Business Overview

AXT, Inc.

NASDAQ: AXTI · Compound semiconductor substrates

7 September 2026

Evidence base: AXT's SEC filings 2016–2026 — ten annual reports through the FY2025 10-K (filed 17 March 2026), the 10-Q for the quarter ended 30 June 2026, the 2026 proxy and the 2025–2026 8-K record — supplemented by independent industry sources (USGS, USITC, MOFCOM announcements, and the filings and disclosures of named competitors and customers). No earnings-call transcripts, shareholder letters or investor-day materials were available; where management's own framing would normally be evidence, that gap is stated.

This document explains how the business works. It is not a valuation and not a recommendation.

1. Executive Snapshot

Figures: FY2025 10-K; 10-Q for the quarter ended 30 June 2026; 8-K filed 8 July 2026.

2. What the Company Does

Some devices cannot be built on silicon. A laser needs a material that emits light efficiently, which silicon does not; a satellite solar cell needs a material that converts a wide slice of the spectrum in a thin stack; a handset power amplifier needs electrons that move faster than silicon allows. Those devices are built on compound semiconductors instead — indium phosphide, gallium arsenide, germanium. Before anyone can build them, someone has to grow a near-perfect single crystal of the compound and slice it into flat, polished discs. That is AXT's business, and only that: the company states it does not design or manufacture chips and does not grow the epitaxial layer (FY2025 10-K).

The unit: one polished substrate wafer

Trace one through. A consolidated Chinese subsidiary refines raw gallium to 6N or 7N purity, or produces the starting material for InP; another, BoYu, makes the pyrolytic boron nitride crucible the melt sits in. Tongmei, the operating subsidiary, loads the charge and grows an ingot by vertical gradient freeze — a slow, controlled solidification whose output is judged by etch pit density, the count of crystal defects per square centimetre. The ingot is sliced, lapped, polished to an epi-ready finish, and shipped against a purchase order. Contracts are short: the company describes the majority as a single performance obligation, “usually less than six months”, priced on a negotiated, formula, list or fixed basis (FY2025 10-K). Revenue is recognised on transfer of control, and cash follows on ordinary terms.

What that unit earns is deliberately opaque. AXT has never disclosed wafer volumes, average selling prices, capacity utilisation, backlog or revenue by end market — the word “backlog” does not appear in the FY2016, FY2020 or FY2025 10-Ks. The only price disclosure is directional and downward: “In certain years, we have experienced an average selling price decline of our substrate selling prices of approximately 5% to 10%” (FY2025 10-K). Consequently every statement in this document about unit economics is an inference from consolidated gross margin, and the volume-versus-price split of recent growth cannot be established from the filings at all (unknown). That is the single largest evidence gap in the company's disclosure.

Two product lines, and what each is for

The substrate group was 67% of revenue in FY2025 and 79% in Q2 2026 (FY2025 10-K; Q2 2026 10-Q). It contains three families with different customers and different fates:

• InP — the growth engine. Supplied in 2″, 3″ and 4″ diameters, with 6″ in pilot production and customer qualification (FY2025 10-K). It ends up in the lasers inside datacentre optical transceivers, in passive optical networks and in coherent telecom optics. Every incremental AI accelerator port is an optical port, and InP is where the light comes from.
• GaAs — the mature leg. Semi-insulating for handset RF front-ends and satellite communications; semi-conducting for VCSELs, LEDs and sensors. Diameters run 1″ to 6″, with 8″ developed and “selling in small quantities”. This is the family blocked from the United States: “no permits for the export of gallium arsenide to the U.S. have yet been approved” (FY2025 10-K).
• Ge — shrinking on purpose. 2″, 4″ and 6″ wafers for triple-junction satellite solar cells and infrared optics. AXT states it “deliberately reduced” germanium wafer sales in 2025 because “raw material costs needed for the germanium wafers increased substantially in 2025 but pricing remained competitive” (FY2025 10-K). This is a clean illustration of the model's limit: when a feedstock is a by-product under export control and the customer will not pay through, the right answer is to sell less.

