is about. And this is very readable to a to an LLM.
If it had to instead get the HTML of your web page and try to parse it, this is very error prone and difficult and we'll screw it up and it's not gonna work. So we can just directly speak to the LLM. It's worth it. A huge amount of documentation is currently written for people, so you will see things like lists and bold and pictures, and this is not directly accessible by an LLM. So I see some of the services now are transitioning a lot of their their docs to be specifically for LLMs.
So Vercel and Stripe as an example are early movers here, but there are a few more that I've seen already and they offer their documentation in markdown. Markdown is super easy for LLMs to understand. This is great. Maybe one simple example from from my experience as well. Maybe some of you know three blue one brown, he makes beautiful animation videos on YouTube.
Yeah. I love this library. I said that he wrote Manon. And I wanted to make my own. And there's extensive documentations on how to use Manon.
And so I didn't wanna actually read through it. So I copy pasted the whole thing to an LLM and I described what I wanted and it just worked out of the box. Like, LLM just by coded me an animation exactly what I wanted and I was like, wow, this is amazing. So if we can make docs legible to LLMs, it's going to unlock a huge amount of kind of use and I think this is wonderful and should should happen more. The other thing I wanted to point out is that you do unfortunately have to it's not just about taking your docs and making them appear in markdown.
That's the easy part. We actually have to change the docs because anytime your docs say click, this is bad. An LLM will not be able to natively take this action right now. So Vercel, for example, is replacing every occurrence of click with the equivalent curl command that your LLM agent could take on your behalf. And so I think this is very interesting.
And then, of course, there's a model context protocol from Anthropic, and this is also another way. It's a protocol speaking directly to agents as this new consumer and manipulator of digital information. So I'm very bullish on these ideas. The other thing I really like is a number of little tools here and there that are helping ingest data that in, like, very LLM friendly formats. So for example, when I go to a GitHub repo like my Nana GPT repo, I can't feed this to an LLM and ask questions about it because it's you know, this is a human interface on GitHub.
So when you just change the URL from GitHub to git ingest, then this will actually concatenate all the files into a single giant text and it will create a directory structure, etcetera, and this is ready to be copy pasted into your favorite LLM, and you can do stuff. Maybe even more dramatic example of this is DeepWiki, where it's not just the raw content of of these files. This is from Devon. But also, like, they have Devin basically do analysis of the GitHub repo, and Devin basically builds up a whole docs pages just for your repo. And you can imagine that this is even more helpful to copy paste into your LLM.
So I love all the little tools that basically where you just change the URL and it makes something accessible to an LLM. So this is all well and great and, yeah, I think there should be a lot more of it. One more note I wanted to make is that it is absolutely possible that in the future, LLMs will be able to it's not even future, this is today. They'll be able to go around and they'll be able to click stuff and so on. But I still think it's very worth basically meeting LLM halfway, LLMs halfway, and making it easier for them to access all this information because this is still fairly expensive, I would say, to use and a lot more difficult.
And so I do think that lots of software, there will be a long tail where it won't, like, adapt because these are not, like, live player sort of repositories or digital infrastructure, and we will need these tools. But I think for everyone else, I think it's very worth kind of like meeting in some middle point. So I'm bullish on both, if that makes sense. So in summary, what an amazing time to get into the industry. We need to rewrite a ton of code.
A ton of code will be written by professionals and by coders. These LLMs are kind of like utilities, kind of like fabs, but they're kind of especially like operating systems, but it's so early. It's like nineteen sixties of operating systems, And and I think a lot of the analogies crossover. And these LLMs are kind of like these fallible, you know, people spirits that we have to learn to work with. And in order to do that properly, we need to adjust our infrastructure towards it.
So when you're building these LLM apps, I described some of the ways of working effectively with these LLMs and some of the tools that make that kind of possible and how you can spin this loop very, quickly and basically create partial autonomy products. And then, yeah, a lot of code has to also be written for the agents or directly. But in any case, going back to the Ironman suit analogy, I think what we'll see over the next decade roughly is we're going to take the slider from left to right and I'm very interesting it's gonna be very interesting to see what that looks like and I can't wait to build it with all of you. Thank you.
reSee.it AI Summary
MilkRoadAI kickstarted with Jensen Huang’s 2011 Stanford talk and Nvidia’s later ascent to a multi-trillion value, urging preservation. lkondeth asked for a 10-point summary and 5 actions. grok’s version covers: lasting impact comes from a unique perspective; Nvidia started in 1993 with 3 engineers targeting 3D graphics; early rejections but persistence; 3D graphics enabled broader apps; strong competition; GPUs became programmable; embrace failure and honesty; passion and big markets; leadership cultivation. Five actions: pursue a unique view; take calculated risks; reinvent often; foster honesty; value passion and learning.
@MilkRoadAI - Milk Road AI
Forget the $200,000 degree. In 2011, Jensen Huang gave a lecture at Stanford that explains strategy better than any professor. NVIDIA is now worth $5 Trillion. Save this post. You won't find this video again once your feed refreshes: https://t.co/T79TtP74Mv
Video Transcript AI Summary
Jensen Huang opens by inviting an interactive conversation about building a company, noting that it is both gratifying and incredibly hard, with perspectives on company building shaped by diverse experiences. He recalls NVIDIA’s beginnings sixteen years ago with three engineers and introduces the idea that perspective, more than grand vision, drives entrepreneurial direction. He distinguishes vision from perspective, arguing that vision is not exclusive to a few, while everyone has a perspective—the way you see the world and identify opportunities.
