As Chinese firms scale and U.S. export rules tighten, Nvidia is fighting to keep a foothold in China
Nvidia CEO Jensen Huang says China is just “nanoseconds behind” the U.S. in chipmaking and that Washington should stop trying to wall off the market.
Speaking on the BG2 podcast, Huang argued that allowing companies like Nvidia to sell into China would serve American interests by spreading U.S. technology and extending its geopolitical influence. “We’re up against a formidable, innovative, hungry, fast-moving, underregulated [competitor],” Huang said, talking about the pedigree of China’s engineers and controversial 9-9-6 working culture.
His comments come as Nvidia hopes to ship its H20 AI GPU to Chinese customers again, following a months-long pause tied to new U.S. export rules. The Commerce Department is understood to have begun issuing licenses for the H20 in August, and Nvidia is already working on a successor chip designed to comply with current restrictions while offering better performance. The company has not confirmed specs, but it would be Nvidia’s second attempt to tailor an AI accelerator specifically for the Chinese market since the original A100 and H100 bans took effect.
You are right it’s a critical industry to China, not only security, but also commercial because USA is blocking China from commercial use too.
China claim to be able to make 7nm now, if that’s true they are 7 years behind.
They also claim they will be able to make 5nm next year, if that’s true they are catching up.
China is forced to develop their own technologies, and if they can make EUV within a few years, there is practically nothing stopping them from catching up and maybe even surpass ASML/TSMC.
That is interesting. It is important to note that these claims are kind of arbitrary. 7nm doesn’t really mean anything because they can measure any feature and say that. What would be more important is seeing the IPC per core. Moving beyond 7nm is also extremely difficult but even getting to 7nm is extremely impressive. This will probably actually just end up helping us as the consumer tbh because chips are so expensive right now, since basically a handful of companies control the market.
That is true, but even Intel has adjusted to use numbers that reflect similar processes to TSMC, and Samsung does too.
I have no doubt SMIC (China) could be exaggerating a bit, Intel does that too. But I’m sure this is supposed to be comparable to Intel/Samsung/TSMC.
Yes, that’s why the claim of 5nm is extremely interesting, that’s almost impossible without EUV.
But we don’t know enough yet, for instance error rate/yields are very important too. Without acceptable error rates, they can only make small dies, which complicate complex designs for demanding tasks, as they have to be spread among more dies.
Something AMD mastered with Ryzen, maybe in part because they were limited to Global Foundries (inferior production process to Intel at the time) when they designed the Ryzen chip, and they managed to turn a disadvantage in production into an advantage in design, using chiplets that worked extremely well together.
Also power consumption is important, because without lower power consumption reducing the scale of production is near meaningless.
That’s the design side of the problem, and IDK where China is at regarding efficient AI chip design, but they did manage to make the worlds fastest super computer based on very good design a few years back. As I see it, design will be the lesser problem for China on chip production, because many companies have been working on their own design for years now. Even Xiaomi have their own AI design in their new chips, and those chips are very good from what I hear.
I agree, if the alternative is near monopoly by Nvidia and TSMC, I have no doubt that not only China but also the rest of the world will benefit too.
Nice info, thanks. I’m going to look up IPC comparisons if I can.