The raw-material group — 33% of FY2025 revenue, 21% in Q2 2026 — sells purified gallium, boron trioxide, gallium-magnesium alloy and pBN crucibles to third parties. It exists mainly for a different reason. AXT holds positions in over ten Chinese raw-material companies, five consolidated and three carried by the equity method, and states that this “provides us with pricing advantages, reliable supply, market trend visibility and better sourcing lead-times”, adding that it believes it is “the only compound semiconductor substrate supplier to have a position in raw materials” (FY2025 10-K). Because gallium is a by-product of aluminium refining and germanium of zinc and coal processing, neither responds to price in the short run — so owning a slice of the supply is a way to buy priority rather than to buy cheaply.

Two structural facts about that group are worth keeping. First, AXT has been retreating from ownership where it does not control: Beijing JiYa was deconsolidated in March 2019 when AXT's stake fell from 46% to 39%, removing raw gallium revenue outright, and the Emeishan Jia Mei stake was sold down through 2025 and fully exited in December. Second, the joint ventures are managed locally — “the daily operations … are managed by local management and not by us” (FY2025 10-K) — and some sit physically inside partners' plants, so AXT's supply security depends on facilities it does not run.

3. Industry, Competitive Position & Moat

The industry exists because the value of a compound semiconductor device is set almost entirely downstream, while the risk of the whole stack sits upstream. A defective crystal ruins every die built on it, so device makers qualify substrate suppliers slowly and change them reluctantly — but they also refuse to pay much, because the wafer is a small fraction of the finished module. Yole sized the entire compound-substrate market (SiC, GaAs, InP, GaN, Ge combined) at $945m in 2021 growing to $2.3bn by 2027, against an InP bare-die market alone forecast above $5bn in 2026 (Yole Group). The substrate stage gates a value pool many times its own size.

Where the profit sits

The reported margins across the chain are unusually clear about which stages earn. Merchant epitaxy — the stage immediately downstream of substrates — is the worst business in the chain: IQE plc reported FY2024 revenue of £118.0m, adjusted EBITDA of £8.1m (6.9%) and an operating loss of £33.0m (IQE FY2024 results). Substrates do better: JX Advanced Metals' Semiconductor Materials segment earned an 18.0% operating margin on ¥148.0bn in FY2024 (JX Integrated Report 2025). Devices and modules do best of all: Lumentum reported a 32.2% non-GAAP operating margin in its third fiscal quarter of 2026 on revenue up 90% year on year (Lumentum Q3 FY2026 release). The stages that touch the end customer captured the AI cycle; the merchant middle did not.

That matters for AXT because AXT is upstream of the squeezed stage, not inside it. Its economics are set by scarcity of qualified crystal, not by proximity to the customer — which works well when crystal is scarce and poorly when it is not.

Market structure: a genuine oligopoly, with a disputed map

InP substrates are made at scale by a short list: Sumitomo Electric, JX Advanced Metals, AXT's Tongmei, Coherent, Freiberger Compound Materials, and a growing Chinese cohort including Yunnan Germanium and Vital Materials. AXT names its own competitors as “Sumitomo Electric Industries, Japan Energy (JX), Freiberger Compound Materials, Umicore, China Crystal Technology Corp. and Vital Materials” (FY2025 10-K) — essentially the same list it gave in 2016.

Share estimates conflict materially and should not be relied on. JX Advanced Metals self-reports roughly 40% of the global InP substrate market (JX Integrated Report 2025), while Reuters reporting in June 2026 put AXT and Sumitomo at roughly 80% combined with JX near 10%. Both are cited here; neither is corroborated. What is not in dispute is that the qualified supplier set thins sharply at 6-inch — Freiberger caps at 4″ InP, Coherent describes its Sherman, Texas line as “the world's first and largest volume-production 6-inch InP platform”, and AXT is still in qualification at that diameter. The scarce asset in this industry is not wafer capacity; it is qualified capacity at the largest diameter.

Which barriers actually bind

Capital is not the barrier. Vital Materials' approved Zhejiang project commits RMB 2bn for 6m wafers a year of 4″–6″ GaAs and InP, and Sumitomo is spending ¥18bn to take InP capacity to 3.1× its FY2024 level by FY2028 (TrendForce, July 2026). These are two to three orders of magnitude below a leading-edge silicon fab. Three things bind instead.