In 1993, with Windows 3.1 era and no networks or wireless tech, Huang explains NVIDIA’s perspective: a PC could run three-dimensional graphics programs to explore new worlds, enabling video games as the killer app. The business plan was to take advanced graphics technology from expensive workstations, reinvent it, and make it affordable. He recounts pitching to Sand Hill Road, who doubted a video game market existed, and a parental nudge to get a real job. Yet the team believed video games would be a large market, a view later validated by today’s status as the world’s largest digital media industry. They also anticipated broader uses for the technology beyond games, such as a notable example with Keyhole (which Google acquired to become Google Earth, the world’s largest downloaded application).
He emphasizes that perspectives often differ even among seemingly obvious opportunities. He cites Yahoo!, AltaVista, Lycos, and others, illustrating how two similar cores (search) could lead to different outcomes based on what each company chose to become (destinations/portals, etc.). Competition was intense as hundreds of three-dimensional graphics startups emerged, yet NVIDIA remains the only surviving graphics company. The lesson is that perspective matters because different viewpoints shape strategic focus.
Huang then discusses the core business principle: Moore’s Law—though framed as a competition-driven efficiency—drives GPU advancement. The early approach was to make three-dimensional graphics insatiable—improving performance year after year even if customers initially resisted due to cost. For the first five years, NVIDIA “turned off the blinders” and ignored customer constraints, eventually cannibalizing its own products when a new generation proved more capable and profitable.
Innovation is risky, he notes, and sustaining a leading position required reinvention. By the late 1990s, NVIDIA shifted from a fixed-function graphics accelerator to a programmable shader architecture with the GeForce FX (a gamble that nearly killed the company but ultimately paid off). The introduction of programmable shaders kept NVIDIA at the forefront, enabling GPUs to be used for general-purpose computing (GPGPU), which has become a major trajectory.
On company culture, Huang stresses the importance of fostering risk-taking and a tolerance for failure, teaching people how to fail quickly and cheaply, and maintaining intellectual honesty to pivot when necessary. He contrasts older, more rigid corporate cultures with modern, beta-form experimentation found in companies like Google, where many applications operate in beta to test ideas rapidly.
Regarding cofounders and governance, he notes that equity was divided equally among the three founders (each initially contributing $200 and receiving 20% each). He explains that leadership should be clearly established (Jensen as CEO) to avoid decision-making gridlock, while still valuing collaboration with strong, trusted partners.
Asked about the venture capital process, Huang explains that VCs invest in people and a sufficiently large, novel market, not just a polished business plan. He shares that their reputations and prior work with notable figures helped, and he emphasizes the ongoing importance of great people and a focused, strategic vision.
He addresses mentors and best advice—focus intensely on a few things, learn from diverse sources, and remain adaptable. On succession, Huang argues against rigid, preselected succession planning, favoring the cultivation of future leaders within the company so that many internal options exist if leadership changes become necessary.
Finally, he speaks about the finance side in the early days: cash is king and survival is paramount, constantly raising or conserving funds. He closes by reiterating the core message: ideas are plentiful, but a unique, passionate perspective and perseverance are what sustain a company, along with a culture that embraces calculated risk and continuous reinvention.
Full Transcript
Speaker 0: Instead of giving you a company presentation today, what I thought I would do is just have a conversation with you. At any time, if you have a question, if you would like to change the direction of conversation, just raise your hand, and we'll talk about whatever comes up. A lot of people talk about and write about building companies. And I can tell you firsthand that building a company is extraordinarily gratifying. It is also incredibly hard.
And so the things that you want to talk about with respect to the company building process is rather expansive. You could talk about company building processes from a lot of different perspectives. And so I'm going to try to touch on a few of them that I think are particularly important in my experience. So sixteen years ago, NVIDIA had three people, three engineers.
Speaker 1: Did
Speaker 0: I do something? Was it me? We had a perspective that if I was just
Speaker 2: You think it's me? Think it's.
Speaker 0: How about I just did that? See, it's you can't be you can't be you can't control user stupidity, you know. You're good. You're good. It was probably me.
I was sitting on it. All right. So sixteen years ago, we started NVIDIA. And the insight that we had, some people call it vision. Vision is an awfully big word to me.
Vision is an awfully big word to me because I believe first of all, vision matters. Let me tell you that vision matters, and I'll help you understand that in a second. But I like to use the word perspective because it makes it possible for anyone to have one. When you say vision, it feels like only a few selected visionaries of the world can have one. But everyone has a perspective, and that's, in fact, all vision means, that you see the world in a way that is either different or otherwise, okay, than somebody else.
And you see opportunities that I think are that you believe are particularly important to go in and address, that you can address in a particular way. And so perspective. Our perspective at the time, this is 1993. You guys won't remember this, but the PC was Windows 3.1. CD ROM was about to be introduced.
There were no PCs with networks. Wireless technology, the no if you said some if somebody said radio, I think you would the word that would come to mind is FM radio. And so wireless technology didn't exist. The fastest microprocessor in the world was a 66 megahertz, 46DX2, and I don't think any of you would even use it in your tennis shoes today. And we would run our computers with that.
And the PC was becoming used for desktop or for office automation. Our perspective was that this particular device was going to be unique in the sense that it has the ability to run programs. And what if we gave it the benefit of running three d graphics programs so that you could explore new worlds, play games, play games. And so we started a company, and the business plan basically read something like this. We're going to take technology that was available only in the most expensive workstations.
We're going to try to make it reinvent the technology and make it inexpensive. And the killer app was video games. And so I took this idea to Sandhill Road, and they told me there was no video game market. People don't start companies to play games. And my parents might I remember calling my mom and telling her that I'm gonna start this company.
And she says, you know, what do you what do you guys do? And I said, we build these things called three d graphics chips, and and people would use them to play games. And then she said, why don't you go get a job? And so now of course, games was, we believe, going to be a very large part of the marketplace. Now we had that perspective for very obvious reasons.
