What has Gavin Baker said recently about specific companies?

Gavin Baker's analysis of the AI landscape highlights a significant concentration of value and performance at the frontier model layer, dominated by a few key players. He identifies Anthropic, OpenAI, and XAI's Grok 4.3 as defining the current "Pareto frontier" of AI capabilities . He argues that the overwhelming economic returns in AI are accruing to this frontier . This is driven by a structural shift in pricing from "all-you-can-eat" subscriptions to "pay-by-the-drink" usage-based models . Baker predicts this shift will enable OpenAI and Anthropic to achieve well over **$200 billion in ARR** this year [8, 25]. In this competitive landscape, he characterizes Google's strategy as that of a low-cost producer, which rationally "sucks the economic oxygen" out of the broader AI ecosystem . Meanwhile, he notes that XAI's cloud computing deal with Google is structured to generate more operating profit per gigawatt than deals for Anthropic, Meta, Google, or OpenAI .

At the hyperscaler level, Baker offers a nuanced view of corporate strategy and execution. He praises Microsoft CEO Satya Nadella's decision to prioritize using internal compute to improve Microsoft's own products, viewing it as the correct long-term strategy despite initial negative market reaction . However, this praise for software and product strategy does not extend to hardware development. Baker has been disappointed by the custom ASIC development efforts at both Microsoft and Meta, suggesting a failure to execute on in-house chip design . This contrasts with his view of the application layer more broadly, where he argues that AI has so far been a net negative, having **destroyed trillions of dollars of value** . He sees profits accruing to energy, data centers, chips, and models, but explicitly not to applications .

In the semiconductor space, Baker emphasizes the pivotal roles of TSMC and NVIDIA. He credits TSMC with having single-handedly prevented an AI bubble by strategically constraining its wafer supply . This supply constraint has significant implications for NVIDIA's potential revenue; Baker speculates that if TSMC were to meet NVIDIA's full demand, the chipmaker could sell between **$2 trillion and $3 trillion** worth of GPUs in 2026 or 2027 . Beyond hardware, Baker believes NVIDIA has the capability to become a major player at the model layer, stating that CEO Jensen Huang can develop a model that is "pretty close to the frontier" whenever he chooses [11, 17]. He also predicts NVIDIA is highly likely to become the world's dominant provider of open-source AI . Despite this bullish long-term outlook, he has observed that NVIDIA and Broadcom have been laggards in the most recent semiconductor market rally .

Baker often draws on historical examples to frame current market dynamics. He recounts how many venture capitalists initially refused to fund Roku, fearing competition from Amazon, Google, and Apple . He notes that Fidelity, by contrast, led three consecutive funding rounds after correctly assessing that competing with Roku was not a top priority for these tech giants . This case study illustrates his approach to evaluating competitive threats from large incumbents. His commentary suggests that while the current AI cycle is still in its early stages—comparing ChatGPT's emergence to Netscape Navigator in year three of a potential 30-year trend —the value is currently captured by a select few companies at the foundational layers of the technology stack.