reSee.it - Related Video Feed

- XAI is two and a half years old and has achieved rapid progress across multiple domains, outperforming many competitors who are five to twenty years older and have larger teams. The company claims to be number one in voice, image and video generation, and to be leading in forecasting with Grok 4.20. Grok is integrated into apps like Imagine and Grokipedia, with Grokipedia positioned to become Encyclopedia Galactica—much more comprehensive and accurate than Wikipedia, including video and image data not present on Wikipedia. - XAI has achieved a 100,000-hour GPU training cluster and is about to reach 1,000,000 GPU-equivalent hours in training. The company emphasizes velocity and acceleration as the key drivers of leadership in technology. - The company outlines a four-area organizational structure: Grok Main and Voice (the main Grok model), a coding-focused model (Grok Code), an image and video model (Imagine), MacroHard (digital emulation of entire companies), and the infrastructure layers. - Grok Main and Voice will be merged into one team. In September 2024, OpenAI released a voice product, but XAI states it started later and, in six months, developed an in-house model surpassing OpenAI, with Grok in over 2,000,000 Teslas and a Grok voice agent API. The aim is to move beyond question answering toward building and deploying broader capabilities, such as handling legal questions, generating slide decks, or solving puzzles. - Product vision stresses that Grok Main’s intent is genuinely useful across engineering, law, and medicine, aiming to be valuable in a wide range of areas necessary to understand the universe and make things useful. - MacroHard is described as the effort to digitally emulate entire companies, enabling end-to-end digital output and the emulation of human workers across various functions (rocket design, AI chips, physics, customer service, etc.). MacroHard is presented as potentially the most important project, with the Roof of the training cluster bearing the MacroHard name. The team emphasizes that most valuable companies produce digital output and that MacroHard could replicate the outputs of companies like Apple, Nvidia, Microsoft, and Google, among others, across multiple domains. - Imagine focuses on imaging and video generation; six months into the project, Imagine released v1 and topped leaderboards across several metrics. The team highlights rapid iteration with multiple product updates daily and model updates every other week. Users are generating close to 50,000,000 videos per day and 6,000,000,000 images in the last 30 days, claiming this surpasses other providers combined. The goal is to turn anything you can imagine into reality. - Hakan discusses longer-form video capabilities, predicting end-of-year capabilities for generating 10 to 20-minute videos in one shot, with real-time rendering and interaction in imagined worlds. The expectation is that most AI compute will be real-time video understanding and generation, with XAI leading in this trajectory and continuing to improve Grok code toward state-of-the-art performance within two to three months. - MacroHard details: the team envisions building a fully capable digital human emulator to perform any computer-based task, including using advanced tools in engineering and medicine, like rocket engines designed by AI. The project is framed as a response to the remaining gap between AI and human capability in this domain, making it a high-priority area for recruitment of top talent. - XChat and X Money are described as major products in development. XChat is planned as a standalone standalone messaging app with full features (encrypted messaging, audio and video calls, screen sharing, etc.), with no advertising or hooks in Grok Chat. X Money is currently in closed beta within the company, moving toward external beta and then worldwide, intended to be the central hub for all monetary transactions, including mortgages, business loans, lines of credit, stock ownership, and crypto. - The presentation also emphasizes the synergy between XAI and SpaceX, noting that SpaceX has acquired xAI and that orbital AI data centers are being pursued to dramatically increase available AI training compute. FCC filings indicate plans to launch a million AI satellites for training and inference, with annual launches potentially reaching 200–300 gigawatts per year, and longer-term goals including moon-based factories, satellites, and a mass driver to launch AI satellites into orbit. The mass driver on the moon is described as a path to exponentially greater compute, potentially reaching gigawatts or terawatts per year, with the broader ambition of enabling a self-sustaining lunar city and interplanetary expansion. - The overall message stresses extraordinary progress, a relentless push toward greater compute and capability, and aggressive growth in user adoption and product scope. The company frames its trajectory as a fundamental shift toward real-time, scalable AI that can transform work, communication, and the management of digital assets across the globe and beyond Earth.

all of the companies here are building just making huge investments in in the country in order to build out data centers and infrastructure to power the next wave of innovation. "How much are you spending, would you say, over the next few years?" "Oh, gosh. I mean, I think it's probably gonna be something like, I don't know, at least $600,000,000,000 through '28 in The US. Yeah. It's a lot." "It's it's significant. That's a lot." "Thank you, Mark. It's great to have you. Thank you."

