Locking Up the Wafer Upstream: Decoding the Coherent–AXT 3-Year InP Supply Agreement and How It Pushes the Shortage Into a Capacity Grab

2 days ago

6 min read

On June 25, 2026, AXT subsidiary 北京同美 signed a three-year "Master Development and Supply Agreement" with Coherent for 6-inch indium phosphide (InP). On the surface it is a supply contract; in reality, one of the largest buyers is paying cash up front to lock in the scarcest wafer upstream. It sends three signals: 6-inch InP substrates have moved from "waiting in line to order" to "grabbing capacity and locking the source"; 6-inch volume production is an irreversible direction; and under the shadow of China's export controls, major optical module makers are being forced to choose between "locking in China's cheapest capacity" and "Non-China supply insurance."

1. Why Now: One Contract Puts the Most Boring Link in the Spotlight

Over the past year, the market's attention has gone almost entirely to GPUs and co-packaged optics (CPO). But what truly has AI optical interconnect by the throat is a 6-inch wafer nobody photographs — InP. The EMLs and CW lasers that "emit light" in high-speed optical modules are all grown on InP; even silicon photonics (SiPh) solutions don't emit light themselves and need an external InP CW laser to light up. We broke down this shortage chain fully in The "Engine of Light" Shortage Map: InP Is the Deepest Bottleneck in AI Optical Interconnect.

The timing of this deal has hard context behind it. Since China imposed export licensing controls on InP-related materials in February 2025, 6-inch InP substrate prices have been pushed up roughly 250%; Lumentum's capacity is reportedly sold out through 2028 and still short of demand; AXT's own InP revenue jumped more than 250% quarter-over-quarter in Q3 2025, a three-year high. When a material is simultaneously the scarcest, the most expensive and the most concentrated in capacity, a contract is no longer just procurement — it's positioning.

2. Breaking Down the Deal: Three Years, 6-Inch, US$22.3M Prepayment

Lay out the publicly disclosed terms, and the point of this contract isn't "how much is bought" but "how it's locked in":

Nature: A three-year Master Development and Supply Agreement — not a one-off purchase order but a long-term contract for joint development plus volume supply, targeting 6-inch InP wafer substrates.
Capacity expansion commitment: AXT will expand its Beijing plant capacity from 2026 to 2028 and reserve a portion of the agreed volume "exclusively" for Coherent.
Prepayment: Coherent prepays about US$22.3 million, to be credited against future wafer purchases; if unused when the agreement expires, it is refundable subject to minimum purchase volume and capacity conditions.
Priority rights: The money buys not only product but also the right to "obtain additional capacity on the same terms with priority" — effectively cutting the line in the next round of expansion.

In plain terms: Coherent uses cash to back AXT's expansion plan, and AXT pays it back with capacity and line-cutting rights. It's a "you fund the expansion, I get first dibs" lock-in, one level deeper than an ordinary long-term contract.

3. What This Structure Says: Not Procurement, but "Supply Insurance"

To grasp the weight of this contract, first clear up a common misconception: many assume that since Coherent has built 6-inch InP lines in Sherman, Texas and Järfälla, Sweden, it shouldn't be short of substrates. Wrong. Those Coherent lines do epitaxy and device fabrication; it doesn't grow InP ingots itself — blank substrates still have to be bought externally, long secured through 3-to-5-year contracts with five or more suppliers. What it controls is "what grows on the wafer," not "the wafer itself."

So this AXT contract fills exactly the bottom layer: blank substrates. Trading a prepayment for expansion and priority capacity is essentially using cash to lock in price and volume of "the scarcest, most expensive raw material for the next three years" ahead of time — this isn't purchasing, it's supply insurance.

Compare it with another contract 北京同美 signed recently, and the difference is clear. On June 11, 2026, 北京同美 signed a long-term contract with Nanjing-based Casela: Casela committed to buying InP substrates for all of 2027, worth about RMB 173 million (about US$25.4 million, roughly 26% of AXT's annual revenue), delivered monthly, with a 50% prepayment and a minimum 80% purchase or else liquidated damages. Casela's is a "single-year, volume-and-price-locked pure purchase contract"; Coherent's is a "three-year, joint-development, priority-capacity" strategic lock-in. The former locks a year of supply; the latter holds a position for three years.

