The Bottleneck Under the Bottleneck

Why Indium Phosphide Is the Chokepoint

Every high-speed optical link in an AI data center needs a light source — a laser — and the fastest, most efficient lasers and photodetectors are built on indium phosphide (InP), a III-V compound semiconductor. Silicon, for all its dominance in logic chips, is a lousy light emitter; InP is a natural one. That makes it effectively irreplaceable for the laser and modulator layer of both pluggable optics and next-gen co-packaged optics (CPO).

The problem is supply. InP is genuinely hard to manufacture: the crystals are brittle, the wafers are small (the industry is still largely on 3-inch and 4-inch substrates while silicon runs 12-inch), and yields are low. The world's substrate capacity sits with a handful of players. Then geopolitics landed on top: in February 2025, China added indium phosphide to its export-control list, so shipments out of China now require case-by-case government permits. When demand for the material is inflecting and the supply is both scarce and politically gated, you have the textbook definition of a chokepoint. That's the setup here.

AXT: The Pure Play Just Inflected — Hard

AXT (AXTI) is the only US-listed pure-play substrate maker — it produces InP, gallium arsenide and germanium wafers, and InP is now the whole story. For years it was a sleepy, cyclical, loss-making materials company. Then AI optics found it. The June quarter, reported July 30, was a step-function:

Read the trajectory: revenue up 164% year over year, InP revenue a company record and more than double the prior quarter, gross margin from 8% to 45% as idle capacity filled and fixed costs leveraged, and a swing from chronic losses to real profit. Then management guided Q3 to ~$66M and $0.30–0.32 EPS — against Street consensus near $39M and $0.10. That's not a beat, it's a re-rating: backlog exceeds $100M, the InP lines are running full-out, and incremental output is spoken for before it comes online. The capacity roadmap runs from ~$35M/quarter of InP by end-2026 to $65–70M/quarter by 2028, now funded from cash flow rather than new raises.

The stock reflects the whiplash: AXTI ran from under $3 to $143 on this theme, then gave back more than half to ~$67. So even a quarter this good arrived after a run that had already gotten ahead of itself. You're buying a genuine inflection at a still-demanding valuation — roughly 50x forward earnings and ~11x a forward-revenue run-rate — on a company whose quarterly revenue is still measured in tens of millions.

The Catch: The "US-Listed" InP Play Is Really a China Producer

Here's the part that decides everything. AXT is headquartered in Fremont, California — but it manufactures its indium phosphide in China, through its subsidiary Tongmei. And because China now controls InP exports, AXT needs a Chinese government permit for every shipment leaving the country. Its own risk factors lead with exactly this: export-permit receipt, Chinese private-equity redemptions in Tongmei, the administrative maze of listing Tongmei on Shanghai's STAR Market, and US–China geopolitical tension.

So the paradox: the cleanest way to buy the "Western AI supply chain" bottleneck is a company whose product sits behind China's export-control wall. Two implications fall out of that, and they cut in opposite directions:

• Bearish: AXT's ability to convert its record backlog into US- and Europe-bound revenue is partly at the discretion of Beijing. A permit delay is an air pocket in the numbers — it's happened before — and it's entirely outside the company's control.
• Bullish (and underappreciated): shipments within China don't need export permits, and Chinese demand is inflecting on its own domestic AI and semiconductor buildout. So AXT is increasingly a play on China's internal optics supply chain, where it has a home-field advantage — plus a potential value-unlock if the Tongmei STAR listing goes through.

Net: AXT is the purest and highest-torque way to own the InP bottleneck, but it is a geopolitically-hostage micro-cap that can gap 20% on a policy headline in either direction. That's the trade you're actually making.

The Swing Factor Nobody Prices: 6-Inch Wafers

One technical wrinkle worth understanding, because it could break the whole constraint. The industry runs InP mostly on 3- and 4-inch wafers; the transition to 6-inch InP roughly doubles the usable area per wafer and slashes per-die cost. Whoever masters 6-inch at yield resets the supply curve — and their own margins. AXT is investing R&D directly here, and it's arguably the single most important long-term lever in the substrate story. It's also a two-edged sword: success relieves the scarcity that's currently handing everyone pricing power. For now it's optionality; watch it.