We grew up in the video game generation. I was the video game generation. I was the beginning of the video game generation. And so the entertainment value of video games, computer games, was very obvious to me. And I could imagine how it could be a very large market and a very large industry.
For a lot of the people that were older, that sensibility didn't exist. And so notice, I've just described to you a perspective about the world that we had that is apparently, obviously, now true because video games is the world's largest digital media industry today. It is apparently true. And yet at the time, our common sense was unique. Nobody would have created the technology, nobody would have created the company with the sole purpose of building technology to make video games possible.
And so that was our perspective. Now we felt that video games would, of course, fuel the technology development, but you could use this technology for a whole bunch of other reasons. And one of my favorite applications, this happened about, I guess, about five years ago, a small company, struggling company here in Silicon Valley called Keyhole. And they were they created a three d virtual world, and it had no application. This three d virtual world, and you start out in space, you see the earth, you zoom into any location you wanted just by typing in the address.
I thought it was such a fabulous way of exploring the world, going to places you've never been, and they couldn't raise a penny. And so we I was so excited about the company. We put money into the company. And I went everywhere and showed that demonstration. I would tell people that this is the way we're going do search someday.
If you want to search for something, look for an address, you would type it in and we would fly you there, okay. And satellite images will continue to download. And before you know it, you're right there on the street and you might even see some buildings. That small company was eventually purchased by Google and became Google Earth. Google Earth is now the single largest downloaded, most frequently downloaded application in the history of mankind, over 200 plus million downloads.
So three d graphics could be used for a lot more than video games. Now that vision, if you will, that perspective, was unique at the time and hard to sell. And so we had to go and explain it to venture capitalists who had to figure out whether the technology was going to be possible, how big was the market because it was $0 at the time. So how do you extrapolate, how do you scope the size of a market when its apparent size was zero at the time? And you look at analyst reports and you study market research, and all of it would say approximately zero.
It would never show up. It's a noncategory, a nonmarket. And so it's incumbent upon the venture capitalists and, of course, the founders to try to figure out how to inspire each other into doing something together. And so Sequoia Capital and Sutter Hill were our venture capitalists, and we got the company going with $2,000,000 Now the question about perspective becomes very interesting in other opportunities in the company, other circumstances in the company. And let me give you some examples.
Many years later, Sequoia Capital came to me and said, you know, there's a couple of kids at Stanford, and they have this thing, and it's an Internet thing. And you just type what you're looking for, and it shows up puts up the website. And I said, yes, Yellow Pages. I mean, no duh, right? We use it.
We and there's a variety of versions of it on the web at the time. We used the Internet, just like everybody else, to do FTP and also to visit various websites. And they said, should we invest in this company? And I said, there's no freaking way they're going to make money. That stuff is free, right?
And so they said, well, we can't figure it out ourselves, but it doesn't cost much to give them $1,000,000 or 2,000,000 And they invested in a small company called what eventually became Jerry's company called Yahoo! Notice, although I had the perspective about one thing, I didn't have the perspective about something else. Just because you're a visionary doesn't mean you're a visionary by everything. Your perspective stems from your life experiences, what's commonsensical about you, what's interesting to you. And so that's important to realize that you have perspective too.
Therefore, you have vision too. Now what's interesting about these websites to follow on the Yahoo! Story, if you remember, were several other searches out there. AltaVista at the time, Xcite at the time, Lycos at the time, right? And now the question is, they're all doing search and now they did a reasonably good job.
Now comes the question is what was their perspective? How did they how was their perspective different from one another? One website thought that they were a destination. Do you guys remember that? We would be a destination, kind of like a channel.
Somebody said, in fact, since we're going to be a destination, we would serve up content. And therefore, the search part of it is a commodity. We'll outsource that. So all of the search engines, which started out as search, turned into destinations or portals, and they outsourced the search to someone else, which made it possible for Google to start. And so notice two companies doing exactly the same thing started with the exact same fundamental core technology, ended up in radically different places because they had different perspectives.
They saw the world differently. So perspective matters. Vision matters. Now in our industry, shortly after we were started, three d graphics for PCs and consumer three d graphics became the hottest, hottest thing. And so everybody in Silicon Valley was starting a three d graphics company.
We were, in 1993, the only consumer three d graphics company in the world. Silicon Graphics up the street was the professional, if you will, three d graphics company. By the end of a couple of years or so, 1995, there were probably fifty, seventy start ups doing exactly the same thing we were trying to do. And over time, we competed with about 200 companies. NVIDIA today is the only surviving computer graphics company in the world.
And so the question is then, what happened? Competition is intense. Everybody has smart people. Everybody has money. We competed with IBM.
We competed with HP. We competed with Silicon Graphics. We competed with Sony, three d FX, S3, Sirius Logic, big, small, international, local. We competed with companies all over the world. So the question is what happened.
I would argue that 300 companies armed with exactly the same technology, armed with exactly the same people, the company that wins and let's say they all execute, and they did. With 300 companies, 50% of them are going to execute at any given point in time. And so the question is, why does one survive? Well, I think that it matters to have perspective, and let me give you some examples. I always believed that you need to understand the reason why your business work.
What is the essence of your business? What makes it work? Now the foundation of my business, at its core, is semiconductor technology. Here in Silicon Valley, we usually like to refer to semiconductor technology as Moore's Law. Moore's Law is not so much a physical law as it's a law of competition.
It is a law of challenging engineers. It's a law almost of setting pace. And Moore's Law approximately gives you twice the performance every year or two. And so understanding the fundamental ingredient of our business improves by a factor of two every year and simultaneously reduces in cost by a factor of two every year. The question is what makes a survivable business.
And so our first perspective was that three d graphics was insatiable. It was insatiable. That if I made something twice as good every year, even if the customer never asked for it, even if the customer told us it was too expensive, even if the customer, when you went to float that product specification to them, told you that they're not interested. And in fact, that was the case. I took our product spec to Dell and HP and IBM and Gateway, and they all told me it was too much money.