• Yield learning. Capacity announcements are uniformly multi-year because, as TrendForce puts it, yield improvements take years to achieve. Equipment arrives faster than crystal quality does, which is why the announced expansions all land in 2028–2029 rather than next year.
• Feedstock concentration, which is state-controlled. China produced 839,000kg of the world's 848,000kg of primary gallium in 2024 — 99% — and the United States has had no primary gallium production since 1987 and 100% net import reliance through 2021–2025 (USGS Mineral Commodity Summaries 2026). China holds roughly 60% of germanium production and, per USGS 2024 data, 70% of global indium output. A by-product metal under licence is the hardest input to substitute in this chain.
• Qualification. AXT states that a customer “typically … has at least two qualified substrate suppliers” (FY2025 10-K), which caps the switching cost — a customer can move share between two qualified sources without a new qualification. No independent source states the duration of a qualification cycle (unknown); the behavioural evidence is that buyers signed multi-year contracts rather than switching spot when supply tightened.

AXT's specific position — and the thing that would be hardest to reproduce

Two assets are genuinely difficult to copy, and they point in opposite directions. The first is the vertical position in Chinese raw materials: AXT is, on its own account, the only substrate maker with equity in its feedstock supply, and it has 198 VGF patents plus 130 held by the consolidated raw-material companies. The second is the location of every gram of that production. All AXT manufacturing is in China (FY2025 10-K, Item 2), which since February 2025 has meant that InP, GaAs and Ge substrates all require a Ministry of Commerce export permit.

This is not a conventional moat, because AXT did not build it and cannot control it. China's export-control regime tightened in three steps — gallium and germanium items from 1 August 2023 (MOFCOM Announcement 2023 No. 23), a US-directed prohibition on 3 December 2024 (Announcement 2024 No. 46), and indium phosphide added by customs code on 4 February 2025 (Announcement 2025 No. 10). The named scope covers finished substrates, not merely the feedstock metal: “germanium epitaxial growth substrate” and indium phosphide under code 2853904051 are listed items. The November 2025 suspensions covered Announcement 46's Article 2; the February 2025 indium controls were not suspended.

The effect on prices was immediate and large. Germanium metal moved from $1,392/kg in 2023 to $4,100/kg in 2025, and gallium from $365/kg to $580/kg over the same span (USGS 2026). Reuters reported that average 6-inch InP wafer prices rose 250% to about $5,000 after the February 2025 restrictions. Chinese germanium metal exports fell from 36,656kg in January–September 2023 to 7,520kg in the same period of 2025. For a company whose crystal is grown inside the export-control perimeter and whose customers are outside it, that is both a price umbrella and a permanent political dependency.

What the outside evidence supports, and what it does not

Closing the section: on this evidence, the companies that win here own or recycle their feedstock, sit at the largest qualified diameter, and are close enough to the device to capture the cycle rather than supply it. AXT satisfies the first condition well, is still qualifying on the second, and fails the third by design — it is a pure merchant upstream supplier with no device business. Its current position rests on being a large qualified InP source inside an export-control perimeter at the moment that perimeter became the industry's binding constraint. That is a real advantage and a borrowed one.

4. Growth Engine

AXT has made no acquisitions or divestitures of any size in the period examined, so reported growth is organic growth. The one break in comparability runs the other way: the March 2019 deconsolidation of Beijing JiYa removed raw gallium revenue from the consolidated line, which is why the raw-material product line steps down between the FY2016 and FY2020 filings without any change in the underlying business. Growth since is decomposed by geography and product line, not by deal.

Where the growth actually came from

Revenue rose from $17.97m in Q2 2025 to $47.59m in Q2 2026, an increase of $29.6m or 165%. Almost the whole increase is substrates, and almost the whole geographic increase is in destinations where export permits were issued during the year.

US$ millions. Source: 10-Q for the quarter ended 30 June 2026. Product and geographic splits are separate disaggregations of the same total and should not be added together.

The company attributes the Taiwan, Japan and Europe increases explicitly to “additional export approvals granted by the China government”, and the North America decline to “the impact of China export restrictions” (Q2 2026 10-Q). North America was 8% of revenue in FY2024, about 2% in FY2025, and 0.3% in Q2 2026 — the largest market for AI optical infrastructure is presently the one AXT barely serves.

What cannot be decomposed is the split between volume and price. AXT discloses neither, and industry evidence points to a large price component: Reuters reported 6-inch InP wafer prices up 250% since February 2025, and TrendForce describes an InP supply-demand gap exceeding that in DRAM and NAND, with customers “placing orders further in advance and paying substantial deposits”. A 165% revenue increase built substantially on scarcity pricing is a different proposition from the same increase built on units, and the filings do not permit the distinction (unknown). This is the first question for deeper work.

