The AI gold rush is less about genius algorithms and more about silicon, specifically who makes it. For the last two decades, semiconductor spending held steady. Then, in just four years, it doubled. Author Chris Miller, whose book 'Chip War' lays bare this new reality, pins the blame—or credit—squarely on AI.

But here’s the kicker: the biggest bottleneck isn't a lack of capital or demand. It’s a single company: TSMC. If you’re building an AI startup or investing in this space, ignoring this reality is like launching a car company without understanding the global battery supply. This isn't just about geopolitics; it's about the very foundation of your future business.

Key Takeaways

  • AI drove a doubling of semiconductor spending in the last four years, a radical shift after two decades of flat growth.
  • TSMC's manufacturing capacity, coupled with its deeply conservative corporate culture, is the primary choke point for advanced AI chips.
  • China's seemingly modest $295 billion AI buildout is a puzzle, suggesting either a longer AGI timeline, a focus on domestic Huawei chips, or a lack of internal urgency.
  • US export controls are shaping the global chip market, but China's internal strategy to circumvent them, or its actual long-term AI goals, remain opaque.
  • Western AI firms have sabotaged public trust by frequently marketing AI as a threat to jobs and humanity, creating a drag on adoption and policy support.

The Silent Choke Point: TSMC's Bottleneck

Miller cuts through the noise: “The big change is that for the last two decades if you look at semiconductor spend as a share of GDP it was roughly flat. And then in the last four years it's roughly doubled and that has been all been driven by AI.” This isn't abstract growth; it's a compute arms race. And right now, one company holds the reins.

“I think right now it is it is gated by TSMC and manufacturing capacity more broadly,” Miller states. TSMC, a Taiwanese giant, isn't just a supplier; it’s the supplier for cutting-edge chips. Their cautious corporate culture, combined with deep supply chain choke points, means they won’t simply ramp up production overnight to meet insatiable AI demand. This translates directly to higher costs and longer lead times for anyone needing serious compute power, from hyperscalers to your own AI-powered SaaS.

China's Puzzling AI Under-Spend

Amidst the frenzy, China presents a paradox. Despite announcing a massive $295 billion AI buildout plan, Miller sees it as an underspend, especially given Beijing's supposed AGI ambitions. “The puzzle is is why isn't Xi Jinping more AGI pill? China's been underspending for the last four years on AI,” he asks.

Miller offers a few theories for this surprising restraint. It might reflect a longer-term AGI timeline, where immediate, aggressive chip acquisition isn't seen as essential. Or, it could signal a strategic pivot towards shoring up the domestic Huawei ecosystem, even if it means sacrificing access to the most advanced foreign chips. “If you were AGI, you would buy the H200's. But if you're not and you're focused on domestic manufacturing and supporting the Huawei ecosystem... then maybe say ban the foreign chips,” Miller suggests. For founders, this means global supply lines are not just economic; they are profoundly political, shaped by national tech priorities that can shift without warning.

The Self-Inflicted Wounds of Western AI

Finally, Miller touches on a critical, often overlooked aspect: the terrible marketing of US AI firms. “US AI firms have been uniquely bad with publicity promising to destroy jobs promising to threaten the future business of humanity. That's been the marketing uh of AI over the last couple of years,” he points out. This messaging doesn't just alienate the public; it creates an environment of fear and skepticism that can hamstring adoption, regulation, and even access to talent. When you constantly frame your product as an existential threat, don't be surprised when people don't line up to buy it.

What to Do With This

If you're building an AI product or service, stop treating compute as a commodity. This week, pull your projected chip requirements for the next 18 months. Then, assume those costs will rise by 30% and lead times will double. Diversify your cloud providers, or at least understand their underlying hardware dependencies. If your business model crumbles under these assumptions, you need to adjust your strategy now, before the TSMC bottleneck squeezes you dry. Consider hardware-agnostic solutions or even exploring non-Western supplier relationships if your target market allows. Your future depends on more than just code; it depends on silicon you can actually get.