The Alternative: Coherent Owns Its InP Outright

Now the other way to own the bottleneck — and the reason the COHR-vs-Lumentum debate keeps circling back here. Coherent (COHR) doesn't buy its indium phosphide; it grows it, in its own fabs in Sherman, Texas; Sweden; and Switzerland, and it's doubling that output. Then it captures the laser and module margin stacked on top. Nvidia is funding that capacity build directly. In other words, COHR internalizes the exact bottleneck AXT sells into — with none of the China export-permit exposure.

The trade-off is clean. AXT gives you undiluted exposure to the scarcest link, with explosive growth and geopolitical landmines. COHR gives you the same material moat — Western-owned, Nvidia-funded — inside a large, profitable, diversified business, so the InP scarcity shows up as one driver among several rather than the entire P&L.

The Verdict

Best risk-adjusted way to own the bottleneck: Coherent (COHR). For a Western investor, owning the InP chokepoint without the China export-permit risk — and getting the downstream laser margin on top — is simply the higher-quality expression of the same idea. COHR's owned InP fabs are the direct hedge against exactly what makes AXT fragile. It's the core holding.

Purest and highest-torque, but a satellite: AXT (AXTI). If you want the undiluted bottleneck and can stomach the risk, AXT is the sharpest instrument in the market — a real, just-confirmed inflection (164% growth, 45% margins, sold-out capacity, a Q3 guide that lapped consensus) trading at ~50x forward after already halving. But it manufactures in China, it's a ~$3B micro-cap that moves violently, and its revenue is still tens of millions against a theme worth trillions. I'd treat it as a small, catalyst-driven satellite — sized so a permit headline or a hyperscaler timing shift can't hurt the core — and I'd trade it around the catalysts: the Q3 print, permit news, the Tongmei STAR listing, and the 6-inch InP ramp.

How I'd hold them together: COHR as the position, AXTI as the option. That pairing gives you the durable, Western-supply-chain version of the InP moat, plus a small slug of pure, high-beta upside if the substrate scarcity keeps inflecting. Don't invert it — a micro-cap hostage to Beijing is not where the core belongs.

The Headwinds That Could Break This

• China export permits (AXT). The defining risk. A permit freeze turns record backlog into missed revenue overnight, and it's entirely outside the company's control.
• The 6-inch transition cuts both ways. It's the long-term unlock, but if it succeeds broadly it relieves the very scarcity that's handing AXT (and COHR) pricing power today.
• AI-capex digestion. InP demand is a leveraged read on optical-interconnect orders, which are a leveraged read on hyperscaler capex. A slowdown hits the smallest, highest-multiple name (AXT) hardest.
• Technology substitution. Silicon photonics and thin-film lithium niobate aim to reduce InP content per link over time. It's not a near-term threat to the laser itself, but it caps the very-long-term ceiling.
• Valuation and size. AXT at ~50x forward on ~$48M quarterly revenue has priced in a lot of the inflection already; the stock's own 50x-then-halved round trip is the warning label.

Bottom Line

The AI-optics trade has layers, and indium phosphide is the deepest one that's still investable — the scarce material every laser is grown on. There are two doors. Coherent (COHR) is the one I'd walk through for a core position: it owns Western InP capacity, captures the laser margin, is Nvidia-funded, and carries none of the China permit risk. AXT (AXTI) is the purest, most explosive way to own the raw bottleneck — and it just proved the inflection is real — but it's a geopolitically-hostage micro-cap that belongs in the satellite sleeve, traded small and around catalysts. Own the moat; rent the option. And keep one eye on the 6-inch wafer, because the day scarcity eases is the day this whole thesis changes.

Prices and technicals pulled live from Yahoo Finance as of the August 25, 2026 close. AXT Q2 2026 results (reported July 30, 2026) and Q3 guidance from the company's BusinessWire release and StockTitan/Pomegra/InfoArb coverage; Coherent figures from its FQ4 FY2026 release. Valuation metrics are approximate and depend on share-count and consensus assumptions. This is a small-cap with elevated volatility and real geopolitical risk. Levels and multiples are point-in-time and will move.