4. Market Impact: Four Lines Pulled at Once

First, an already tight InP substrate market loses another chunk of freely circulating volume. With AXT reserving agreed volume "exclusively" for Coherent, a slice is pulled out of the open market before total capacity catches up. For smaller module makers and epi houses further back in line without long-term contracts, getting supply will only be harder and pricier. Taiwanese companies feel this most — VPEC and LandMark Optoelectronics have both been squeezed before by disrupted AXT supply.

Second, 6-inch volume production gets stamped once again. This contract targets 6-inch, not 4-inch. When one of the biggest buyers is willing to lock 6-inch capacity with a three-year deal plus prepayment, it's a heavy bet that "6-inch is the volume mainstream for InP." A 6-inch wafer yields more than twice as many dies as a 4-inch one; once the cost curve gets going, it will squeeze legacy capacity still centered on 4-inch.

Third, supply strategies are clearly diverging. On one side is Coherent — locking in China's lowest-cost VGF 6-inch capacity (while still keeping five-plus suppliers to spread risk); on the other is LandMark Optoelectronics — which in April 2026 signed a long-term contract with Japan's Sumitomo Electric instead, taking a de-China alternative-source route. It's a route choice of "locking in China's cheapest capacity" versus "Non-China supply insurance," and behind it is a single variable: are you betting that China's export licenses will keep being granted?

Fourth, this is one link in a series of "grab materials upstream" moves. The same month, IQE and Tower Semiconductor signed a multi-year InP epiwafer supply agreement; earlier, NVIDIA took a US$2 billion stake in Coherent and broke ground on a 6-inch InP fab in Texas. When buyers keep moving upstream — taking stakes, locking capacity, signing long-term contracts — it means the whole industry has concluded: the next bottleneck isn't the GPU, it's this wafer. For the context of this capital and positioning, see NVIDIA Buys Up Optics: When Optical Communications Goes From Growth Stock to Index Heavyweight.

5. Risks and Counterarguments: This Contract Also Places a Bet

Looked at calmly, this kind of lock-in isn't free.

It locks in exactly the capacity with the highest political risk. AXT's expansion is in Beijing, and the Beijing plant is a direct target of China's export controls. By locking China's 6-inch capacity with a three-year deal, Coherent is buying supply insurance in a place where "the policy itself could be frozen by controls" — which also explains why the contract includes the "refundable prepayment" proviso: it's a hedge, not an unconditional commitment.

Demand isn't one-way up either. The market has already pushed the real ramp of CPO out to 2028, extending the shelf life of pluggable optical modules; if the pace of AI capex slows, whether this prepaid capacity lands at the right time remains uncertain.

The real choke point of 6-inch yield and large-diameter high-grade wafers isn't solved yet either. What's short has never been just "InP," but "enough large-diameter semi-insulating substrates of good enough quality." Expanding capacity doesn't equal expanding yield; this still depends on the actual ramp from 2026 to 2028.

6. Conclusion

What's most worth remembering about the Coherent–AXT three-year deal isn't the US$22.3 million figure, but the behavior pattern it reveals: when a material is simultaneously the scarcest, the most expensive and the most concentrated in capacity, a rational big buyer no longer waits in line to order — it uses cash to lock expansion, lock priority, lock three years. The InP shortage has escalated from "grabbing product" to a positional war of "grabbing capacity and locking the source."

Three signals to watch next: first, whether China tightens InP export licensing further — that directly determines whether Coherent's insurance policy pays out; second, who else follows with long-term contracts or upstream stakes, especially whether Taiwan's epi and laser makers choose "lock in China" or "Non-China"; third, the actual ramp speed of 6-inch capacity from 2026 to 2028, the real physical time lock of this war. Whoever first locks in enough good 6-inch substrates gets the first ticket to the 1.6T and 3.2T generations.