What has Gavin Baker said lately

Gavin Baker characterizes the current AI trend as being in its early stages, viewing it as year three of a **20 to 30-year cycle** comparable to the early internet era initiated by Netscape Navigator . He projects that total inference revenue will significantly surpass $200 billion by the end of the current year . While acknowledging market froth, he believes the current "rolling bubble" is concentrated in publicly-traded nuclear and quantum computing companies rather than mainstream AI . Baker also notes a unique challenge in evaluating the technology's progress: the true intelligence of each new generation of AI models may never be fully understood because new, more capable models are released before their predecessors can be thoroughly evaluated . This rapid pace of development underscores the long-term nature of the investment cycle he envisions.

Looking at the infrastructure layer, Baker posits that the long-term energy shortage for AI will ultimately be solved by orbital compute, which involves deploying individual server racks in space . He estimates the total cost to deploy a gigawatt of AI compute in space, including silicon, would be **approximately $30 billion** . In the nearer term, he believes the investment strategy of "finding the next bottleneck" in the semiconductor supply chain is now over . He has been disappointed by the custom ASIC development efforts at Meta and Microsoft and predicts that the first AI models trained on NVIDIA's Blackwell architecture will be released in early 2026, with xAI likely being the first to do so . This suggests a continued reliance on established chip designers for frontier model development in the immediate future.

Regarding specific companies and competitive dynamics, Baker asserts that NVIDIA has the capability to develop its own model that is "pretty close to the frontier" whenever it chooses and is highly likely to become the world's dominant provider of open-source AI [9, 16]. This contrasts with his assessment of Meta, where he claims Mark Zuckerberg's January 2024 prediction of having the best-performing AI by 2025 was completely wrong . A key competitive differentiator appears to be proprietary data; Baker claims that both Cursor and Anthropic individually possess **more tokens of proprietary coding data** than exist on the public internet, creating a significant moat . The importance of non-public information extends to talent acquisition, with Baker citing Anthropic researchers who state that hiring is influenced by candidates' public discourse on platforms like X and Substack [1, 12].

This market structure, with a mix of public and private companies competing at every level of the technology stack, informs Baker's investment philosophy . He argues that a crossover investing approach combining public and private markets is paramount in the AI sector . He is skeptical of the venture capital model, stating that the operational value-add from many VC firms is "wildly overstated" and that there is more bad behavior in private markets than is commonly realized, citing FTX as an example [5, 6]. Ultimately, he believes the performance of any investment organization is driven by a core group of **two to ten key individuals**, and that the peak performance years for professional investors are between the ages of 50 and 70 [3, 4].