What Happens When the AI Boom Runs Out of Money
Speakers
Ben Thompson
— Guest
Summary
Ben Thompson discusses the economics, business models, and geopolitical implications of the AI boom. He analyzes how the massive capital requirements of AI are reshaping big tech, comparing Google's AI investments to Berkshire Hathaway's railway acquisition, and warns that hyperscalers will eventually commoditize Nvidia's chips. The conversation also covers why Amazon and Apple have the deepest moats, how AI threatens Microsoft's core software business, and why compute scarcity has saved Intel.
- Google's massive AI capex mirrors Berkshire Hathaway's shift from high-margin candy to high-absolute-profit railways.
- Amazon's core retail business is impervious to AI, and its playbook of internalizing infrastructure costs provides a massive moat.
- Meta is uniquely positioned to monetize AI by using its liquid ad market to verify and optimize AI-generated content.
- Microsoft is playing a defensive 'IBM middleware' strategy as AI threatens to disrupt its core systems of record UI.
- Extreme compute scarcity and geopolitical risks have forced the tech industry to save Intel as an alternative to TSMC.
- Nvidia faces long-term existential threats from hyperscalers who have cheaper capital and are commoditizing AI chips.
- Memory makers risk structurally impairing their own market by keeping prices high, driving algorithmic efficiency.
Ideas
Google's AI transition mirrors Berkshire's railway acquisition.
Google is undergoing a transition similar to Berkshire Hathaway's shift from See's Candies to the BNSF Railway. While Google's core search business has perfect margins, the AI opportunity targets all white-collar work, offering absolute profits that are astronomically larger despite lower margins. This justifies Google's massive capex and equity issuance to capture the space.
High prices will structurally impair memory demand.
Memory makers have created a massive target on their backs by maintaining high prices and constrained supply. This dynamic is incentivizing major customers like Apple to seek Chinese alternatives and driving the industry to develop algorithmic changes specifically designed to use less memory, which will structurally impair the memory market.
Compute shortages incentivize routing around TSMC's monopoly.
TSMC's conservative approach to capacity expansion has offloaded significant risk onto big tech companies, resulting in massive foregone revenue due to compute shortages. This scarcity, combined with geopolitical concentration risks, is forcing the industry to route around TSMC by investing heavily in alternative foundries.
Compute scarcity and geopolitical risks saved Intel.
Intel has been saved by the extreme compute shortage and the geopolitical risks associated with TSMC's monopoly. Big tech companies are now economically incentivized to endure the pain of bringing Intel's foundry business up to speed to secure alternative supply chains, setting Intel up for major future partnerships.
Amazon's core business is impervious to AI.
Amazon possesses the deepest moat in technology because its core retail and logistics businesses are highly insulated from AI disruption. Furthermore, Amazon has a proven playbook of building massive infrastructure (like AWS, logistics, and custom silicon such as Graviton and Trainium) to serve its own internal needs first, which gives it the scale to iterate and eventually sell these services externally as highly competitive commodities.
Apple's physical device moat insulates it.
Apple is well-insulated from AI disruption because its moat is built on deterministic physical goods and unparalleled distribution. Apple does not need to build frontier AI models; it can act as an aggregator, leveraging its massive user base to force AI suppliers to come to them and provide on-device capabilities.
AI will massively improve Meta's ad targeting.
Meta is uniquely positioned to monetize AI by integrating leading-edge models into its advertising engine. Using its massive, liquid ad marketplace to verify AI-generated content and improve ad matching by even a few percentage points will yield billions in incremental returns, making its massive AI investments highly rational.
AI threatens Microsoft's core software UI business.
Microsoft's core software and systems of record UI businesses are highly vulnerable to disruption by AI models that can easily replicate repetitive tasks. To survive, Microsoft is executing a rational but defensive 1990s IBM playbook—acting as the safe, backwards-compatible middleware for enterprises to adopt AI, which caps its upside compared to frontier AI leaders.
Hyperscalers with cheaper capital will commoditize Nvidia.
Nvidia's current margins are unnatural and partially propped up by assuming risk through circular financing with neo-clouds. In the long run, Nvidia faces an existential threat from hyperscalers like Google and Amazon, who possess a lower cost of capital and are actively commoditizing AI chips (like TPUs and Trainium) to sell externally.
