Writing/AXT, Tongmei, and the InP supply contracts

AXT, Tongmei, and the InP supply contracts

AXT is a Delaware holding company whose whole manufacturing operation is a Beijing subsidiary called Tongmei, sitting inside a group of part-owned Chinese raw-material producers. Its three new long-term supply agreements are fixed-price, and the one exhibit that got filed has terms the press releases left out.

September 20, 2026

After writing about why indium phosphide ended up in the data center, I wanted to look at the supply side straight from the filings instead of from market-sizing reports, which mostly come from people selling something. AXT is the only InP substrate maker that reports to the SEC. It’s US-listed, headquartered in Fremont, California, and named in most coverage of the InP shortage. So I assumed it was an American manufacturer with a China exposure problem.

The filings describe something closer to the opposite. AXT is a holding company. The manufacturing is done by a Beijing subsidiary, the raw materials come from a group of Chinese companies AXT partly owns, and in the quarter ended June 30, 2026 the company booked $161,000 of revenue in North America out of $47.6 million total.

Sources are the Form 10-K for 2025 (filed March 17), the Form 10-Q for the quarter ended June 30, 2026 (filed August 13), Exhibit 10.3 to that 10-Q, which is the full text of the Coherent supply agreement, and the Form 8-K of July 29.

The operating company is Tongmei

The subsidiary is Beijing Tongmei Xtal Technology, and it’s where the crystal growth actually happens. As of June 30, Tongmei’s noncontrolling and redeemable noncontrolling interests together came to about 14.5%, and the 10-Q says AXT “remains the controlling stakeholder of Tongmei and holds a majority of the Board of Director positions of Tongmei.” That’s a holding company describing control of a subsidiary, not a manufacturer describing a plant.

The 14.5% AXT doesn’t own was sold to Chinese private equity investors in a January 2021 raise, done ahead of a planned listing of Tongmei on the Shanghai Stock Exchange’s STAR Market. That application got withdrawn. The plan now is Hong Kong, and the 10-Q lays out where the 2021 investors stand:

We are in discussions with each Investor to determine whether it wishes to continue its investment as Tongmei advances toward an application for a Hong Kong Stock Exchange listing, or to have its investment redeemed in whole or in part. We have sufficient funds to redeem the Investors’ investments in full should full redemption be required.

The redeemable noncontrolling interest was carried at $38.1 million at the start of 2026. No Hong Kong application had been filed as of the August 10-Q, and the filings don’t give a target date. The board went from five directors to six on July 16, adding two independent directors, both put on the audit, compensation and nominating committees.

The raw materials come from companies AXT part-owns

AXT holds equity in eight Chinese companies that produce its inputs. Five are consolidated, three are carried at equity. Ownership percentages and carrying values are as filed at June 30:

Consolidated, $14.2 million:

Equity method, $14.1 million:

The 10-K lists what they make: raw gallium (4N), high purity gallium (6N and 7N), starting material for InP, arsenic, germanium, germanium dioxide, pyrolytic boron nitride crucibles, quartz tubing and boron oxide. One of them recycles scrap. So the group does reach the InP feedstock itself, not just the gallium side of the business, though the filing only calls it “starting material for InP” and doesn’t say which company makes it or how much.

AXT spells out the reasoning pretty directly: the arrangement “provides us with pricing advantages, reliable supply, market trend visibility and better sourcing lead-times,” and “in the event of industry-wide supply shortages we believe our vertically integrated supply chain strategy will be even more advantageous.” The consolidated raw material companies hold 130 Chinese patents between them (80 of those at BoYu and 34 at JinMei). AXT buys part of their output and they sell the rest to third parties, which is why “Raw Materials and Other” shows up as a reported revenue line instead of an internal transfer.

Two things limit how far to take this. The combined carrying value is about $28 million, which is small against the balance sheet, so these are strategic stakes, not a materials division. And the 10-Q is explicit that AXT doesn’t run them:

Although we have representation on the board of directors of each of the privately held raw material companies, the daily operations of each of these companies are managed by local management and not by us.

The structural point holds anyway, and honestly it’s what separates AXT from its competitors. The 10-K notes that “our primary competitors have their major manufacturing operations in Germany or Japan,” meaning Freiberger, Sumitomo Electric and JX. Those companies buy their high-purity feedstock. AXT owns equity in the firms refining its gallium, making its crucibles, and producing its InP starting material.

What it sells, and where

Second quarter 2026, from the product table:

Substrates covers indium phosphide, gallium arsenide and germanium wafers, grown by VGF in diameters the 10-K gives as one inch to eight inches. InP alone came in at $30.7 million, a record.

Revenue by where the product shipped:

North America is three tenths of one percent of revenue. The 10-Q traces how it got there: North America was 8% of worldwide revenue in 2024 and about 2% in 2025. The causes it gives are US tariffs on wafer substrates, which hit 70%, and China’s export controls. Indium phosphide substrates are named as “the primary revenue generator derived from imports into the U.S.”

Europe went the other way, and the 10-Q puts the increase down to “additional export approvals granted by the China government.” So the geographic revenue line is pretty close to a direct readout of Chinese export licensing, which is a specific mechanism, not the general “country risk” these stories usually talk about.

As for who the European customers are, the filings don’t name them and the concentration disclosure rules out a single account. No customer was more than 10% of revenue in 2025, 2024 or 2023, and the top five together came to 29% of 2025 revenue, 30% in 2024 and 25% in 2023. Europe at roughly a fifth of revenue has to be several customers. AXT sells there through its own salesforce instead of a distributor, and only uses independent representatives in Japan, Taiwan, Korea and some other markets.

