Are you suggesting that the sane policy is for Xi to dump in as much subsidies as needed until China catches up in semiconductors with the US and its allies? I haven't seen anyone else argue this
Yes. And perhaps no one else does because they aren't scaling proponents. But from a scaling perspective, accepting a permanent straitjacket around GPUs & a tightened noose is tantamount to admitting defeat & abandoning the future to other countries; it'd be like expelling all your Jewish scientists in 1935 & banning the mining of uranium. It's not the beginning of any story that ends in victory & being #1, only of stories that end with you being #10, or #100. Like the Manhattan Project, you pay whatever it costs, and it costs what it costs.
so how can they hope to win an outright tech race (and starting from about a decade behind)?
Well, it's certainly not easy. It's not a great situation for China to be in, and yet, it is in fact the one they are in, and they have to deal with it. Reality has no obligation to make anything easy for you, nor should you expect something like 'dethrone the global hyperpower and create a new world order around the Middle Kingdom' to be easy. Similarly, it's not easy to defeat a Russian or Chinese invasion, nor was it easy to develop atomic bombs, etc, but if you don't, there will be consequences you may find unacceptable - even, existential, one might say - so you don't get much of a choice. You miss 100% of the shots you don't take.
China has a lot of resources yet untapped, both financial and otherwise, and is doing many other things like stimulating other sectors of the economy, so clearly it can try more than it is right now - but it appears not to be. If it's not a capability issue, then it must be a choice. Xi seems to disagree that the consequences of choosing to accept defeat in chips & AI will be all that unacceptable, and my comment here is about thinking through what the strategic logic could be from his perspective which makes that choice an acceptable one because the the chip race game is not worth the candle.
What are the restrictions?
Broadly speaking, the policy restricts the sale of GPUs and related technology to Chinese companies. The specifics remain unclear because (a) the Department of Commerce hasn't released its official press report yet and (b) licenses to sell compute to Chinese companies will be approved on a case-by-case basis.
Here's the NYTimes description of the restrictions:
Licenses to continue sales to China will be approved on a case-by-case basis by the Department of Commerce. Standards for these licenses will therefore be an ongoing battleground where this policy can become more or less strict. NYTimes:
Policy Environment
The US has taken complementary policy actions recently, including banning NVIDIA from selling A100s and H100s to Chinese companies and blacklisting 13 more Chinese companies from receiving any investment by Americans.
Taiwan seems to be on-board with the plan, promising "very firm" export controls on Taiwanese chips being sold to the Chinese military complex. Taiwan is a crucial player in this battle, home to Taiwan Semiconductor Manufacturing which produces 53% of the world's outsourced chips and all of NVIDIA's GPUs.
The most important background is the CHIPS Act passed this August providing $52B for US chip manufacturing. Here's a Wikipedia summary of the spending:
How will China respond?
Unclear. NYTimes is maximally vague:
While the US is dependent on China for many imports, we do not seem to critically depend on their GPUs. (I don't have strong proof of this claim and would welcome disagreement. My understanding is that most US GPUs are designed in the US and manufactured in Asia outside China using machinery developed in the Netherlands and elsewhere around the world. See here.)
China might respond in turn with similar GPU export controls against the US, but given the limited reach of Chinese compute in the US, I expect the US would welcome those restrictions as they further prevent American dependence on Chinese supply chains. A stronger response could include export controls against the US in other sectors or other actions against US foreign policy interests around the world.
Over the coming years and decades, China could gain more influence over Taiwan, South Korea, Singapore, Japan, and other Asian countries with strong chip manufacturing. This could be a real problem for the US which depends on exports from those countries. But promoting US chip independence by funding domestic manufacturing and cutting ourselves off from global supply chains seems like a good way to counter that threat.
How does this support US strategic goals?
These policies together support independence for the US chip supply chain. This is an important long-term goal for the US given that much of the world's compute is produced in East Asia where China could attempt to restrict our access. It's part of a larger strategy of economic decoupling from China in preparation for another Cold War. (Another policy that would promote US supply chain independence is preventing US companies from using Chinese parts in building chips -- perhaps this is on the agenda, or will be enforced unilaterally by the Chinese government in retaliation to our policies.)
The policies also slow down Chinese AI. Some of the companies that are specifically targeted seem particularly heinous such as SenseTime and Dahua which provide facial recognition software for the concentration camps in Xinjiang, while others are more neutral, such as AMD's partnership with Chinese companies to build CPUs. From a US national security perspective, slowing down Chinese growth in general and military AI growth in particular seem widely accepted as a good strategy. I would expect the same thing from an x-risk perspective, though perhaps decreasing Chinese reliance on US chips only reduces our influence during a future critical time (see Matt Yglesias).