Gavin Baker - AI Market Jitters - [Invest Like the Best, EP.485]
Summary
This episode focuses on the mismatch between public-market sentiment and what is happening on the ground in AI infrastructure. Gavin Baker argues that, despite sharp selloffs in AI-related stocks, demand for GPUs, tokens, and AI compute continues to accelerate. The discussion also digs into the economics of GPU contracting versus spot pricing, the importance of long-term supply agreements, and how financing and credit markets are becoming central risks in AI buildout. Another major theme is chip and memory architecture, including SRAM, DRAM, and the benefits of disaggregating compute to improve ROI. The conversation closes with a look at SpaceX, orbital compute, and the biggest long-term risk Gavin sees: regulation.
Key Takeaways
- 1Public AI markets may be signaling caution, but operational demand for AI compute is still accelerating rapidly.
- 2The spread between contracted GPU prices and spot GPU prices is one of the most important indicators in the current AI cycle.
- 3Open source AI can compress model-level margins while still expanding the overall compute market.
- 4Credit markets and financing are becoming a major constraint and risk in AI infrastructure expansion.
- 5Long-term supply agreements, especially around memory allocation, may determine which AI companies gain competitive advantage.
- 6AI chip design will keep evolving toward more flexible memory and compute tradeoffs, and disaggregated compute can improve ROI.
Notable Quotes
""The best AI and software companies from OpenAI to cursor to perplexity use Work OS to become enterprise ready overnight, not in months.""
""However you cut it, whether you cut GPU availability, whether you cut GPU rental price, say, whether you cut like the spot price of DRAM this month, token growth, everything is actually accelerated.""
""A token is a token, and you need the exact same amount of compute to make a token all else equal.""
""Nvidia has actually, as we record this, at its lowest forward PE of the last 10 years.""
""You just can't beat SRAM in particular for that feed forward network. It no matter how much you try to get the ratio of compute to be of D-RAM, to SRAM, on the chip correct, the workloads are always changing.""
""This is going to be really, really positive for the ROI on AI.""
""It was like the market just utterly absorbed it.""
""I would also just say like I did spend a lot of time at Starbase and orbital compute feels more real every day.""
Episode questions
Why does Gavin think the public market is misreading AI right now?
He says stock prices are falling while the operational evidence on the ground is still accelerating. His examples include GPU scarcity, higher rental prices, and rising token growth, which he views as stronger signals than market sentiment.
Why is the spread between contract and spot GPU pricing so important?
Because many GPUs were financed under long-term contracts that are now far below current market rates. As those contracts reset, the economics of installed compute can reprice materially higher, improving economics for owners and vendors.
How can open source models help AI infrastructure if they lower model margins?
Gavin argues that cheaper models increase total usage and shift demand toward more compute consumption. Even if frontier model margins shrink, the compute layer can capture more volume because tokens still require the same underlying processing.
What is the biggest risk Gavin sees for AI?
Regulation. He says political backlash around data centers, energy use, and job displacement could create a meaningful obstacle even if fundamentals remain strong.