Gavin Baker: Why AI Demand Is Outrunning Compute Supply

Gavin Baker discusses the accelerating demand for AI, emphasizing that it is outpacing compute supply, which could lead to significant global shortages affecting up to 500 million people. He notes that while the AI market may support multiple successful companies, historical trends indicate that major technology shifts often result in overinvestment. Baker highlights the current constraints on compute resources, with usage concentrated among a limited number of users, and suggests that companies like Anthropic and OpenAI must balance their missions with profitability.

Baker expresses confidence in the performance of the leading AI labs, particularly NVIDIA, which plays a central role in the AI landscape. He points out that these labs are heavily investing in compute, often subsidizing their products, and achieving high returns on aggressive spending. He estimates that there are around 30 million heavy paying users of AI, although he believes this figure may be overstated. The productivity gains from AI are significant, as demonstrated by tools like Grockbot, which can complete tasks in seconds that previously took hours.

Baker warns of potential compute inequality if data centers are not built, which could exacerbate disparities in access to AI resources. He addresses environmental concerns about data centers, arguing that they are largely unfounded, particularly regarding water consumption. He stresses the importance of the AI industry effectively communicating its benefits to the public and acknowledges the current challenges surrounding AI regulation in the U.S.

In discussing the future of compute supply, Baker highlights the potential for orbital projects, particularly with SpaceX's Starship, which could significantly reduce launch costs. He predicts that as terrestrial cooling and power costs rise, the economics of orbital compute will become more favorable, leading to a greater share of global compute being located in orbit. Baker believes that skepticism about orbital compute is misplaced, as advancements in terrestrial compute are already underway.

Baker also addresses the competitive landscape, suggesting that the dominance of OpenAI and Anthropic may overlook the potential of reusable Starship technology. He envisions a future where no single AI model dominates, but rather an ensemble of models is utilized, with companies fine-tuning open-source models with their own data. The market for AI abstraction layers is substantial, and Baker anticipates increased competition among companies in this space.

He praises Jensen Huang's strategy of being vertically integrated yet horizontally open, particularly in the semiconductor market, where even a 1% market share can be worth $100 billion. Baker asserts that Huang's data centers are highly financeable, with a significant portion of the necessary components locked up, creating challenges for competitors. He concludes that the financing for Huang's projects is standardized and supported by major financial institutions, further enhancing his competitive position.

Baker highlights Jalapeno's unique achievements in chip technology, noting its accomplishments compared to larger competitors. He reflects on Elon Musk's collaborative approach as a wise decision in the industry. Baker emphasizes the complexities of understanding customer preferences in a supply chain-constrained environment, which complicates market analysis.

He outlines a deal structure where a chip company invests in a customer, suggesting that this can be mutually beneficial as long as the investment is less than the gross profit. Baker discusses scenarios where financing deals can yield benefits through revenue sharing and mentions that warrants tied to fixed prices can be advantageous if chip performance exceeds stock performance. He concludes that analyzing the hierarchy of these deals can provide deeper insights into genuine customer preferences.