600,000 NVIDIA H200 chips are sitting in Hong Kong, unable to enter mainland China.
Domestic tech giants want to buy them, but large-scale imports still haven’t been approved.
Even more astonishingly, an 8-GPU H200 server is now priced at over 6 million yuan in China—about three times more than a year ago.
It used to be that the U.S. didn’t want to sell, while Chinese companies wanted to buy.
Now the U.S. allows sales to approved customers, but China, on the contrary, is in no hurry to give the green light.
For AI in the short term, it’s true that we can’t do without NVIDIA; for AI development in the long term, we can’t rely on NVIDIA forever.
China’s homegrown computing power needs a market and time, but the competition in large language models won’t wait for us.
So the real challenge isn’t whether to buy H200s, but how to strike a balance between domestic alternatives and global AI competition.
Chips can wait to clear customs, but the AI race won’t wait for anyone.
Domestic tech giants want to buy them, but large-scale imports still haven’t been approved.
Even more astonishingly, an 8-GPU H200 server is now priced at over 6 million yuan in China—about three times more than a year ago.
It used to be that the U.S. didn’t want to sell, while Chinese companies wanted to buy.
Now the U.S. allows sales to approved customers, but China, on the contrary, is in no hurry to give the green light.
For AI in the short term, it’s true that we can’t do without NVIDIA; for AI development in the long term, we can’t rely on NVIDIA forever.
China’s homegrown computing power needs a market and time, but the competition in large language models won’t wait for us.
So the real challenge isn’t whether to buy H200s, but how to strike a balance between domestic alternatives and global AI competition.
Chips can wait to clear customs, but the AI race won’t wait for anyone.