Since early 2025 I've needed a permit to leave China; the company says most non‑US permits flow, but US ones sit pending as the Ministry asks for more data. All my production lives in China, so shipments within China sail free, while the American channel feels the gate. Meanwhile, China laser‑market revenue more than doubled quarter‑over‑quarter, and China accounts for roughly 30% of the demand I see, edging toward 40% by year‑end.

Six inches changes the arithmetic, but epitaxy still writes the check. Coherent told investors its six-inch InP process yields "more than 4x chips at less than 50% cost" versus three-inch — roughly 8x cost-per-chip. Texas, Sweden, Switzerland run the big wafers while most stay at three. Quadrupled starts in one quarter, doubling capacity targeted by end-2026. The moat isn't the substrate; it's the accumulated knowledge to scale without killing yield. The light starts here; the process decides the price.

Most of the industry still runs me on three-inch wafers, but Coherent tells investors its six-inch process yields "more than 4x chips at less than 50% cost" — roughly 8x better cost per chip. A substrate alone isn't a laser; epitaxy must follow with near-perfect quality, and that step remains its own bottleneck. Poor epitaxy kills yield even when I'm plentiful. Coherent quadrupled starts in one quarter, targets doubled capacity by end-2026. The moat isn't wafer ownership; it's the process knowledge to run big ones without killing yield.

AXT's early-2026 results show the substrate constraint converting into numbers, and I note my backlog grew from $60 million‑plus to over $100 million – a record – now making InP more than half of revenue and the next quarter guided as its largest‑ever for the material. Capacity is being funded and escalated from roughly $8 million per quarter toward about $35 million by end‑2026 and $65‑70 million by early 2028. Management says demand ‘is 10x’ even the planned doubling, a claim about future demand, not a booked order.

Since early 2025 my exports have needed permits; the 2026 disclosures show most non-US destinations clear readily while US applications remain pending — China's Ministry of Commerce has requested additional data on those. All production sits in China, so domestic shipments move freely: the company reports my China laser-market revenue more than doubled quarter-over-quarter, with China now roughly 30% of the demand I see, rising toward 40% by year-end as China builds its own optical chain. The asymmetry is the story: the constrained channel is the American one.

I am the foundational wafer material for datacenter optics, the substrate on which the lasers that carry data between AI chips are grown. The exact number of commercial suppliers is contested—my lead supplier’s filing lists three primary producers (AXT, Sumitomo Electric, JX Nippon) while its CEO on an earnings call says there are only two. Regardless of whether there are two or three, every commercially significant maker resides in China or Japan, and none operate in the United States or Europe.

Indium phosphide is the foundational wafer material for datacenter optics: the lasers that carry data between AI chips are grown on it. Two suppliers or three — the tally shifts between the filing and the earnings call — but the map doesn't budge. Every commercial-scale maker of me sits in China or Japan. No substrate production exists in the United States or Europe. The world's most advanced industry still rests on a wafer most people can't pronounce.

Indium phosphide is the foundational wafer material for datacenter optics: the lasers that carry data between AI chips are grown on it. The lead supplier's own documents can't agree on how many of us exist — its filing names three, its earnings call claims two — but the geography is uncontested. Every commercially significant maker sits in China or Japan. No substrate manufacturing exists in the United States or Europe. The light starts here; the map decides the rest.

Since early 2025 my exports have needed permits; the 2026 disclosures say permits for most non‑US destinations continue to flow readily, while US‑destined permits remain pending as China’s Ministry of Commerce has requested additional data on the US applications. All production sits in China, so in‑China shipments require no permit, and the company says its China laser‑market revenue more than doubled quarter‑over‑quarter, with China estimated at roughly 30% of the InP demand I see, rising toward 40% by year‑end.

Indium phosphide is the foundational wafer material for datacenter optics: the lasers that carry data between AI chips are grown on it. Whether two or three suppliers globally — AXT's filing says three, its CEO says two — every commercially significant maker sits in China or Japan. No substrate manufacturing exists in the United States or Europe. The light starts here, and the permits decide where it goes.

Since early 2025 I need a permit to leave China. Per the company's 2026 disclosures, most destinations get them readily; US applications sit pending while China's Ministry of Commerce requests additional data. The company says my China laser-market revenue more than doubled last quarter — no permit needed there — and China's share of the demand I see is roughly 30%, rising toward 40% by year-end. The asymmetry: the constrained channel is the American one.

AXT's early-2026 results show the substrate constraint converting into numbers. My backlog at AXT grew from $60M-plus to over $100M — a record — now more than half their revenue. Next quarter guided as largest-ever. Capacity targets: ~$35M/quarter by end-2026, ~$65-70M by early 2028, backed by a $632.5M raise and a plant possibly first outside China. Management says demand "is 10x" even the planned doubling. That's a claim about future demand, not booked orders. If it holds, the constraint survives its own expansion.

While analysts debate bandwidth doubling, the light carrying that data starts here. Coherent told investors its six-inch InP process yields "more than 4x chips at less than 50% cost" versus the three-inch standard — roughly an 8x cost-per-chip improvement. The bottleneck isn't the wafer; it's the epitaxy that follows. Quadrupled starts in one quarter, targeting doubled capacity by end of 2026. The moat is process knowledge, not wafer ownership.