The drivers, ranked

• 1. Export-permit release. Cyclical / policy-driven. The largest single driver and entirely outside management's control. Tongmei received its first InP permits on 11 June 2025 and further permits in August 2025; the Q2 2026 filing still reports “a backlog of orders for which we have not yet received permits”. Every permit converts existing demand into shipments. It reverses as easily as it arrived.
• 2. InP demand from AI datacentre optics. Structural. Optical port count scales with accelerator deployment, and InP is where the laser light originates — silicon photonics does not remove the InP laser, it packages it. Lumentum's cloud and networking revenue rose 30% between FY2024 and FY2025 and total revenue 90% year on year in its March 2026 quarter.
• 3. Scarcity pricing on InP wafers. Temporary. Prices are elevated because the export perimeter and the qualified 6-inch supplier set are both narrow. Announced capacity from Sumitomo, Coherent, Vital and Yunnan Germanium clusters in 2027–2029, and Fraunhofer has demonstrated 150mm InP-on-GaAs engineered substrates targeted at below 20% of today's InP price.
• 4. Contracted capacity reservations. Management-driven. Three agreements in seven weeks converted spot demand into committed volume with cash up front: Nanjing Casela (RMB 173m for calendar 2027, 50% prepaid, 80% take-or-pay floor), Coherent (three years of 6-inch InP, $22.3m prepayment), and Lumentum (six years, two deposits of $43.5m). Roughly $135m of customer money, with the shortfall risk sitting on the customer.
• 5. Six-inch InP qualification. Structural, and unproven. Six-inch is where demand and pricing are concentrated and where the qualified supplier set is thinnest. AXT is in pilot production and customer qualification; the Coherent agreement is specifically for 6-inch product. Success would move AXT to the diameter frontier; failure leaves it selling into the commoditising part of the market.
• 6. Raw materials and germanium. Structurally weak. Raw materials grew 50% in Q2 2026 against substrates' 233%, and equity income from the unconsolidated gallium ventures fell from $3.4m in FY2024 to $0.8m in FY2025 as Chinese gallium suppliers “reduced their pricing as they competed for market share within their domestic market” (FY2025 10-K). Germanium wafer sales were deliberately cut. This leg does not participate in the AI cycle.

5. Margin, Cash & Capital Allocation

AXT's margin is a utilisation story with almost no ambiguity. The company states plainly that “because many portions of our manufacturing costs are relatively fixed, high utilisation rates are critical to our gross margins”, and attributes the FY2025 collapse to “lower revenue resulting in fixed costs being spread over fewer units and manufacturing variances” (FY2025 10-K). Crystal growth is a slow batch process in an owned building with a trained workforce; the furnaces run whether or not the wafers ship. Cost of revenue also carries the initial cost of the raw material elements and, critically, yield — the company names yield first among its cost factors.

What that produces in practice

The quarterly path through the permit shutdown and reopening is the clearest illustration of operating leverage available in the filings: gross margin ran −6.4% in Q1 2025, 8.0% in Q2 2025, 22.3% in Q3 2025, then 29.6% in Q1 2026 and 44.9% in Q2 2026. Revenue roughly doubled between those endpoints; gross profit went from negative to $21.4m. Cost of revenue rose only 59% while revenue rose 165%, which is the fixed base doing its work in the favourable direction.

Two things sit inside the reported margin that a reader should separate out. First, inventory: the excess and obsolete reserve stood at $32.0m at 30 June 2026 against a carrying value of $96.3m, meaning gross inventory is roughly $128m and about a quarter of it is reserved. That reserve was a critical audit matter in the FY2025 audit. Product sold out of previously reserved inventory carries no cost, so a period of rising demand after a deep reserving cycle flatters gross margin by an amount the filings do not quantify (unknown). Second, earnings: of $15.1m of Q2 2026 pre-tax income, $4.7m was net interest income on the cash from the April equity raise. Roughly a third of the quarter's pre-tax profit was earned by the balance sheet, not the factory.

The financial spine

Sources: FY2024 10-K (FY2022), FY2025 10-K (FY2024, FY2025), 10-Q for the quarter ended 30 June 2026. Comparability: 6M 2026 is a half-year and is not annualised. Operating cash flow has been negative in four of the last five reporting periods including the record-margin half-year, because working capital absorbs the growth. The March 2019 deconsolidation of Beijing JiYa removed raw gallium revenue and makes pre-2019 product-line figures non-comparable with later years.