You're well outside of the boundaries of what they were willing to pay for. When your customers all tell you not to do something, the question is then what do you do? In our case, because we had this unique perspective that three d graphics was insatiable and Moore's Law was our friend, therefore. We should make our graphics processors twice as good every year. And so for the first five years of our company, we just turned off our blinders and said, we're going to ignore customers.
Now which one of you guys are going to go through your marketing courses and the lesson that it teaches you is ignore your customers? Well, sometimes you have to ignore your customers. And the reason for that is because they don't know the nature of your business. And while the industry is being created, before there's common sense about the rules of that business, there is no way they can possibly know. And so we I took the last few million dollars of the company's money and built a chip that is way, way, way too big.
And our customers told us they we were way out of bounds on cost, and they weren't going to buy any. Until the day we showed up with the processor, we were in allocation throughout the entire life of that project, until our next generation product, which was twice its price, cannibalized the previous one. And so we grew and grew and grew for several years. Then the question became, what now? Now you guys are going to learn that innovation is a rather dangerous thing.
On the one hand, once you discover a great idea, once you discover a great idea, you rinse and repeat, rinse and repeat, and you make that idea better and better and better. Whether it's a laptop computer that you guys have here or a car or a microprocessor or, in our case, a graphics processor, we made it better and better every year. At some point, it becomes good enough. Moore's Law is a wonderful thing. Semiconductor technology enables you to make amazing leaps and bounds in technology.
And at some point, it becomes good enough. And so in the this is probably in the late '90s, about seven years into our company, maybe six, seven years into our company. I came to the conclusion that three d graphics was not going to be sustainable as an accelerator or a fixed function device that renders texture maps and polygons on the screen. That we had to change the company to make the three d graphics processor programmable so that it could be an artistic medium for expression. Now this is a weird word.
Here we are, an engineering company, and we now want to change this chip to become an artistic expression, an artistic medium so that all of the video games, so that all of the applications that were developed on our chip would be stylistically different. And we believed unless we could figure out a way to make the content richer and more interesting and stylistically different from one game developer to another game developer to another, we would limit the life of our medium. And if our medium reached its end, we, as the world leader, would also see our end. And so we decided to make the GPU programmable and make it a medium for artistic expression and invented a technology called programmable shader. Almost every single video game that you guys see today has our fingerprint on it, whether it's a Xbox three sixty or PS3 or any PC game today.
You could see elements of what programmable shaders made possible. That started a whole new innovation curve for us and kept our industry vibrant. But the crossing from one generation of technology to the next generation almost killed the company. And so that process of reinventing the company, the perspective that led us to a new idea also risks the company in the process. And those are interesting conversations that we can have.
So there's perspective matters. When a company gets larger, you guys are going to learn that as the founder or as the CEO, you have to learn new things. And many of the new things that you'll learn has to do with building products at first. And I've just talked to you about building products. Soon, you'll be talking about and learning about building companies.
And building companies means things that are soft and hard to explain, like building a company with a culture. What does that mean? How does the culture of one company different from a culture of another company? And why is it that this particular culture is better for your company and not for another? So the culture of a company is important to find out, to put your arms around and to create and develop.
How do you organize? Are you we were just talking earlier with one of the guys. Are you functionally organized? Are you organized in business units? How do you deal with multiple products and multiple geographies and multiple customers?
And so that's the company building process. It's mechanical. It's interesting. Lots of trial and error. It's organic.
People matter. Personalities matter. And if you guys are interested in talking about that, I'm happy to talk about that as well. And then I would say probably the most important thing above that is to realize that building a when you're building a company and building a product, skill matters, intellect matters, training matters, but it's not enough. The part of it that is important to realize about building companies is that it's a challenging and painful and oftentimes extraordinarily scary thing to do.
And so unless you have passion, unless you really love the process of building the company and what you're trying to do, it's going to be incredibly, incredibly challenging. And so what I would leave you with is when you're building a company, if you decide to build a company, you have to ask yourself what is the purpose that you're building the company for. Is it that you would like to build a company so that you can sell it, make a fortune? Is it that you would like to build a company so you can take it public? You're just a serial entrepreneur.
You want to build something, let somebody else run it, build something, sell it. Whatever your reason is, be honest to yourself. It turns out that, in my case, I just love the process of building things. And I love being part of something. You know, being part of NVIDIA and being part of a people and being part of a cause that's inspiring to me keeps me vibrant, and and it's something that I'm willing to do for a very long period of time.
And so I've been in it now for sixteen years, and I've learned a lot in the process of being the CEO and the founder of the company. And so maybe the thing to do is why don't I open it up now and let you guys ask me whatever type of questions you guys have. Yes, sir.
Speaker 3: What did your investors say about your idea of reinventing your company to include programmable
Speaker 2: GPUs? And Jensen, could you please
Speaker 0: repeat Yes, I will. The The question is what did our investors say about us taking the big risk of adding programmable shaders and reinventing that product category and reinventing our company. First of all, it's not a conversation you really have with your shareholders, but you do have the conversation with your management team first, your employees second, your Board of Directors third, typically is the way the process that I take. And when you're in high-tech, when you're in a technology industry, when the technology moves this fast, if you're not reinventing yourself, you're just slowly dying. You're just slowly dying, unfortunately, at at the rate of Moore's Law, which is the fastest of any rate that we know, right?
The compounded rate of Moore's Law is pretty unbelievable. We have a very
Speaker 3: successful product that was generating a lot of money.