Three long-term agreements, signed six weeks apart

All three were signed in June and July 2026.

Casela, June 11. Nanjing Casela Technologies, a Chinese company. Tongmei reserves production capacity and raw-material allocation and gives it supply priority, in exchange for a binding commitment to buy a fixed quantity of InP substrates during calendar 2027 for RMB 173,000,000 (about $25.4 million), delivered monthly. Casela pays half within 15 business days and the rest by December 31, 2026. It has to take at least 80% or pay a cancellation fee. Falling below the 80%, cancelling orders, refusing delivery, or going more than 30 days delinquent each count as fundamental breach, on which Tongmei “may, in its sole discretion, terminate the Agreement and retain all amounts previously paid by Casela.”

Coherent, June 26. A Master Development and Supply Agreement for 6-inch InP substrates: three-year initial term, a prepayment of $22,288,500, and a commitment by AXT to increase capacity at Beijing.

Lumentum, July 26. A Capacity Reservation Agreement over six years, extendable in one-year terms. A $43.5 million deposit within 30 business days and a second $43.5 million whose timing gets set during 2028, both applied as shipment credits. If Lumentum buys less than its annual commitment, it owes the shortfall. If AXT fails to deliver the minimum capacity, that’s a material breach and the unapplied deposit gets refunded. On termination for convenience or “changes in demand,” AXT keeps the unallocated deposits.

The agreements are fixed-price

The revenue note in the 10-Q:

As of June 30, 2026, the aggregate transaction price allocated to the remaining performance obligations under the long-term supply agreements was approximately $98.5 million, excluding applicable taxes. We expect to recognize this amount as revenue through the first half of 2029.

Under ASC 606 a transaction price can only be allocated to remaining performance obligations if the price is determinable. A contract indexed to a floating spot market doesn’t produce a single number like that. Casela confirms it from the other side: a fixed quantity for a fixed RMB 173,000,000.

So contracted volume doesn’t move with the spot price of a wafer. The reports of InP substrate prices roughly tripling across 2025 and 2026 are describing the uncontracted market. The $98.5 million runs through the first half of 2029 and doesn’t include Lumentum, which was signed after the balance sheet date. Against it, the agreements require $47.7 million of advance payments, $12.7 million of which had come in by June 30. Contract liabilities went from $103 thousand at the end of 2025 to $12.1 million.

What the Coherent exhibit adds

The full Coherent agreement is filed as Exhibit 10.3, and several of its terms never made it into the 10-Q summary or the coverage.

The price is capped by a most-favored-nation clause, which shows up twice in the document:

Products shall be priced no higher than the lesser of (i) the agreed upon pricing set forth in this Section 3 or (ii) the lowest price AXT offers any third party for the same/substantially similar Products.

Wafer pricing “shall in no event increase by more than [***]% on an annual basis.” So increases are capped at a redacted figure, and decreases aren’t.

There’s a right of first refusal on new capacity. If AXT builds capacity beyond the commitment, it has to offer a redacted percentage of it to Coherent first, “on the same terms and conditions (including price, lead time, and volume).”

Export-permit failure is carved out of force majeure by name:

The parties agree that AXT’s failure to obtain export licenses from the Chinese government does not qualify as a Force Majeure Event.

Coherent can also terminate immediately if licences fail and no alternative manufacturing location can be found. Since the geographic revenue table is mostly a licensing readout anyway, AXT is carrying that risk by contract instead of sharing it.

The refund terms are lopsided in both directions. If Coherent misses its minimum order quantity, AXT can terminate on 90 days’ notice and the unused prepayment becomes nonrefundable. If AXT misses its capacity commitment in a given month, it gets 60 days to cure and then has to refund the unapplied balance tied to the shortfall within ten business days.

Coherent owns the specifications: “Coherent owns all right, title and interest to the Specifications and AXT has no right, title, or interest in or to the same.” The licence AXT gets to Coherent’s “Precision Notching” trade secret is revocable and nontransferable.

And the agreement says right in its recitals that AXT’s 6-inch line isn’t in production yet:

Coherent understands and affirms that, as of the Effective Date, AXT’s 6-inch InP program covered by this Agreement is in the development/pilot production phase, and as such agrees there are risks in development of additional capacity in order to achieve the target capacity numbers in Table A. Additional risks include, but not limited to, lower than expected yields, issues in scaling crystal growth capacity, and export permits.

Coherent has moved its Sherman, Texas fab to 6-inch. What it signed with AXT is a development agreement for a substrate line that was still at pilot stage in late June.

What the filings do not disclose

Capacity in wafers doesn’t appear anywhere. Neither the 10-K nor the 10-Q gives a unit figure. Exhibit B to the Coherent agreement, Table A, is the whole capacity commitment schedule, and it’s redacted to [***], same as every per-wafer price in the document, the annual escalation cap, the right-of-first-refusal percentage, and the wafer specifications in Exhibit A. The only capacity figures attached to this company in public are dollar targets given on earnings calls. So any wafer count you see quoted for AXT is derived from revenue and an assumed price, not read off a disclosure.

The Lumentum agreement text isn’t public yet either. The July 29 8-K describes the terms and then says the description “does not purport to be complete and is qualified in its entirety by reference to the full text of the Agreement, which will be filed as an exhibit to the Registrant’s quarterly report on Form 10-Q for the period ended September 30, 2026.” That filing is due in late October, and it’ll show whether Lumentum’s six-year reservation carries the same most-favored-nation and first-refusal terms Coherent negotiated, which together would cap pricing across most of the contracted book.

← All writing