AI Chip Shortage and Post-COVID Economic Recovery

Note: I wrote this in May 2021, in the middle of the pandemic chip shortage. I have added an update at the end on how it actually played out.

As economies started to reopen after COVID vaccinations, the technology world hit a serious shortage of computer chips, driven mainly by disrupted semiconductor production. Almost every modern technology, from consumer electronics to AI, 5G, and cloud computing, depends on chips, which is why some called it an "AI chip shortage."

AI researchers and enterprises lean heavily on GPUs (largely NVIDIA's) for training and inference, and demand was climbing further as companies pushed toward custom AI silicon. Intel had recently acquired the AI-chip startup Habana Labs, and others like Graphcore were attracting heavy investment, part of a broader move to design chips specifically for machine learning.

The supply side is concentrated and fragile: roughly 75% of semiconductor manufacturing sits in East Asia, led by TSMC in Taiwan and Samsung in South Korea. COVID, extreme weather, and a factory fire all disrupted production, leaving a reported ~25% shortfall. The US and Europe flagged it as a top economic priority. At the time, the optimistic expectation was that the shortage might ease within about six months.

Update (looking back)

Six months was wishful thinking. The shortage stretched well into 2022 and only really eased through 2023 as demand cooled and new capacity came online. It also pushed semiconductors up the political agenda for good: the US passed the CHIPS Act in 2022 and the EU followed with its own Chips Act, both aimed at bringing more manufacturing onshore. The underlying point held up, though, and matters even more in the current AI build-out: AI's progress is bound to the physical supply of chips, and that supply is geographically concentrated and easily disrupted.