Speaker 0: I know. So it's scary. And so you have to and this is so the question is, but we have a product that's generating a lot of money and it's very successful, how do you cannibalize it? There is a theory that if you don't cannibalize it, someone will and it surely will be cannibalized. And so if you want to be a market leader, you have to take the initiative to cannibalize your own products and have your ideas cannibalize your own ideas.
When we went from a fixed function graphics accelerator, a texture mapping engine for games like Quake III and Doom and those kind of games, to a programmable shading architecture, our first chip almost killed the company. It was called GeForce FX. I don't know if any of you have ever owned one of those. GeForce FX is a chip, is a processor that it's a baby only a mother can love. I mean, it's we took enormous chance in building GeForceFX, but it almost broke our back.
But if it if we didn't build that chip, I am sure NVIDIA would be dead today. I am absolutely certain we'd be dead. It was one of the biggest gambles in our history. We had to create instead of an API, we had to go to a processor with a language, we call it CG, and a compiler. So it's kind of like a processor.
We introduced a new programming paradigm to the world that it never understood in the beginning. And so it took a lot of evangelism, a lot of marketing, a lot of education. But CG inventing CG took us to unbelievable places. And one of our most important work today is related to using GPUs for general purpose computing. And it's extraordinary that the results of what we're seeing.
And it wouldn't have been possible if it wasn't because of CG that we started then. Okay. So you have to take these leaps. Questions? Yes, sir.
Speaker 1: Some people are sort of see that gaming is sort of moving more towards consoles and
Speaker 0: away Is from
Speaker 1: it harder to make profit with when you have to negotiate with Sony or Nintendo or whoever, some big company, than it would be when you sell individual cards?
Speaker 0: Yeah. The economics the economics of it of building anything ultimately comes down to the amount of competition you have. You don't set the price, the competition sets the price. The market doesn't set the price, the competition does. And so if your competitor wishes to build PlayStation three as much as you do, then the economics will be challenging.
And so there, you just need to decide, is an economic decision for you? For example, there's there are rumors that we were not enthusiastic in building some of the game consoles, and we were more enthusiastic in building other game consoles. It came down to this for me. I think the you have to realize what is the finite resource, what is the scarce resource that you have as a manager. The function of a manager is to allocate resources properly for the best return.
And so if you think about our resources, our resource is the finite number of extraordinary engineers and how much time they have in a day to pursue whatever opportunities that are out there. Now if the number of opportunities that are out there is less than my supply of engineers, so therefore, exceeds is less than my supply, then obviously, I'm very enthusiastic about it. But if it's the other way around, then the opportunity to build a game console at terrible economics or any project at terrible economics is simply not worth it. And so I look at it irrespective of competition. The competition sets the price, but then I get to decide whether I want to engage in that project or not.
You are in charge of your own company as the CEO, right? And so we decide whether it's economic. And once you decide, then it is what it is. So you have to be thoughtful about what is your critical resource. Do you have more of it or less of it than the market demands?
Do you have more opportunity or less opportunity than your resource can support? And then what's the appropriate return on that investment, thinking about not just your cost, but more importantly, your opportunity cost? And so we look at it from that perspective every single time.
Speaker 1: Can you tell us a little bit about the culture that you try to set at NVIDIA?
Speaker 0: Yes. That's a good question. The question is what is the can I talk about the culture that we're trying to set at NVIDIA? At the core of our company's success is innovation. Now a lot of companies say innovation is important to their company.
Invention is important to their company. However, I don't believe you can fundamentally say that innovation that you want as a CEO to nurture the spirit of innovation, to encourage innovation unless you have a culture of risk taking. We have to encourage our engineers excuse me, our marketing people, right, all of our employees to take calculated risks. In order to encourage them to take calculated risks, first of all, you have to teach them how to do that. That's a skill, a matter of skill.
Then the second part of it is a matter of courage. Most people hate to fail. Do you guys agree with that? Well, unless you guys want to be successful let me say it the positive way. If you want to be successful, I would encourage you to grow a tolerance for failure, to develop a tolerance for failure.
Now when I mean a tolerance for failure, I don't mean, gee, what Jensen just told me is sleep in until noon, okay? Don't do any of my homework. Flunk out of all my classes because that defines failure. Right? That's not what I said.
What I said is what I was trying to say is that I want you to try things even though it is impossible to calculate precisely that it would lead to success, that your instincts and your intuition is something you ought to follow. If it wasn't because of following my own instincts or the founder's instincts or many of our employees' instincts, why would we be where we are today? And why would we have invented things that the markets never had before, the world's never had before? So you have to have this culture or tolerance for risk taking. But the thing about failure is this.
If you fail often enough, you actually might become a failure. And that's different than being successful. And so the question is, how do you teach someone how to fail but fail quickly and to change courses as soon as you know it's a dead end. And the way to do that is we call it intellectual honesty. We assess on a continuous basis whether something makes sense or not.
And if it's the wrong decision, let's change our mind. And a lot of people say CEOs are always right, and they never change their mind. That doesn't make any sense at all to me, especially when it violates the first principles of what we want the company to become, an innovative company that invents amazing things, that solves problems for the world that it sometimes didn't even know it had, If you want to do that, then you have to cultivate that tolerance for risk taking. And you have to then teach people how to fail but fail quickly and inexpensively and how to be direct with each other that this is the wrong approach and what's the better approach and then, you know, be flexible enough to change courses and quick. And so that type of culture, if you will, in today's if you guys were to start a company and you were build building a website with an Internet service of some kind, Internet based service of some kind, with the competition coming from all over the world, and it's twenty fourseven, and ideas take no time to experiment, and it a particular website or a particular company could be throwing ideas out into the world 20 a day.
And so unless you are thoughtful about risk taking and being able to change your mind, reacting to the market conditions and being flexible, how are you going to stay alive? And so you could almost see what I just described in the nature of older companies and the nature of the newer companies. The modern companies, if you guys you got I'm sure you guys all go to Google's website. Almost every single application is in beta form. They're trying all kinds of stuff.
