Liang Wenfeng: China-made AI chips have a historic opportunity to replace incumbents; within the next year, the claim that the “ecosystem for domestically made chips has no issues” will be proven.

Golden Finance reported that on July 23, Liang Wenfeng, founder of DeepSeek, said at a recent investor communication meeting, “We currently have about 20,000 H-equivalent computing power, most of which has just arrived—arrived in the past one or two months. Our gap with the US is mainly in resources. There’s almost no gap in terms of people, because it’s the same group of people.
Our gap with the US may be 12 months behind; the US may be 12 to 18 months ahead, or 6 to 12 months ahead. Nvidia CUDA’s moat is being quickly dismantled. On the one hand, there is now AI—AI can write code, and I can use AI to build this ecosystem. On the other hand, there are some new technologies. For example, we developed a technology called TileLang (an open-source high-performance AI operator programming language developed by a team from Peking University’s School of Computer Science). With this higher-level language to write CUDA operators, we can quickly write out Nvidia’s entire ecosystem. Also, previously the market for computing cards was smaller than that for gaming cards; now the market for computing cards is larger than that for gaming cards, so there’s no reason for these two to still need to be coupled together.
The replacement of domestic AI chips now presents a historic opportunity. We believe that within the next year, we will be able to see something being validated: that the ecosystem for domestic chips has absolutely no issues. People previously thought there were issues—that the chips couldn’t be used or weren’t good to use—but I think within a year we can reverse that perception.
The purpose of buying Huawei 950 is still to help Huawei build a strong ecosystem. Huawei 950 supernode can completely match Nvidia’s GB200 and GB300 in performance and price. The price will definitely be higher, but the increase is limited. The only cost is that four Huawei cards can only top one Nvidia card, while still being two years behind. Being two years behind is a time gap: Huawei 950 supernode is shipping this year in Q3 or Q4, while Nvidia GB200 is from two years ago in Q3.
You can basically depreciate Nvidia’s cards over five years. Huawei’s cards are probably depreciated over at most three years. Huawei 950 can still be used this year—that’s quite good. Next year, I think it can still be used. After that, maybe it would really use too much power.”
NVDA2.36%
View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments
  • Pinned