Everyone knows that the rapid rise of AI is powered by processor chips. Not many people know that almost all the chips used for AI are made by one, vulnerable. company.
The most popular chips for use in AI are GPUs from Nvidia. GPUs – Graphics Processing Units – were originally designed to do the maths required to create the detailed, realistic, quick-changing graphics displays we expect from modern games. The calculations used for this are similar to those used for the neural networks that power AI, so GPUs have been widely adopted for AI.
Lots of companies have also designed their own chips specifically for AI, including Google, Amazon and Microsoft. Neither these companies, nor Nvidia, have a way of fabricating chips: they are fabless designers, to use the jargon. The UK’s ARM has always been fabless. They all rely on one company for the chip fabrication: the Taiwan Semiconductor Manufacturing Company (TSMC).
TSMC is a big company. It has a market capitalization of about half a trillion dollars. It is one of the 15 most valuable companies in the world. It’s built it’s dominance by continually improving and investing. In 2021, TSMC announced that it would spend $100 billion over three years to expand its fabrication capabilities.
TSMC is also vulnerable. It’s based less than 200 miles from mainland China. Tensions between China and Taiwan have never been higher. The US has banned the export of all high-end AI chips to any entity operating in China. TSMC must look like a very tempting target to China. Perhaps all the governments and private investors looking to invest in AI might like to think about making the supply chain more resilient.