Right? They're trying all kinds of stuff. If they call it production and it doesn't work well, you guys would just be upset at them. So they call it beta. Have you noticed?
They call it beta so that they could try a lot of things. And if it fails, take it out. If it's bad, take it out. If it works, do more. And so innovation requires a little bit of experimentation.
Experimentation requires exploration. Exploration will result in failure. Unless you have a tolerance for failure, you would never experiment. And if you don't ever experiment, you would never innovate. If you don't innovate, you don't succeed.
You'll just be a dweeb. That's it. Any other questions? Yes, sir.
Speaker 2: How did you choose your cofounders at additional time?
Speaker 0: Don't ever go into business with anyone you don't deeply trust. And they were two my closest colleagues, and I trust them completely. And they're wonderful friends even today. So by the way, as a CEO, selecting people is 99% of the job.
Speaker 1: What applications do you see driving demand in the general purpose EV market, which is like that's going to be like the next big thing the semiconductor?
Speaker 0: So the question is what is what do I see driving the demand for this thing that we're pushing right now called GPU computing, using the GPU for much more than just graphics. For graphics, there's a model of graphics that we call computational graphics. So it's using programs to generate the images. The algorithms are no longer cast in the silicon. The algorithms are actually software.
And it could be ambient occlusion. It could be ray tracing. It could be all kinds of interesting algorithms that people are going to explore for the future. We observed and it was, in fact, this is to give you another example of innovation and with Stanford is fabulous. We when we invented GeForceFX, although it wasn't a very successful GPU for graphics, Researchers around the world noticed that it had a programming language called CG, C for graphics, and that you could program this GPU to do other things aside from graphics.
And it had 32 bit floating point, IEEE compatible 32 bit floating point. And so some smart researchers, many of them were here at Stanford, just bought a graphics card from Fry's and started writing programs. And they discovered that if they really worked hard and do all these algorithmic gymnastics, they could get something, an algorithm it could be nanomolecular dynamics, it could be computational fluid dynamics to run 20x faster. And they couldn't believe it. How do you speed up an application 20x?
Well, the interesting observation that we made is that we speed up three d graphics applications, which is basically something you can do in software, 1000x over a CPU all the time. So what if we took all of those parallel processors that were inside our GPUs and make it completely programmable and expose it through a programming language called C, right? C and now in the near future, C plus plus Imagine the type of problems we could help solve. And so whether it's weather prediction or seismic analysis or taking your CT scans and reconstructing the human image for the body from it, all kinds of very computationally intensive applications, we could accelerate 50x, 100x. Now just to put it in perspective, 100x is a Moore's Law time, approximately ten years.
Now to put ten years' worth of computing resources in the hands of scientists, researchers, engineers, unbelievable benefits. And so the risk was really large, however, to make our GPU even more general purpose. Because every time you make something general purpose, you know what, right, a Swiss Army knife? It's dangerous because whenever you make something general purpose or a Swiss Army knife like, you move away from your core business. It's much, much better to have a very specific niche, to have intense focus on a particular market segment.
And you guys will learn all of this in marketing. When you make something general purpose, you're all things to all people, you become, you know, what is it? What is it? Jack of all trades, master of none. Very, very dangerous move.
Now we thought that it was just too important for us not to do it, so we decided to make that move. And it's fabulous results. Okay. So those are some of the things that we're seeing now. Yes, ma'am.
Speaker 2: Mentioned you're the cofounders of NZD. So especially in the initial stage, how can you find each of your investment in the company? And how can you distribute the profit?
Speaker 0: Okay. So her question is we were friends in the beginning, we're friends now. How do we figure out who's the right position and how do we distribute the profits? Okay. And so I won't say anything funny just as I'm not misunderstood, but all of our pay were identical.
And we all had identical share in the company. So that's that's just simply fairness. Now the question becomes governance. There's the part of it which is equity. Equity is another way of saying what's fair.
Right? So we all had the same salary. All three of us were making $100,000 a year. Okay? And we all had a percentage of the company, equal percentage.
Now you can't run a company, though. You can't build a great company. When you have three people who has to vote on everything and with equal share of responsibilities. You simply can't. That becomes a leadership question.
That becomes a governance question. That becomes a management question. Right? That becomes a question about building a great company. I don't recall exactly the conversation, but I think it kind of went like this.
All right, Jensen, you're the CEO, right? Okay. That was done. That was basically the process. I think that some people are I'm not particularly I'm not from a personality perspective, I'm not particularly outgoing.
And so that's not a necessity for being a good CEO. But as a personality, I've always been able to see around the corners, if you will. I can see around the fuzzy edges. And I think CEOs and leaders need to be comfortable with ambiguity. Ambiguity meaning that, you know, what does the future look like?
Well, it's hard to say. Some people hate that. Some people just say, Jensen, tell me what you need to have done and for how with how much resource and by when. Okay? Some people rather me tell them that, hey, look, there's this there's this opportunity out there, not sure what it is, not sure how big it is, but it kind of feels like this.
Let's go figure it out, and let's build a business. Some people can are very comfortable with that. And so this this ambiguity is is important to to to to be comfortable with, I guess. And I think that all CEOs that are very successful are comfortable with ambiguity. And I'm very I am very comfortable with ambiguity.
Yes?
Speaker 4: Yes. What percentage of your initial investment was yours? Like how do you get the rest? And also, like, how many times was your proposal for an for help with the investment was rejected?
Speaker 0: Mhmm. I was 30 years old, and I'd never taken a single business class, and and I've never taken any marketing classes. And and I've never used never used the at the time, it wasn't PowerPoint. It was called Persuasion on the on the Mac. It was called Persuasion.