This is the alchemy of intelligence. This newly manufactured intelligence will spawn a new chapter of unprecedented productivity and development, and that will serve to improve human quality of life. The IDC estimates that AI will generate $20,000,000,000,000 in economic impact by 2030. So even if you can earn a small slice of that, that hundreds of billions of dollars of investment will earn an amazing return. For each dollar invested into, business related AI, it's expected to generate $4.60. As my friend Jensen would say, the more you buy, the more you save. Or in this case, the more you buy, the more you make. And we can grow the pie together and usher in a new era of AI driven

The speaker believes humanoid robots will be the biggest product ever, with insatiable demand, like having a personal C-3PO and R2-D2. They mentioned that "tens of billions of robots" is at least a decade away, but the growth will be very fast. The speaker's goal is to produce a million robots by 2029 or 2030, which they consider a reasonable target, and then move towards sustainable abundance.

A VP at NVIDIA flagged that, for months, his team’s costs were higher for AI than for humans, and the issue was emerging “in droves.” Uber’s CTO said he had already blown out his entire 2026 budget on AI-related costs, implying spending on AI exceeded spending on human workers. Startup founders were also described as “bragging” about high AI bills as a form of demonstrating they were “ahead,” essentially “flexing” that they were blowing cash on AI. The original purpose of AI spending was described as reducing costs and expanding profits, especially for public companies, but it was characterized as unclear whether that would remain tenable over time. One factor referenced was “the curve,” with discussion tied to the idea of costs not necessarily declining as expected. Data cited in the report stated that worldwide IT spending is expected to rise by 13 and a half percent this year compared to last year, exceeding $6,000,000,000,000. The question raised was where the money is going. A significant portion was said to go toward token costs, described as the currency of AI use, and toward subscriptions, including enterprise contracts with OpenAI and Anthropic. It was also described as flowing to AI labs. The transcript added that ordinary queries entered into AI do not cost very much, but costs rise for activities like coding or using an autonomous agent overnight. It further stated that some companies, especially tech firms like Meta, encourage high token use because they want to “see and seem like they’re really ahead in the AI race.”

The chart referenced is a few months old, and it’s worth examining the recent developments to understand the current situation better.

We can't track $2.3 trillion in transactions. That's two trillion, three hundred billion dollars.

The discussion centers on the ongoing battle between Google and Nvidia in AI hardware, with Google focusing on TPUs and Nvidia offering a full GPU stack. Blackwell, Nvidia’s next-generation chip, faced a delayed first iteration (Blackwell 200) and was followed by a difficult, complex product transition from Hopper to Blackwell. The transition required moving from air cooling to liquid cooling, increasing rack weight from about 1,000 pounds to 3,000 pounds, and boosting power from roughly 30 kilowatts to about 130 kilowatts. The speaker likens the change to a homeowner needing to overhaul power infrastructure, cooling, and the physical environment to support a new, denser, heat-intensive system. As a result, many Blackwell SKUs were canceled, and true deployment only began in the last three or four months, with scale-out starting recently. Google is viewed as having a temporary pre-training advantage and, notably, being the lowest-cost producer of tokens. The speaker argues that, in AI, being the low-cost producer has become a meaningful factor, a rarity in tech markets. This dynamic enables Google to “suck the economic oxygen out of the AI ecosystem,” making life harder for competitors and potentially altering strategic calculations across the industry. Two key upcoming shifts are highlighted. First, the first models trained on Blackwell are expected in early 2026, with the first Blackwell model anticipated to come from XAI. The rationale is that even with Blackwells available, it takes six to nine months to reach Hopper-level performance due to Hopper’s tuning, software, and architectural familiarity. Since Hopper outperformed its predecessor after six to twelve months, Nvidia aims to deploy GPUs rapidly in coherent data-center clusters to work out bugs fast, enabling Blackwell scaling. XAI is positioned to accelerate this process by building data centers quickly and helping debug for others, thereby likely producing the first Blackwell model. Second, the GB200’s difficulties gave way to the GB300, which is drop-in compatible with GB200 racks. The GB300 will be deployed in data centers capable of handling the new heat and power requirements, replacing not the GB200s but fitting into existing, scalable racks. Companies using GB300s may become the low-cost token producers, especially if they’re vertically integrated; those paying others to produce tokens would be disadvantaged. These hardware developments have broad strategic implications for Google: if it maintains a decisive cost advantage and potentially operates AI at negative margins (e.g., -30%), it could continue to extract economic oxygen from the market and solidify a dominant position, affecting funding dynamics for competitors. The shift from training to inference with Blackwell deployments and the arrival of Rubin are anticipated to widen the gap versus TPUs and other ASICs, altering the economics and competitive landscape of AI at scale.

