The United States exports weak AI chips to China. Is it a plan to return to the mountain or a clever strategy? Gavin Baker, a well-known technology investor, posted onblockThese Nvidia chips.

Is it a mistake to export GPUs to China? Beijing itself blocks Huida chips first

Baker pointed out that starting in the summer of 2025, it is not only Washington that is blocking Huida GPUs from entering China, but Beijing itself is also imposing controls. He said that after U.S. Commerce Secretary Lutnick proposed making Chinese developers "addicted" to U.S. technology stacks, within weeks Beijing began to pressure technology giants to stop purchasing H20; then it ordered a halt to the purchase of RTX Pro 6000D, sent a customs team to the port to inspect Huida chip imports, and banned the use of foreign AI chips in new state-funded data centers.

Citing a Reuters report, he pointed out that about 600,000 H200 chips are currently stranded in Hong Kong because Beijing does not approve large-scale imports of this product, which is still superior to mass-produced chips produced by China's own industry. Regarding the public’s doubts, Baker asked: “If releasing weaker GPUs is a policy mistake that allows China to take advantage, then why is it China that blocks them?”

He also admitted that China's AI laboratories still have a very high demand for Huida chips.black market H200 server prices have roughly tripled in one year. But in his view, that’s exactly the point: Beijing’s willingness to bear the real costs of its own AI labs while also avoiding reliance on U.S. GPUs shows that Beijing sees imported GPUs as a threat, not a gift.

（Huida requires data centers to reduce Chinese customers, and the price of B300 skyrockets by 5 times）

The cloud becomes a political tool: Let China become addicted and shut down if necessary

Baker's core logic is "leverage". He believes that whether China uses American AI hardware directly within the country or remotely accesses it through Southeast Asian clouds (including the availability zones of large American cloud operators in Southeast Asia), as long as it relies on American chips, the United States will have huge leverage in any future conflict, and "the ability to immediately shut down the cloud" is the most powerful political means.

On the contrary, if China establishes an alternative AI hardware ecosystem that the United States cannot control or influence, it may be shared with Russia, Iran, North Korea and other adversaries over time. Baker emphasized that his remarks on China on

In his reply to netizens, he further expected that there is a "high probability" that Beijing will ban access to U.S. GPUs on Southeast Asian clouds; even if not, the B200 scale that Southeast Asian clouds can obtain will not be enough for China to compete with U.S. laboratories in remote computing power, because U.S. laboratories are gradually turning to Vera Rubin and B300.

H20 lessons and B30A hypothesis: Domestic chips are catching up

Baker believes that the lesson of H20 is: deliberately castrated chips cannot retain China. In his view, the reason why Beijing took action against H20 and RTX Pro 6000D was because it judged that China’s domestic chips had caught up with these products. He therefore argued that it would have been smarter in retrospect if the B30A had been made available in 2025, rather than the H200 which was later approved on a case-by-case basis. He also pointed out that the H200 is not cutting-edge, and while Vera Rubin is in full mass production, it is two generations behind products procured by U.S. laboratories.

The latest sign he cited is that Chinese laboratories are increasingly adapting the MoE architecture to adapt to Chinese domestic hardware: for example, DeepSeek allowed Huawei and Cambricon to obtain V4 early instead of Huida and AMD; subsequent teams led by Huawei have also conducted post-training on V4-Pro on Ascend chips.

Baker believes that this makes Huida CEO Huang Renxun’s long-term lobbying position “look smarter every month” and warns: If the United States lags behind China in AI in the future, those U.S. China hawks who have spawned China’s alternative AI hardware ecosystem will bear considerable responsibility.

The subtle consensus between China and the United States in the chip war: the grand chess game of acquiescence and renting

Jukan, a Citrini researcher, believes that China’s efforts to build its own GPU industry are certainly the main reason, but “injured pride” also plays a role, because Lutnick said that the United States sold “fourth-rate GPUs” to China, which must have been quite harsh for Beijing.

Another user, Ros Thain, said that on the surface China has banned these chips, but in fact it deliberately acquiesces to Chinese companies "informally" leasing Blackwell and other chips to catch up. He also said that both China and the United States can accept the status quo of about 10% to 15% of Huida chips being used illegally. This is "unspoken diplomacy."

Baker replied that this was a "big chess game" and emphasized again that this approach is almost like a smart and deliberately pro-US defense policy implemented by Huida, the Biden administration, and the Trump administration.

this article Is Beijing itself blocking Huida? Gavin Baker: Exporting low-end GPUs to China is smart defense policy first appeared in Chain News ABMedia。