How concerned are you about the possibility of China developing competitive AI chips due to these constraints?
China has some competitive advantages.
Yes, it’s not at the scale of data centers, but the Huawei Mate 60 smartphone released last year received attention for its 7-nanometer chip.
They are limited by their semiconductor processing technology, but they can still build large systems by combining many of these chips.
Overall, how worried are you that China will be able to keep up with the US in generative AI?
Regulations will restrict China’s access to cutting-edge technology, giving non-restricted countries better technology that is advancing rapidly. This limitation will increase costs for China as they try to aggregate more chipmaking systems to compensate.
Does your need to build compliant chips for the Chinese market impact your relationship with TSMC, Taiwan’s semiconductor leader?
No, regulations are specific and are like speed limits.
You’ve mentioned that only eight out of 35,000 components in your supercomputer are from TSMC. Is this downplaying your reliance on TSMC?
No, not at all.
What point are you trying to make with that statement?
I’m highlighting that building an AI supercomputer involves numerous components from various semiconductor industry partners, not just TSMC. We work closely with companies like Samsung, SK Hynix, Intel, AMD, Broadcom, Marvell, and others. Success in our AI supercomputers benefits many companies.
How often do you communicate with Morris Chang or Mark Liu at TSMC?
Regularly, continuously.
What are your discussions focused on?
We discuss advanced packaging, capacity planning, and equipment for future advanced computing needs. Support from TSMC is crucial for these developments.
Who do you see as your biggest competitor in the AI chip market?
Everyone. Teams like the TPU team, AWS Trainium team, AWS Inferentia team, Microsoft’s Maia project, as well as numerous startups and existing semiconductor companies are all competitors.