Speaker 0 notes that latest AI chips use somewhere between six and ten times the amount of memory of the earlier H100, leading to a huge consumption requirement and creating a memory bottleneck. Building a new memory fabrication plant takes between three and five years, intensifying the supply constraint. Samsung, the world’s largest memory chip maker, will be impacted negatively because it also serves smartphones, PCs, and TVs; while it gains in some areas, it loses in others, and the problem is expected to worsen. Hynix, another memory producer, says it will get worse before it gets better in terms of being able to supply to meet demand. Overall, memory supply issues are a major concern for the industry, with wide-reaching implications. Speaker 1: Investor sentiment around AI disruption on management calls is rising sharply. The question is how this translates to markets. The speaker confirms there is nervousness, in part because it’s not clear how AI will affect business models. A concrete example mentioned is CBRE, the large commercial real estate firm, which said it can use AI to reduce its research costs by 25%. Despite this potential internal efficiency, CBRE’s stock was hit hard, because investors wonder what external AI models could do for even lower costs, and fear that the competitive advantages from internal efficiency might be replicated externally at a much lower price. The overarching concern is the unknowns: while companies are attempting to address AI head-on, there is a risk that others can replicate or surpass the benefits quickly, given the speed and breadth of AI developments, making it hard to keep up.

This infrastructure, like the Internet and electricity, requires factories, but these are unlike data centers of the past, which are part of a trillion-dollar industry providing information and storage. While originating from the same industry, these new factories will be completely separate from the world's data centers. These AI data centers are better described as AI factories. Applying energy to them produces something valuable: tokens.

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.

The speaker discusses building AI factories to run companies, describing it as more significant than buying a TV or bicycle. They state that the world is building trillions of dollars worth of AI infrastructure over the next several years, characterizing this as a new industrial revolution. The speaker compares AI factories to historical innovations like the steam engine and railroads, but asserts that AI factories are much bigger due to the current scale of the world economy. They claim that with a $120 trillion global GDP, AI factories will underpin a substantial portion of it, suggesting that trillions of dollars in AI factories supporting a hundred trillion dollars of the world's GDP is a sensible proposition.

