Anatomy of a viral thesis: China is throttling indium phosphide — the substrate behind every laser in every AI data center — and one widely-shared tweet has drawn a 9-stock map of winners and losers. We pressure-tested all of it: the physics, the fundamentals, and the consensus arc from before the dip to today.
On 13 June 2026, @TomSzczypka posted a tightly-argued thread framing China's indium phosphide (InP) export controls as deliberate "supply-chain warfare over AI supremacy," and named a basket that supposedly explains "the recent price action." Here it is, decoded.
InP is the only commercially viable direct-bandgap substrate for the high-speed lasers (EMLs, CW-DFBs) that move data inside AI clusters. Silicon can't emit light efficiently at 1.3–1.55µm. There is no drop-in substitute at the volumes and speeds the 800G→1.6T transceiver ramp demands.
China ≈70% of global indium (USGS)
AXT (in China), Sumitomo & JX (Japan)
Lumentum, Coherent, AAOI, Sivers
800G / 1.6T modules
GPU-to-GPU interconnect
| 6-inch InP wafer price | ~$1,400 → ~$5,000 | +250% |
| China share of global indium | ~70% | USGS 2024 |
| AXT InP order backlog (Q1'26) | >$100M | vs $96M TTM rev |
| Supplier qualification cycle | 2–3 yrs | tweet says 12–18m |
The control regime began with Decision No. 10 of 2025 (4 Feb 2025), which placed indium and indium-related items — explicitly including InP substrates — under export licensing. Critically, when China suspended its gallium/germanium/antimony bans for the US in the November 2025 trade truce, InP was left out. It remains the live lever. AXT's Beijing subsidiary saw its first permits only in late June 2025, then a stop-start cadence through 2026 — each permit batch takes ~60 business days, giving Beijing a tap it can turn on and off at will.
Here's the arc, not just a snapshot: how the "China-weaponizes-InP" narrative evolved before the dip, during it, and now. The short answer — it was a niche compound-semiconductor view in early 2025 that went fully mainstream the week of the tweet.
By 14 June 2026 this is a crowded trade. The trifecta that made it consensus: Nvidia's $4B investment into Coherent + Lumentum (2 Mar 2026), Lumentum being sold out through 2028, and the Reuters wire (10–11 Jun) that put it on every screen. The tweet itself rode the very top of that wave — it's a clean summary of consensus, not a discovery of edge.
Gallium/germanium controls (Aug '23) seen as a radar/EV/defense story. AI-optics thesis was purely demand-side ("hyperscalers buy more 800G"). AXT a distressed sub-$2 micro-cap. InP-as-chokepoint lived only in niche III-V circles — the GaAs permit template was hiding in plain sight.
Edge — ignoredFeb 4 controls land. AXT's Q2'25 revenue collapses −36% on zero permits. Bulls frame it as temporary; the contrarian "bottleneck owners win" view starts forming on specialist Substacks (TheComputeChain, Yianisz). AXT itself: "InP export permits are the most significant challenge we currently face."
Edge → specialist consensusNvidia's $4B (Mar), Coherent's CEO flying to Beijing with Trump (May), and Reuters (Jun) make it mainstream. SemiAnalysis, Albright Stonebridge, JPMorgan, Needham all on record. The tweet (13 Jun) is the retail crystallization point — and the stocks already reflect 2–3 years of scarcity premium.
Crowded — mostly pricedCompiled from Reddit, X, YouTube, Hacker News & GitHub activity for the window 15 May – 14 Jun 2026 — the social-pulse layer folded into this report.
The tweet is the loud center of a tight cluster. The same TomSzczypka post surfaced independently (74 cluster score) sitting inside a wall of corroborating June 11 posts — @kyleichan ("China produces 70% of the world's indium," 81 likes), @pequityresearch (Coherent CEO joined Trump's China trip), and @LIWEI_TWCapital, who posted the identical ticker list — $AXTI $LITE $COHR $AAOI $SIVE. When five accounts independently converge on the same basket in 72 hours, that's the signature of a crowded, late-stage narrative, not a hidden gem.
The contrarian voice is the most valuable signal in the set. Buried in the X data, @arithesis threw the cold water everyone else avoided:
And on YouTube, Denver Stock Lab flagged the specific downside trigger: "sustained pressure from US export controls could see LITE retesting support at $854.96." The crowd is euphoric; the smart skeptics are watching earnings quality and CPO timing — exactly the right things to watch.
Each card carries the tweet's claim, our accuracy verdict, the price action (the "dip" and the run), the fundamentals, and Jeff's conviction on the InP linkage. Click any card to expand.
The thesis is real but late. Owning the bottleneck is correct; paying any price for it is not. Ranked by thesis-validity × valuation reasonableness:
The cleanest claim in the thread and the best business. Japanese substrate supply (China-immune), Nvidia lock-in through 2028, $2.49B revenue actually inflecting. Risk is entirely valuation (31× sales, 61× fwd P/E) and SiPh encroachment at 1.6T+. Own it; add on optical routs, don't chase vertical days.
Direct, primary-source confirmation: Lumentum placed G10-AsP MOCVD orders for InP. Aixtron wins regardless of which laser-maker wins. FY26 guide raised to €560M on optoelectronics; 13× sales is the most digestible multiple among the high-linkage names. Veeco is the one real competitor.
Down −27% in a month, 8.5× sales — by far the cheapest. But the linkage is second-order: Soitec's Photonics-SOI benefits if SiPh takes share from InP. It's a hedge on the thesis as much as an expression of it. Core risk is its depressed RF-SOI base, not photonics.
Genuine beneficiary as the dominant external III-V foundry, but "only pure-play InP foundry on earth" is wrong — it's primarily GaAs, with InP a growing minority, and IQE/GCS exist. Already up ~8× off the low; optical is only ~15% of revenue.
Tagged a loser for AXT dependence, but it's the most actively mitigating name — first to 6-inch InP, 5+ substrate suppliers, $33M CHIPS money. Profitable, +375% in a year. Near-term substrate risk is real; the "loser" framing is backward on a 24-month view.
The thread's risk analysis is precisely right — the entire equity is a China-permit binary, $100M+ backlog it can't fully ship, $730M raised in dilution. But calling a stock that's up 4,000% a "loser" is absurd; it's the purest beneficiary of scarcity pricing (margins −6%→+30%) for as long as some permits flow. A call option on Beijing, not a short.
The thread says it "can't secure laser chips" — but AAOI fabricates its own InP lasers; that's its moat. Real risk is raw substrate (covered ~12 months out). It's a mid-chain winner with upstream vulnerability, not a loser. And it's the name @arithesis says its own earnings contradict. +899% in a year; show-me on margins.
Genuine InP DFB-laser maker at the CPO/scarcity intersection — but ~80–90× a still-loss-making $31M-revenue business, trading 2.4× above its own bull-case analyst target. Q1 revenue fell 22%. The "14-quarter constraint" claim is unverifiable. Pure momentum.
The weakest link in the thread. Aehr's actual orders are silicon photonics burn-in, not InP — every press release says SiPh. 72× sales, loss-making, and consensus PT ($64) sits 41% below the price. Benefits from AI optics broadly, but the InP attribution doesn't hold.