And so I bought a Mac so I could I could use Persuasion. And and and then I tried to create a company presentation to take it to venture capitalists. The process kind of went like this. We started my first official day of work was my thirtieth birthday, February 17, and we got the company funded. And so once we got started, the question is, what are we gonna do?
You know, how does it all work out? How do we start the company? And so we met every day, the three of us, in in one of the founder's townhouse in Fremont. And and we would get together, and and there would be nothing to do. I mean, what do you do?
You get three guys and get together. You just talk. You know? So what did you guys do last night? What did you have for dinner?
I mean, so you talk about that for about six months. Okay? And the big event of the day would be, hey, where do guys want to go to for lunch? And so Philly cheesesteak today or some Chinese food tomorrow or whatever. That would be like a big deal.
And then after a while, it was like, could you put some donuts in the fridge in the morning for when we come? I mean, so that would be a big deal for a while. And so that lasted for a few months, just the three of us like that. I know it sounds pathetic, but it's it's it's true. Because at that time, I'm reading about books on how to start companies, and I'm trying to figure out, you know, how to go raise money and, you know, what's a venture capitalist and how do you incorporate the company and those kind of things.
And pretty soon, I met a I met a met a lawyer, went to a went to a law firm called Cooley Godward, and they helped us incorporate the company. And the amount of money that he he he says, you know, we need we need some money from you so that we could price the shares and also to incorporate the company. So he says, how much money do you have in your pocket? I said, $200. So he says, okay.
Give me $200. I gave him $200. And for $200, I bought 15%, I think it was 20% of NVIDIA. So it was a good deal. Yeah.
20%. Yeah. And then I went I went back to the house, and then I went back to the condo, and and they all they both gave me $200, and they both got 20%. And that's how it worked, literally. Yeah.
It's not that much more. You know, don't here here's here's the thing. NVIDIA, I never finished my business plan. I know it. I know it.
We we never finished a business plan, never could figure out how to finish a business plan, to tell you the truth. And and if if I would have finished that book, and I I went to went to Borders and got Gordon Bell's book, How to Start a High-tech Company, it's like this thick. If I would have read the whole thing, I would have been dead now. We would have run out of money, run out of time. And so I I I read I read the first three or four chapters, and I, you know, I I gotta go to work.
And so so I I incorporated the company. They introduced us to two venture capitalists, and I just went to their office and told them what I'd like to do. The thing that gets the company funded and when you're when you get to that point, you just have to remember a few things. VCs don't invest in business plans because business plans are easy to write. I couldn't write it, but other people could.
Right? And so so they invest in this. They invest in great people. And so the so the question is is do they trust you? Your reputation matters.
Your history matters. Because because I had done so much work with Andy Bechtelsheim, which was another graduate of Stanford of Stanford and the founder of Sun and and worked with the founders of Synopsys and LSI Logic. And the and, you know, we we we were all very successful, and we did good work. Your reputation will precede you even if your business plan writing skills are inadequate. And the second thing is you need to have a vision that's sufficiently large to invest in because their statistics, their probability of success is rather low.
And if they need to put in $10,000,000 if the market is only $20,000,000 large, they'll never get that $10,000,000 back with reasonable return. But if it's a $200,000,000,000 market, then, of course, it's a rather different thing. Okay? So the size of the market. And they want to know that at least there is a clever idea that the market has never done before.
So that last part is probably second, you know, last. I said it last because also I think it's least important. You have to you might have to reinvent yourself over time. And if you want to reinvent yourself, you need to have great people. That's why great people is so important.
Yes, sir. Who
Speaker 2: were some of the people that you considered to be your mentors when you were getting started? What was some of the best advice that you got from them?
Speaker 0: So the question is what what who are some of the mentors, and what were the best advice I got? I I truly believe that if you wanna be successful, you a successful habit is to have the capacity and the willingness to learn from just about anybody. And I do. I learn from just about anybody. And it could be a little thing, could be a big thing.
You know, if it wasn't because of my kids, I would be I would miss the whole Internet age. You know, I would have missed YouTube and Facebook and Twitter and you know, I mean, without so you you need to you need to know that that the world changes, and and you wanna be able to learn from just about anybody. And so I I'm surrounded with extraordinarily talented executives and professionals of of all walks of life. And so you just have to make sure that you're you're willing to learn from just about anybody. Some of the some of the great advice that I've I've had over the years, focus.
Laser beam focus. You know, don't do too much. Do a few things well and do it with extraordinary intensity. And focus matters. If you look at what I do with my time, I wake up in the morning, and the first thing of my time is NVIDIA, and the last thing I do is NVIDIA.
And I do that 20 fourseven. And if I could figure out a way to do that for another fifty years, we're going to be in good company. Yes, ma'am?
Speaker 2: Successful entrepreneur and female,
Speaker 5: so what's your greatest challenge at the current?
Speaker 2: With your ambiguity as CEO, what's your best and worst estimation of the future of you and your company?
Speaker 0: Okay. So her question is she started out by saying that I'm a successful entrepreneur. And what are my biggest challenges now? And considering that I like ambiguity, what's my best estimate of the future for the company or and and for myself? The biggest challenge with building a company is the reinvention of the company.
Every successful thing needs to be torn down at some point and be rebuilt. It is unfortunate but true. And the reason for that is because the technology either gets good enough, and therefore, you have to reinvent. And sometimes the invention process is disruptive. Sometimes it's, in fact, destructive.
And it could be it could destroy you could destroy what you have built in the past. And so the reinvention process is very challenging. It's gut wrenching. It takes a lot of courage, and it it really tests your conviction. In in the technology industry, reinventing the company every ten years is almost a necessary thing.
And so that's when I say challenging, I don't mean bad challenging. I think that's fun challenging. I love the process of reinvention. Okay. So that's fun challenging.
