China’s export-licensing grip on indium phosphide (InP), the crystal that makes AI data centers’ optical links possible, has pushed wafer prices up as much as 250% since Beijing added the material to its control list in February 2025. Washington is now drafting a mirror-image restriction of its own, on the Chinese-made transceivers that use it.
The Mirror, Named
It’s a trade war fought in wafers.
Beijing’s permit regime on indium phosphide substrates and the U.S. Federal Communications Commission’s draft ban on new Chinese optical transceivers look, on paper, like two unrelated regulatory actions on two ends of the same supply chain. Read them as gates instead of policies, though, and the resemblance stops being coincidental: each side controls the point its rival cannot manufacture around, and each is exercising that control at the same moment, over the same industry, for the same underlying reason.
Some analysts, including Jefferies, read the US move as a negotiating tactic timed to Xi Jinping’s coming Washington visit rather than a settled policy, a caveat worth keeping in view before treating the mirror as more than a mirror.
The Confirmed Half — What China Already Did
AXT Inc., the California-headquartered substrate maker that supplies roughly a third of the world’s InP, disclosed the arithmetic of the restriction in its most recent quarterly filing with the composure of routine bookkeeping: revenue from Taiwan up 368.5% quarter over quarter, Europe up 473.1%, Japan up 107.7%, all attributed plainly to newly granted Chinese export permits.
North America, over the same quarter, fell 28.1%. The filing states, without elaboration, that this is “primarily due to the impact of China export restrictions.” A company’s shrinking American revenue line, filed under ordinary disclosure obligations, is doing the work a diplomatic protest usually does.
Coherent and Lumentum have responded not with public complaint but with money: a $22.3 million prepayment and an $87 million capacity reservation, respectively, both to keep AXT’s Chinese production running.
The Untested Half — What America Might Do
The FCC’s draft would bar new-model imports of Chinese transceivers, a rule about a single component.
Widen the frame and it’s a rule about which country’s hardware is allowed inside a data center. Widen it again and it’s a rule about whether Chinese-made hardware is allowed inside American infrastructure at all, the same anxiety that shaped the government’s decade-long fight against Huawei. Widen it once more and the transceiver ban stops being a transceiver ban entirely: TrendForce warns the same logic could extend to co-packaged and near-packaged optics, the switch-level engines that transceivers plug into, at which point the rule is no longer regulating a product category but regulating an architecture.
None of this has been finalized. The FCC’s own sourcing on “new models,” on which manufacturers count as Chinese, and on transition timing remains, in TrendForce’s phrasing, undefined enough that “it is believed to be too early to interpret the proposal as a comprehensive ban” — a gap the market has already priced in anyway, in the roughly 9% single-day slide of China’s CSI300 telecom index.
Displacement, Not Evacuation
$8 billion rotated into Hong Kong-listed names in the days after the FCC draft leaked, even as Zhongji Innolight and its optics peers slumped roughly 10% on the same news.
Capital didn’t leave China. It moved to a different part of it.
The mechanism was a valuation gap that predated the news: Hong Kong-listed names trading around 12.2x earnings against the S&P’s 25.8x and the mainland CSI300’s 14.2x. The FCC draft didn’t create that gap. It gave institutional money a reason to close it on a Tuesday instead of eventually.
The same underlying move shows up a layer over, in memory chips instead of optics.
CXMT and YMTC, China’s two largest DRAM and NAND makers, have gone from price-takers to price-setters over the past year, now dictating supply terms to customers that include ByteDance ($7 billion) and Tencent ($3 billion). CXMT’s revenue is up roughly 700% year over year; YMTC is reportedly targeting a $148 billion valuation.
Beijing has since revised its IC design-protection rules in a way that locks the shift in as policy rather than leaving it a market accident. Different component, different customers, same shape: a restriction lands on one node of the supply chain, and the leverage simply resurfaces one layer over, in whatever the restriction didn’t name.
Both moves share the same ceiling.
CXMT and YMTC’s leverage runs only as far as ASML’s export-controlled EUV lithography tools allow it to; the Hong Kong rotation only works as long as a valuation gap exists to be closed. Neither is a sovereign chokehold. Both are leverage that arrived because something specific was left unrestricted, which is the same read the InP dossier’s own supply data supports: control one gate and the material, or the capital, finds the gate you didn’t build.
The Chip Version of the Same Gate
Think of supply-chain leverage less like a wall and more like water: block one channel and it doesn’t disappear, it finds the next-lowest point. Google’s planned 2027 shift of Pixel assembly to Vietnam and India is the water finding a new channel at the visible layer, the one investors and headlines actually track. It solves nothing at the layer beneath it.
Assembly relocation looks like decoupling. It isn’t.
Because the component pricing power that actually governs the phone’s cost structure sits with CXMT and YMTC, China’s memory chipmakers, and neither company’s address changes when Google’s factory does. Samsung proved in 2019 that the assembly move itself is executable; Apple’s harder, partial version of the same move (roughly 25% India-assembled, at 18 times Google’s volume) proves it scales.
Proves it’s finish-able. Doesn’t prove it’s sufficient.
Open up the Matryoshka doll one layer further and the recursion actually closes: assembly sits inside components, components sit inside the tooling that makes them, and the tooling bottoms out at ASML’s EUV lithography systems, the one constraint even CXMT and YMTC can’t route around.
That’s the floor. Everything above it, including a factory move to Hanoi, is rearrangement, not resolution — the same structural verdict the InP coverage above reaches about export permits: the visible fix and the actual leverage live at different depths, and moving the visible one doesn’t touch the other.
Aklatan’s news and analysis drills down to the structural mechanics, geopolitical shifts, and hidden constraints truly driving AI and Asian tech ecosystems and knowledge work.
See coverage span here: Aklatan’s News and Analysis
Generative AI Transparency:
This news article was written primarily with generative AI, specifically SupraGraphos’ A.C.E. News Module. Reviewed with human post-editing, all sources and claims are confirmed as of the time of writing.