The shape of the chip market in the wake of the Chinese ban
It is very unclear what America is trying to achieve
The dust is now settling on the ban on high-speed chips for China. We can now get an outline of what the new world looks like. Reuters have run an excellent overview of how chipmaker Nvidia are responding to the restrictions.
The chipmaker appears to be purposely downgrading the capacities of one of its high-speed chips, the A100, so that they can sell this new downgraded version to China. The new downgraded version is called the A800:
A distributor website in China detailed the specifications of the A800. A comparison of the chip capabilities with the A100 shows that the chip-to-chip data transfer rate is 400 gigabytes per second on the new chip, down from 600 gigabytes per second on the A100. The new rules restrict rates of 600 gigabytes per second and up.
“The A800 looks to be a repackaged A100 GPU designed to avoid the recent Commerce Department trade restrictions,” said Wayne Lam, an analyst at CCS Insight, basing his comments on the specs shared by Reuters, and noting that eight is a lucky number in China.
Lam said the chip-to-chip communications abilities of the A800 represented a clear performance downgrade for a data center where thousands of chips are used together.
This strategy raises a key question for these restrictions moving forward: is the goal of the US government to freeze China into having access to chips that produce 400gb/s, or is the goal to keep a permanent lag of ~200gb/s on Chinese products relative to American products?
The law itself does not help in interpreting American intentions. Here is the relevant section of the October 7th statement from the Bureau of Industry and Security:
6.) Adds new license requirements for items destined to a semiconductor fabrication “facility” in the PRC that fabricates ICs meeting specified. Licenses for facilities owned by PRC entities will face a “presumption of denial,” and facilities owned by multinationals will be decided on a case-by-case basis. The relevant thresholds are as follows:
• Logic chips with non-planar transistor architectures (I.e., FinFET or GAAFET) of
16nm or 14nm, or below;
• DRAM memory chips of 18nm half-pitch or less;
• NAND flash memory chips with 128 layers or more.
So, the law, as it now stands, freezes Chinese tech at <=400gb/s. There is no indication in the document that this will be changed as technology developed. But there is every likelihood that it will be.
Let us consider the two scenarios. We will call the first, where there is a permanent freeze in exports to China at <=400gb/s, a ‘static ban’. We will call the second, where the ban is updated as technology improves and China is held at a -200gb/s lag, a ‘sliding-scale ban’.
A static ban would be truly bizarre. It would be reminiscent of strategy video games like Civilization and Age of Empires, where one civilisation is subject to a permanent restriction on technological development. In this scenario, the US would seek to restrict China to the technological dark age while technology continues to improve.
This would be a very strange strategy. Restricting technology by fiat is remakably difficult. The Catholic Church famously tried to get control over the printing press through its system of ‘impramatur’. But it did not work very well and the Protestant Reformation occured regardless.
Even restrictions on technology for top secret military technology can be difficult if the state actor that is being restricted is sufficiently sophisticated. During the Cold War, the Western democracies tried to restrict the Soviet Union from gaining access to multiple types of technology, but typically the Soviets managed to get access to the technology; usually through espionage. The most famous instance being the theft of Britain’s nuclear secrets by the Soviet spy Klaus Fuchs.
Such restrictions on commercially-available technology seem very unlikely to be enforcable. In the face of a static ban, China would likely feel justified in resorting to whatever tactics necessary to gain access to new chip technology. Since the chips would be commercially-available, China could likely have intermediaries set up shell companies in friendly countries to buy the chips and then transport them to China. Those involved in global black market activities, such as running guns into embargoed warzones, would be massively incentivised to get in on the action.
Trying to restrict China to the technological dark age woulod therefore likely turn the international chip war into something like the international drug war. The US authorities would be constantly fighting against decentralised and highly incentivised black markets elements. As with the international drug war, outright victory seems improbable.
The sliding scale ban may not elicit as aggressive a response from the Chinese. But it would come with downsides of its own. If the definitions of the ban are fluid and change quite regularly — say, every 3-5 years as technology develops — this will most likely create a lot of grey areas in the laws. Companies will be strongly incentivised to look for loopholes to export the best — and most expensive — technology to China.
While a sliding scale approach seems more realistic than a static ban, it still seems very ambitious. Is there a precedent for something like this? One involving a country as sophisticated, wealthy and well-connected as China? I am not aware of any. I would be very surprised if there is precedent for this sort of policy.
No matter how this pans out, however, China now has a massive incentive to develop chip-making technology for themselves. They have plenty of money to do it and have shown how good they are at industrial policy in the past. It also seems likely that they have plenty of capacity for industrial espionage. Klaus Fuchs had to infiltrate top secret government projects; his Chinese counterpart today just has to get a job at a chipmaker!
The Chinese chip ban of 2022 will almost certainly go down in history as a failure. If you want to put your money on China remaining in the technological Dark Ages, be my guest. I’ll sit that bet out.
It doesn't have to keep them in the technological dark ages forever. Just long enough that they don't build skynet first.