Ben Thompson on Big Tech, China, and the AI Boom Running Out of Money - [Invest Like the Best, EP.487]

If the US achieves overwhelming AI superiority, particularly in military applications, it could force adversaries like China into desperate, high-stakes actions like destroying TSMC, ultimately destabilizing the world.

The economic realities and deep dependencies on China make complete decoupling impractical. Companies will not accept the massive cost disadvantages of reshoring unless forced by conflict, leaving the US perpetually vulnerable.

Like the 19th-century railroads, AI has a huge mismatch between massive upfront capital expenditure and future revenues. The industry is rapidly moving down the capital stack, and a funding gap could cause a major blowup long before technical limits are hit.

Analogous to Berkshire Hathaway buying BNSF with See's Candies profits, Google is using its high-margin search business to fund a capital-intensive, lower-margin, but astronomically larger absolute-profit opportunity in AI.

Today's AIs are trained on the final product of human cognition (e.g., articles, code). The next great leap could come from training models on the actual neural "traces of thought," potentially via technologies like Neuralink, to solve for creativity and unverifiable domains.

By shifting from fixed per-seat licenses to variable usage-based models for AI, Microsoft forces customers to scrutinize their monthly bills. This breaks the inertia of its bundled offering and invites evaluation of best-of-breed competitors.

By chasing consumer subscriptions instead of leveraging advertising from day one, OpenAI missed a critical window. An ad-supported ChatGPT could have created a powerful monetization flywheel, severely threatening incumbents' core business models when they were most vulnerable.

Intel's foundry ambitions languished because no major customer would take the risk. However, TSMC's under-building created such a severe compute scarcity that companies are now economically forced to work with Intel, effectively subsidizing its learning curve to secure future supply.

Amazon de-risks new technology development, like custom AI chips, by first deploying them internally for its massive retail and cloud operations. This provides the scale and feedback loops to improve the products until they're ready to be sold externally, a strategy competitors can't easily replicate.

Apple excels at creating flawless, deterministic hardware—a process requiring zero tolerance for error. This cultural DNA clashes with AI development, which is inherently probabilistic and requires rapid, messy iteration, explaining the company's caution in the AI race.

Like IBM in the 90s, Microsoft is positioning itself as the dependable, integrated provider for enterprises wary of new tech. It offers a middleware layer that connects legacy systems to new AI models, selling stability and integration over cutting-edge performance.

Unlike competitors, Meta has a built-in verification machine for its generative AI. It can generate millions of ad variations and measure their real-world effectiveness (clicks and purchases) instantly, creating a powerful feedback loop to improve its models based on direct economic outcomes.

To protect its high gross margins, NVIDIA avoids direct price cuts. Instead, it assumes financial risk for its customers (e.g., guaranteeing compute purchases), which lowers their cost of capital to buy more GPUs. This risk is a hidden price concession not reflected on the income statement.

Historical tech bubbles often leave behind valuable, overbuilt infrastructure. The intense demand for power to run AI data centers is spurring massive investments in energy generation. Even if the AI boom cools, society will be left with a surplus of power, a fundamental economic good.