- The conversation centers on how AI progress has evolved over the last few years, what is surprising, and what the near future might look like in terms of capabilities, diffusion, and economic impact. - Big picture of progress - Speaker 1 argues that the underlying exponential progression of AI tech has followed expectations, with models advancing from “smart high school student” to “smart college student” to capabilities approaching PhD/professional levels, and code-related tasks extending beyond that frontier. The pace is roughly as anticipated, with some variance in direction for specific tasks. - The most surprising aspect, per Speaker 1, is the lack of public recognition of how close we are to the end of the exponential growth curve. He notes that public discourse remains focused on political controversies while the technology is approaching a phase where the exponential growth tapers or ends. - What “the exponential” looks like now - There is a shared hypothesis dating back to 2017 (the big blob of compute hypothesis) that what matters most for progress are a small handful of factors: compute, data quantity, data quality/distribution, training duration, scalable objective functions, and normalization/conditioning for stability. - Pretraining scaling has continued to yield gains, and now RL shows a similar pattern: pretraining followed by RL phases can scale with long-term training data and objectives. Tasks like math contests have shown log-linear improvements with training time in RL, and this pattern mirrors pretraining. - The discussion emphasizes that RL and pretraining are not fundamentally different in their relation to scaling; RL is seen as an RL-like extension atop the same scaling principles already observed in pretraining. - On the nature of learning and generalization - There is debate about whether the best path to generalization is “human-like” learning (continual on-the-job learning) or large-scale pretraining plus RL. Speaker 1 argues the generalization observed in pretraining on massive, diverse data (e.g., Common Crawl) is what enables the broad capabilities, and RL similarly benefits from broad, varied data and tasks. - The in-context learning capacity is described as a form of short- to mid-term learning that sits between long-term human learning and evolution, suggesting a spectrum rather than a binary gap between AI learning and human learning. - On the end state and timeline to AGI-like capabilities - Speaker 1 expresses high confidence (~90% or higher) that within ten years we will reach capabilities where a country-of-geniuses-level model in a data center could handle end-to-end tasks (including coding) and generalize across many domains. He places a strong emphasis on timing: “one to three years” for on-the-job, end-to-end coding and related tasks; “three to five” or “five to ten” years for broader, high-ability AI integration into real work. - A central caution is the diffusion problem: even if the technology is advancing rapidly, the economic uptake and deployment into real-world tasks take time due to organizational, regulatory, and operational frictions. He envisions two overlapping fast exponential curves: one for model capability and one for diffusion into the economy, with the latter slower but still rapid compared with historical tech diffusion. - On coding and software engineering - The conversation explores whether the near-term future could see 90% or even 100% of coding tasks done by AI. Speaker 1 clarifies his forecast as a spectrum: - 90% of code written by models is already seen in some places. - 90% of end-to-end SWE tasks (including environment setup, testing, deployment, and even writing memos) might be handled by models; 100% is still a broader claim. - The distinction is between what can be automated now and the broader productivity impact across teams. Even with high automation, human roles in software design and project management may shift rather than disappear. - The value of coding-specific products like Claude Code is discussed as a result of internal experimentation becoming externally marketable; adoption is rapid in the coding domain, both internally and externally. - On product strategy and economics - The economics of frontier AI are discussed in depth. The industry is characterized as a few large players with steep compute needs and a dynamic where training costs grow rapidly while inference margins are substantial. This creates a cycle: training costs are enormous, but inference revenue plus margins can be significant; the industry’s profitability depends on accurately forecasting future demand for compute and managing investment in training versus inference. - The concept of a “country of geniuses in a data center” is used to describe the point at which frontier AI capabilities become so powerful that they unlock large-scale economic value. The timing is uncertain and depends on both technical progress and the diffusion of benefits through the economy. - There is a nuanced view on profitability: in a multi-firm equilibrium, each model may be profitable on its own, but the cost of training new models can outpace current profits if demand does not grow as fast as the compute investments. The balance is described in terms of a distribution where roughly half of compute is used for training and half for inference, with margins on inference driving profitability while training remains a cost center. - On governance, safety, and society - The conversation ventures into governance and international dynamics. The world may evolve toward an “AI governance architecture” with preemption or standard-setting at the federal level, to avoid an unhelpful patchwork of state laws. The idea is to establish standards for transparency, safety, and alignment while balancing innovation. - There is concern about autocracies and the potential for AI to exacerbate geopolitical tensions. The idea is that the post-AGI world may require new governance structures that preserve human freedoms, while enabling competitive but safe AI development. Speaker 1 contemplates scenarios in which authoritarian regimes could become destabilized by powerful AI-enabled information and privacy tools, though cautions that practical governance approaches would be required. - The role of philanthropy is acknowledged, but there is emphasis on endogenous growth and the dissemination of benefits globally. Building AI-enabled health, drug discovery, and other critical sectors in the developing world is seen as essential for broad distribution of AI benefits. - The role of safety tools and alignments - Anthropic’s approach to model governance includes a constitution-like framework for AI behavior, focusing on principles rather than just prohibitions. The idea is to train models to act according to high-level principles with guardrails, enabling better handling of edge cases and greater alignment with human values. - The constitution is viewed as an evolving set of guidelines that can be iterated within the company, compared across different organizations, and subject to broader societal input. This iterative approach is intended to improve alignment while preserving safety and corrigibility. - Specific topics and examples - Video editing and content workflows illustrate how an AI with long-context capabilities and computer-use ability could perform complex tasks, such as reviewing interviews, identifying where to edit, and generating a final cut with context-aware decisions. - There is a discussion of long-context capacity (from thousands of tokens to potentially millions) and the engineering challenges of serving such long contexts, including memory management and inference efficiency. The conversation stresses that these are engineering problems tied to system design rather than fundamental limits of the model’s capabilities. - Final outlook and strategy - The timeline for a country-of-geniuses in a data center is framed as potentially within one to three years for end-to-end on-the-job capabilities, and by 2028-2030 for broader societal diffusion and economic impact. The probability of reaching fundamental capabilities that enable trillions of dollars in revenue is asserted as high within the next decade, with 2030 as a plausible horizon. - There is ongoing emphasis on responsible scaling: the pace of compute expansion must be balanced with thoughtful investment and risk management to ensure long-term stability and safety. The broader vision includes global distribution of benefits, governance mechanisms that preserve civil liberties, and a cautious but optimistic expectation that AI progress will transform many sectors while requiring careful policy and institutional responses. - Mentions of concrete topics - Claude Code as a notable Anthropic product rising from internal use to external adoption. - The idea of a “collective intelligence” approach to shaping AI constitutions with input from multiple stakeholders, including potential future government-level processes. - The role of continual learning, model governance, and the interplay between technology progression and regulatory development. - The broader existential and geopolitical questions—how the world navigates diffusion, governance, and potential misalignment—are acknowledged as central to both policy and industry strategy. - In sum, the dialogue canvasses (a) the expected trajectory of AI progress and the surprising proximity to exponential endpoints, (b) how scaling, pretraining, and RL interact to yield generalization, (c) the practical timelines for on-the-job competencies and automation of complex professional tasks, (d) the economics of compute and the diffusion of frontier AI across the economy, (e) governance, safety, and the potential for a governance architecture (constitutions, preemption, and multi-stakeholder input), and (f) the strategic moves of Anthropic (including Claude Code) within this evolving landscape.

