![Gavin Baker - AI Market Jitters - [Invest Like the Best, EP.485]](https://image.scripod.com/https://megaphone.imgix.net/podcasts/c8fca7de-8f77-11f1-ae96-ab857574605f/image/e34471e6c9b4af0375abbd710ff0fa11.jpg?ixlib=rails-4.3.1&max-w=3000&max-h=3000&fit=crop&auto=format,compress)
Gavin Baker - AI Market Jitters - [Invest Like the Best, EP.485]
Gavin Baker - AI Market Jitters - [Invest Like the Best, EP.485]
Gavin Baker - AI Market Jitters - [Invest Like the Best, EP.485]
In this episode, Gavin Baker, founding partner and CIO of Atreides Management, returns for a seventh conversation to dissect the recent turbulence in AI stocks. Despite a market sell-off that echoed 2022, Baker argues that the underlying fundamentals—from GPU pricing to token growth—are accelerating, not deteriorating. He offers a contrarian perspective on the gap between public market sentiment and private sector reality, touching on open-source AI's impact, the strategic dynamics of memory supply agreements, and the surprising role of AI models like Claude in shaping market narratives.
Baker contends that the recent 40-60% drop in AI stocks is disconnected from on-the-ground acceleration in GPU availability, rental pricing, and token growth. He highlights that private AI companies like OpenAI and Anthropic are omitted from market comparisons, and notes that old GPU prices are rising, which will boost revenue as contracts reprice. The sell-off was triggered by open-source model releases (Kimi, GLM 5.2) and China's DUV progress, but Baker sees these as demand drivers, not threats. He identifies rising real yields and credit spreads as the real concern, yet argues that hyperscalers' accelerating operating cash flow could fund the buildout without debt. Baker also discusses the game theory of memory LTAs, Nvidia's innovative credit wrappers, and the potential of SRAM-based accelerators. He concludes that the biggest risk is regulation, not technology, and highlights SpaceX's orbital compute ambitions as a dark horse opportunity.
02:00
02:00
The overall balance of fundamentals is improving.
02:43
02:43
All quantitative metrics are accelerating despite stock price drops.
04:40
04:40
Old GPU prices are rising sharply, contrary to expectations.
05:06
05:06
Contracted compute repricing will boost revenue and answer ROI questions.
06:57
06:57
Market misreads open-source AI shift as negative.
08:33
08:33
Open-source AI models drive more compute demand.
10:53
10:53
Debt-fueled buildouts demand immediate repayment and can unwind quickly.
14:56
14:56
The current sell-off is driven by clear, identifiable events, which is oddly comforting.
16:00
16:00
Hyperscalers are under-earning, planning to raise prices.
17:38
17:38
AI inference demand is accelerating with open-source models.
20:00
20:00
Credit issues won't matter during a compute shortage.
21:37
21:37
AI models are now the primary interpreters of market news.
23:56
23:56
Solving continual learning could cause a demand discontinuity.
25:22
25:22
Reduced training demand would be positive for the world.
29:47
29:47
Only about 500,000 people use agentic AI today.
30:52
30:52
Token spend reaches 20-30% of compensation in AI-native companies.
33:40
33:40
Breaking an LTA could jeopardize future supply allocations.
36:50
36:50
Memory supply dynamics have shifted, making LTA breaches costly.
37:45
37:45
Nvidia's credit wrapper bridges cash flow gaps for GPU buyers.
41:51
41:51
Most people are more bullish than me.
44:30
44:30
China's DUV progress is a propeller plane, not a jet turbine.
47:00
47:00
Open-source AI enables better performance at lower costs.
48:23
48:23
Frontier models plan and delegate to smaller models.
49:33
49:33
Economic returns shift from frontier labs to inference clouds.
50:44
50:44
Open-source tokens may dominate volume, but frontier models capture most value.
53:48
53:48
The AI industry must better communicate its benefits to the public.
57:21
57:21
SRAM accelerators may not justify ROI despite their superiority.
1:01:14
1:01:14
Orbital compute is feasible, and the market underestimates its scale.
