reSee.it - Related Video Feed

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."

Cloud providers are investing heavily in data centers to support AI. Microsoft, Meta, Google, and Amazon collectively spent $125 billion on data centers in 2024. These data centers require increasing power to train and operate AI models. Data center power demand is projected to rise by 15-20% annually through 2030 in the US due to the AI boom. The average data center, around 100 megawatts, consumes the equivalent energy of 100,000 US households.

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.

At the end of 2018, there were 430 hyperscale data centers, growing to 597 by 2020 and 992 by the end of 2023. Currently, there are over 1,000, with an additional 100 planned. Microsoft announced a $50 billion investment in data centers from July 2023 to June 2024, aiming to accelerate server capacity expansion. Amazon committed $150 billion to data center growth, with $50 billion allocated for U.S. projects in the first half of 2024. These companies are focused on expanding their operations and meeting increasing computational demands, prioritizing profit over potential social benefits.

This segment juxtaposes everyday living with the expanding footprint of data centers and the perceived costs of the AI revolution. In the home, Speaker 0 demonstrates a high-pressure cold water line used for storage and filling tanks, noting that the water is needed for flushing toilets. Speaker 1 observes sediment in the water coming from the faucet and asks if that sediment comes from the data center, to which Speaker 0 confirms—“Yeah. And this is what's in all the pipes.” Speaker 2 adds that the well itself is likely “20,000” (units implied) and that this figure doesn’t include costs for replacing fixtures, faucets, toilets, and pipes underneath the house. The cumulative burden feels overwhelming, as Speaker 0 describes feeling up against a “huge wall that you can't penetrate” and a sense that “they don't care.” Turned outward, the report spotlights Meta’s new data center in Mansfield, Georgia: a 2,000,000 square foot facility intended to power AI tools such as ChatGPT and other technologies integrated into daily life. Data centers are described as a hot item and an exciting asset class, with Meta building a two gigawatt-plus data center so large it could cover a significant part of Manhattan. Yet this growth comes with significant costs: light and noise pollution, environmental impacts, and potential rises in energy bills. The facilities exert extraordinary demand on the power grid and require entirely new infrastructure. Speaker 0 voices concern that the burden should be borne by those responsible, not residents. Speaker 2 argues that large tech companies—Meta, Amazon, Microsoft—“can afford to pay for their own generation,” urging people to search their profits. The reporters pursued two central questions in Georgia: “What’s the true cost of the AI revolution, and who should be paying for it?” They note the proximity of a house to the data center—“less than 400 yards.” The profile then introduces Beverly and Jeff Morris, who purchased their home near downtown Atlanta in 2016, with deep roots in the community. Beverly characterizes country living as her peace and therapy, while Jeff notes he was raised about five miles away.

- 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.

In 2014, the speaker's company hired Manuela Veloso from Carnegie Mellon to run machine learning. They have a 200-person AI research group and spend approximately $2 billion on AI, with about 600 end use cases. This number of use cases is expected to double or triple next year. The company moved AI and data out of the technology department because it was deemed too important. The head of AI and data now reports to the speaker and the president. The company focuses on accelerating AI development and tests extensively, collaborating with many people. AI will change everything.

Mike Adams discusses concerns about the global build-out of data centers and presents a multi-part theory about their purpose and implications. He notes that a tweet he posted went viral, drawing responses from figures like Jimmy Dore and Rizwan Virk. He frames his talk as a theory, not a confirmed prediction, and plans to cover it in two parts. Key data and observations - There are about 11,000 existing data centers worldwide. The map and graphics Adams shares focus on 3,000 new or planned/construction sites, showing locations, size, power use, water use, land area, and investment needs. - In Piketon, Ohio, and other U.S. sites (including multiple facilities in Ohio and Texas), as well as Abu Dhabi, Shanghai, Tokyo, Malaysia, and other locations, there are large data centers under construction or announced. The lines in the AI-generated map may mis-point geographically, but the cities and nations listed are accurate. - The aggregate planned/under-construction capacity projects to about 190 gigawatts of power draw once completed. - The projected annual power consumption for these new centers would exceed 1,200 terawatt-hours per year, which Adams compares to about 10% of all power produced by China. - The centers would occupy over 1,000 square kilometers and use about 15+ billion liters of water per year, with some water potentially drawn from neighborhoods or households. Revenue and purpose questions - Adams argues there is not enough AI business, web hosting, data storage, or overall demand to justify the scale of the investment, implying the revenue model may be inadequate to pay back these projects. - He contrasts various high-profile tech figures—Tesla, Sam Altman, and Mark Zuckerberg—suggesting that the motives behind these data center buildups extend beyond serving immediate consumer compute needs, hinting at broader or longer-term strategic aims. Foundational ideas about AI and intelligence - He cites Jan LeCun (referenced as a leading AI researcher) arguing that the current structure of large language models (LLMs) is a dead end for achieving AGI or superintelligence due to gaps in physical-world understanding, memory, and long-term planning. Memory is said to be improving with newer context-handling approaches, but physical-world understanding and planning are highlighted as critical gaps. - LeCun’s idea mentioned is the development of world models and JEPPA architectures that learn from sensory inputs to understand and interact with the physical environment, rather than solely processing language statistics. - Adams suggests that the only viable path to practical superintelligence is to train AI systems in simulated three-dimensional worlds, where physics, gravity, time, light, touch, and other sensory inputs are experienced. He argues that simulated worlds can run at speeds far faster than the real world, limited only by compute and hardware bandwidth. - He mentions NVIDIA’s announced world simulator for training robots as an example of three-dimensional world simulations used for reinforcement learning and rapid iteration. - The concept of digital worlds is tied to the idea of digital evolution or Darwinism: billions of parallel simulated worlds could nurture AI entities that grow and potentially be summoned into our three-dimensional reality. He notes that a simulation-based approach could produce agents whose capabilities enable real-world deployment after learning in fast, rich simulations. - Adams discusses practical applications of three-dimensional simulations beyond AI self-improvement, including autonomous vehicle testing (synthetic data), manufacturing and robotics on factory floors, military scenario planning, surgical robotics, and pilot training. He emphasizes that the more realistic the simulation, the more reliable the results for real-world tasks and decisions. - He invokes the simulation hypothesis, suggesting a link between building simulated worlds and the possibility that our own reality could be a simulation. He plans to address evidence for the simulation hypothesis in part two, along with how simulated beings might be “summoned” into our world. Closing - Adams signals a two-part structure, with Part 1 covering data center build-out, AI constructs, and the simulation framework; Part 2 promising to address the simulation hypothesis with evidence and the idea of summoning advanced AI from simulations into the real world. Note: Promotional content regarding gold and silver investments and Battalion Metals has been omitted from this summary to align with content-avoidance requirements.