Recommended: Listen to this story (10:10 - 16:33):

The Story: The AI revolution relies on computing power and a lot of it. Large language models get that computing power from microchips powered by semiconductors. The tech is complicated and very expensive to make, and today, Taiwan produces over 60% of the world’s semiconductors and over 90% of the advanced semiconductors.

Naturally, this concentration of manufacturing has created supply chain bottlenecks that are slowing the distribution of chips to top bidders around the world. Those top bidders include Meta, OpenAI, and Microsoft. Last week, Mark Zuckerberg announced that Meta purchased 340,000 H100 GPUs from Nvidia, the world’s leading AI computing company that, you guessed it, relies on Taiwan’s TSMC to build 90% of its chips.

There’s growing fear that dependence on Taiwan to buttress the AI revolution is becoming increasingly risky. The threat of an invasion from China is not out of the question, and would very likely halt the production and delivery of Taiwan’s chips.

The US wants to reduce this dependence. America’s solution is creating factories that design semiconductor chips on US soil. And the plans have already begun. Taiwan’s biggest chip provider TSMC is building two factories in Arizona, a project that will cost over $40 billion. The original plan was for the factories to be operational by 2026, but TSMC chairman Mark Liu has announced the project is delayed until at least 2027 or 2028 due to labor and licensing issues.

While many American building projects are experiencing similar delays to the Arizona factories, over $200 billion dollars in total has been committed by many investors to support new chip manufacturing infrastructure across the country.

In November, GlobalWafer, another Taiwan-based chipmaker, began building a $5 billion dollar plant in Sherman, Texas. The Biden Administration has also announced funding for 25 National AI Research Institutes.

As the US government gets more involved in supporting American companies to build the AI infrastructure necessary to meet the growing demand for chip tech in the US and around the world, there is also a question that has arisen. Will this lead to a new problem for the US, what has been coined the “AI Industrial Complex?”

Expert Take: Doug Clinton, co-founder and managing partner at Deepwater Asset Management, believes it is “only a matter of time” before the US will be able to rival Taiwan’s semiconductor manufacturing dominance.

Clinton expects that in the next 2-3 years, we will begin to see semiconductor fabrication plants (fabs) pop up around the US. However, he says it will likely take a decade for American infrastructure to catch up to Taiwan, “unless we have some massive capital infusion, probably hundreds of billions of dollars.”

Clinton doesn’t see the growing demand for AI chips and fabs in the US as the dawn of an impending “AI industrial complex.”

He says, “You have hyperscalers. You have Meta, spending close to $10 billion dollars this year on Nvidia chips. I think they are seeing an opportunity in the market, I don’t think that their intent is to have undue control over government policy, I think their intent is that they see a huge opportunity in AI.”

He continues, “I think it’s very much market-driven… could there be a by-product where maybe there is some influence on the government? Sure, I think you could’ve said that looking back on the internet twenty-five years ago… I think we need to figure out the supply-chain more, get more diversity, and the chips away from Taiwan. That to me is more important than domestic influence on government policy.”

*Stay informed about the three biggest stories in venture capital and tech news every weekday morning. 3-min reads only.