Floating point numbers are being produced at high volume and have value because they represent artificial intelligence. These numbers can be reformulated into various outputs like languages, proteins, chemicals, graphics, images, videos, and robotic movements. In the previous industrial revolution, water was converted into steam and then electrons. Now, electrons are input, and floating point numbers are the output. Similar to the last industrial revolution where the value of electricity was not immediately understood, the significance of these floating point numbers is emerging.

Jensen Huang (NVIDIA) discusses how the amount of compute—and the energy required for that compute—is likely to increase dramatically, moving from “a hundred times” to “a thousand times” compared with current levels. He frames future computing as two simultaneous shifts: it will be intelligent and contextually aware with generative outputs, and it will be continuous rather than based on prerecorded retrieval that is initiated only when prompted. The discussion contrasts concerns about today’s AI being “backward looking” and copying previous work, potentially leading to feedback loops where people rely on AI and become stagnant without new regenerative creativity. Jensen Huang’s described future addresses this by arguing that software will not remain static code stored on a hard drive; instead, people will ask AI to write software in real time as needed (for example, generating a Photoshop clone to edit an image or generating an original movie tailored to a preference). Creating such continuous generative experiences is said to require a tremendous amount of energy—“a thousand times more” than today’s levels. Speakers note that existing energy sources cannot easily support this scale. The conversation states that it cannot be done on hydrocarbons, not even on nuclear due to long build-out time, and not on solar because current energy sources are insufficient. It also emphasizes efficiency: having the ability to use vastly more energy does not mean it should be used, and continuous regeneration is not always the more efficient approach. Speaker 0 then argues for limiting market cap and having these groups invest themselves without government backing or government liability protection, suggesting a free-market approach rather than government-directed competition framed as an arms race. Speaker 2 responds that pursuit of “superintelligence” requires centralized power and therefore cannot be decentralized. The conversation claims this centralized effort is being directed toward a quest for superintelligence connected to world domination and competition, particularly framed as an attempt to “beat China,” and concludes that once superintelligence is achieved, humanity’s fate would be in question.

The speaker reframes computers as AI factories, which produce tokens, numbers. These AI factories should be used for three fundamental things, with the first being to train the next frontier model so you can build the best AI and get to market first. The goal is to train it as fast as possible. Regarding performance, Rubin is described as a 4x leap compared to Blackwell, meaning the fourfold improvement could be achieved in one month instead of four months.