[Salon] The Annals of Government Idiocy: #ChinaMemoryWatch: How the 2022 Memory Controls Became the Greatest Unforced Technology Transfer in History




Forwarded this email? Subscribe here for more

#ChinaMemoryWatch: How the 2022 Memory Controls Became the Greatest Unforced Technology Transfer in History

The debate over whether fabs should have to choose between US and Chinese tools ignores how we got here. A short unpacking.

Aug 5
 
READ IN APP
 

My friend Chris McGuire argued this week that it is “extremely obvious” the US government should not permit any fab that uses US chipmaking tools to also use Chinese tools — that just as we tell countries they must pick between US or Chinese AI chips

X avatar for @ChrisRMcGuire
Chris McGuire
@ChrisRMcGuire
It is extremely obvious that the US government should not permit any fab that uses US chipmaking tools to also use Chinese tools. We tell countries they must pick between using US or Chinese AI chips (see: UAE). We should tell fabs they have to pick between US and Chinese tools.
X avatar for @ReutersChina
Reuters China @ReutersChina
EXCLUSIVE: Samsung, SK Hynix test Chinese chip tools as hedge against US risks https://t.co/psObvS81Zy
3:21 AM · Aug 5, 2026 · 36.1K Views
69 Replies · 23 Reposts · 142 Likes

(see: UAE), we should tell fabs they have to pick between US and Chinese tools.

This issue is complex and requires some unpacking to understand how we got here. Bear with me.

Yes. It started in 2022

When the Commerce Department chose to include memory as part of the initial October 2022 controls, there was a problem: no one thought about the major Korean producers in China. Samsung and SK Hynix had billions sunk into DRAM and NAND production at facilities in China. I wrote about this at the time in Memory test: The U.S.-China chip war escalates.

A last-minute (literally, minutes) letter temporarily exempting the Korean facilities in China came on October 9, but no clear solution to this issue came from the administration until much, much later — and the issue is still unresolved. I covered much of this in South Korea: Caught in the Crosshairs of U.S.−China Competition Over Semiconductors for KEI’s Korea Policy journal.

In another major irony, in December 2022 I coauthored a white paper for the administration on why memory was different from logic — in terms of both commercial and technology roadmaps, very, very different. This difference became crystal clear over the ensuing four years. [FLAG: rest of this tweet was truncated in the screenshot — add the closing clause]

Now, fast forward three years

Two major new developments — both foreseeable — have massively changed the equation.

First, the rise of CXMT and YMTC. It turned out that both were already operating beyond the 2022 end-use controls, and because of those controls, US firms had to [FLAG: truncated — presumably “pull equipment and support,” confirm wording]. I examine where this has left China’s semiconductor industry in Innovation under Pressure: China’s Semiconductor Industry at a Crossroads in American Affairs.

Industry observers have called this “the greatest unforced technology transfer in history.” Let that sink in slowly: the reason Chinese toolmakers are so competitive now dates clearly and directly to October 2022. This will become important as we get to the crazy conclusion below.

Second, the AI memory supercycle happened. Ironically, the controls on memory were supposed to slow the ability of Chinese companies to develop advanced AI capabilities. By 2026 this approach had clearly failed, but the memory controls remained in place even as demand for memory [FLAG: truncated — complete the thought on the supercycle/demand squeeze].

I have proposed that, at a minimum, there be an updating of the end-use controls: all Chinese firms were operating beyond them already in 2022, and the industry did not consider either the logic or memory thresholds to be “advanced” even at that time. See: Rethinking the “Small Yard, High Fence” Framework [FLAG: confirm this is the right post — thread showed /p/rethinking-t...].

The situation you cannot make up

So now we have arrived at an incredible situation:

1) US controls have given jet fuel to China’s toolmakers, helping them move up the innovation scale and become competitive in etch, deposition, and cleaning — they can serve all nodes in China and most outside China.

2) The memory supercycle means all memory makers want to ramp production, including SK and Samsung in China. The VEU status that was the outcome of policy deliberations at end of Biden administration does not allow expansiion of existing facilities with US tools. Hence

3) Korean toolmakers are qualifying Chinese tools, which are now competitive with US tools because of the controls. But there is more.

4) Because of the memory squeeze, US electronics companies that rely on affordable memory now also want to qualify CXMT and YMTC for their devices. Apple and others have already qualified some CXMT memory, but want guarantees from the administration that there will be no repercussions for using CXMT memory — because CXMT is on the Pentagon’s “Chinese military companies” list, the 1260H list. I address the weakness of the 1260H arguments in here.

5) And finally, yes, of course, Samsung and SK Hynix — not wishing to abandon not-yet-fully-amortized massive sunk investments in China-based memory fabs — want to qualify Chinese equipment to expand those fabs. This is what companies do. So if the memory cycle continues, fabs will want to do more of this. And it works both ways: if the memory cycle slows, all memory makers would now have to consider buying affordable and capable Chinese tools in order to compete with lower-cost producers like CXMT and YMTC — which are using better equipment because of the US controls.

And of course, ALL of this undercuts leading US toolmakers’ ability to participate in the China market and drive R&D.

You really can’t make this up.


This piece is adapted from a thread on X.



This archive was generated by a fusion of Pipermail (Mailman edition) and MHonArc.