What's my best forecast for our company? I think that NVIDIA has the opportunity to become one of the most important technology company companies in the world. And I hope that I hope that it does. And my best forecast for me is that I am 80 years old, and I'm here talking to students, and I hope I'm still the CEO. Yes, sir.
Speaker 6: Questions about the first few years of when you start, very critical. You actually survive and make sure your cash is positive.
Speaker 0: Yes. And
Speaker 6: the constant cash flow, cash is king. So in NVIDIA case, how do you manage that cash
Speaker 0: or Yes. So the gentleman's question is has to do with in the beginning, survival is important, cash is king. Just so that there's no no no ambiguity about this, survival is always important. Cash is always king. And so as the CEO, you're either making money, saving money, or raising money.
And and if you're not making money, raising money and saving money, you ought to be doing those three things. It's a it's a just stay focused on those three things. And so when you're during the beginning, in the early days, I was raising money all the time. As soon as I was done raising this round of money, I got to raise more money. You know, you're always raising money.
Just maybe maybe there was a week break in between, but I was raising money all the time. I was as a start up, you're always going out of business. Right? That's the definition of a start up, an enterprise that is nearly out of business all the time.
Speaker 2: That's the definition of a start up. Question? Yes, ma'am. Are you prepared for the leadership succession? Well,
Speaker 0: because I want this the question is how do I deal with leadership succession as a CEO and for our company? Well, I want this job until I'm 80, I just said. No. I'm just kidding. One of the primary roles of a CEO, in order to grow the company in order to grow the company, in order to make make NVIDIA one of the most important technology companies in the world and make make significant contribution to society.
In order to do that, you have to cultivate new leaders so that they can have new ideas and grow new businesses and and, you know, maybe run a different geography, run a new new different product line. And so I spend most of my time these days, most of my time these days, sitting with our general managers and sitting with our leaders and helping them think through strategies and helping them think through challenges and helping them think through product roadmaps and helping them think through transitions and, you know, team building, organization creation, you know, how to manage, how to create processes that last the test of time. So so, you know, these things are are lessons that that I'm supposed to pass on, and I do. And I spend a lot of my time doing that. I believe I believe this, that succession planning by a priori picking out three people that the Board should consider in the case that I get run over by a bus is a toxic, toxic process.
I know that it has been it has been thought of as a methodology succession planning, but I think it's just very toxic for the environment because everybody is trying to figure out who got selected and who didn't. I think that it's a much, much better process to focus on ultimately developing the next generation of leaders so that in the case that something happens, where I'm not the right CEO anymore, there are many choices for the Board to choose from, including outside. So those are I think the company building process, focusing on that is the positive way of thinking about succession planning. But good question. How about I take one question, and that's it?
Okay. Yes, ma'am.
Speaker 5: I don't know if it's a good question, but
Speaker 0: It's okay.
Speaker 5: One thing that is a problem with Latin America
Speaker 0: Any question from our Latino crowd here is fabulous.
Speaker 2: I am a model.
Speaker 5: I have many
Speaker 0: friends who are
Speaker 5: engineers. They have many, many brilliant ideas, but they don't feel motivated because they they don't have money and they were looking for for people who could make these ideas real, and they are only working for other people, and they don't realize this idea. I I don't know what to see that is the problem.
Speaker 0: Mhmm. Well, first of all, so her her I think let me see if I can paraphrase the question. You you know you know friends who have a lot of ideas, but they also have a lot of money, and so they don't really feel that motivated to do something with those ideas. I you know you know, first of all, money I hope that if I if I leave you with anything, money is the only singular reason not to start the company. Because starting companies are it's a very, very unlikely probability for success.
And so if you're looking for if that is your reason for doing it, you will likely regret the experience. You should start the company and build a company because you just you so believe in your idea, you so you're so passionate about it, and you want to build something great. The only reason you want to do it. It just comes along. I have plenty of money and but it doesn't motivate me nor does it demotivate me because it wasn't And so I would urge you to ask yourself and for your friends to ask ask themselves, what is the purpose that they want to start a company?
Ideas. Ideas are a dime a dozen. There's so many ideas. If you want ideas, you're going to get a lot of ideas in this room. And so ideas don't really matter.
You have to have a perspective that's unique, that you feel really strongly about it, that you're willing to persevere almost any challenge. Okay? Thank you everybody.
@lkondeth - luqman kondeth
@MilkRoadAI @grok summarize this into 10 bullet points and then give readers 5 actions items
@grok - Grok
Here's a summary of Jensen Huang's 2011 Stanford lecture in 10 bullet points: - Building companies is gratifying but extremely hard, requiring unique perspectives over grand visions. - NVIDIA started in 1993 with 3 engineers, betting on 3D graphics for PCs and video games amid skepticism. - Initial idea faced rejection from VCs and family, but founders persisted due to their gaming generation insight. - 3D graphics enabled apps like Keyhole (later Google Earth), showing tech's broader potential. - Competed against 200+ firms; survived by understanding Moore's Law and insatiable demand for better graphics. - Sometimes ignored customers to innovate, like building overpriced chips that became hits. - Reinvented GPUs as programmable for artistic expression, nearly killing the company but sparking growth. - Emphasized tolerance for failure, intellectual honesty, and quick course corrections. - Passion, not money, drives success; focus on large markets and constant reinvention. - Cultivate leaders, embrace ambiguity, and learn from everyone for long-term impact. 5 Action Items: 1. Identify your unique perspective on an opportunity and pursue it relentlessly. 2. Build tolerance for calculated risks and fail quickly to innovate. 3. Reinvent your approach periodically, even if it disrupts current success. 4. Foster intellectual honesty in your team for adaptive decision-making. 5. Prioritize passion and learning over financial motives when starting ventures. (487 chars)