The row that deserves the most attention is operating cash flow. AXT generated $9.5m of net income in the first half of 2026 and still consumed $0.9m of cash, because receivables rose $9.2m and inventories $12.2m against the ramp. Over FY2023–FY2025 the company produced cumulative operating cash flow of roughly negative $21m while spending $22m on capex. This business has not funded itself from operations in the period examined; it has been funded by Chinese bank loans, by minority investors in Tongmei, and now by public equity.

Where the cash came from and where it went

Ranked by size over the last nine months, capital has arrived far faster than it has been deployed:

• Equity issuance — approximately $700m gross. Two offerings through the same bookrunner: 8.16m shares at $12.25 in December 2025 for $93.9m net, and 9.84m shares at $64.25 in April 2026 for $600.1m net. The second raise took roughly six times the dollars for about 20% more shares. Authorised share count was raised from 70m to 120m at the June 2026 annual meeting; shares outstanding went from 45.4m at end-2024 to 65.6m at 30 June 2026, a 44% increase.
• Customer deposits and prepayments — approximately $135m committed. Lumentum $87m in two deposits, Coherent $22.3m, Casela RMB 173m half-prepaid. This is customer capital funding the same capacity build, with the shortfall obligation on the customer and refund obligations on AXT only if it fails to deliver reserved capacity.
• Chinese bank debt — approximately $99.7m outstanding at 30 June 2026. Short-term facilities across eleven Chinese banks at 2.0–3.8%, partly collateralised by land use rights and buildings at Dingxing and Kazuo and partly by company time deposits. Cheap, short, and rolled continuously.
• Capital expenditure — $8.6m in the first half of 2026, against $6.0m in the whole of FY2025 and $28.5m in FY2022. Capacity investment is only now accelerating; the $700m raised has not yet been spent, and $748.8m of the balance sheet sat in cash, restricted cash and investments at 30 June 2026.
• Returns to shareholders — none. AXT has never paid a common dividend. The 2014 buyback authorisation was last used in 2015; roughly $2.7m remains available and the company states it does not plan to repurchase shares. Accrued unpaid dividends of $2.9m on the Series A preferred block repurchases in any case.

What that ranking reveals is a management team that funds growth by issuing claims rather than by generating cash, and that has been consistent about it for a decade — diluted shares rose from 32.9m in FY2016 to 59.6m in the first half of 2026, an 81% increase, alongside a $49m sale of minority equity in the operating subsidiary in 2020. The December 2025 raise at $12.25 four months before the April raise at $64.25 is the cost of that pattern made explicit.

A transaction that post-dates the reported figures

On 8 July 2026 the Shanghai Stock Exchange accepted Tongmei's withdrawal of its STAR Market IPO application, filed in December 2021 and approved by the exchange in July 2022 but never cleared by the CSRC. The withdrawal triggers a redemption right held by eleven Chinese private equity funds that invested RMB 324.4m — approximately US$49m — for a 7.28% interest in Tongmei; each may require redemption at the original amount invested, without interest. AXT states it has sufficient funds to redeem in full, which the $748.8m liquidity position supports, and intends to pursue a Hong Kong listing for Tongmei instead, emphasising InP rather than the GaAs and micro-LED story of the 2021 application. Separately, on 23 July 2026 AXT agreed to buy roughly 0.58% of Tongmei from a related party controlled by BoYu's general manager for about $6m, taking its stake to approximately 86.09%; that transaction had not closed when the second-quarter financial statements were issued. All of this post-dates the FY2025 10-K and none of it appears in the FY2025 figures above.

6. Cyclicality, Constraints & What to Monitor

This is a high-amplitude cyclical business with a decade of evidence for it. Revenue over FY2016–FY2025 ran 81.3, 98.7, 102.4, 83.3, 95.4, 137.4, 141.1, 75.8, 99.4, 88.3, with gross margin between 12.7% and 36.9%. The FY2019 downturn came from GaAs weakness in automotive and industrial sensors; the FY2023 collapse from the optical and semiconductor inventory correction, when revenue fell 46% in a single year; the FY2025 trough from export permits, which is a policy event rather than a demand event. Three different mechanisms, three deep troughs in seven years.

Where the business sits right now

Above its prior peak on margin and below it on revenue. Q2 2026 gross margin of 44.9% is the highest in the ten-year record, eight points above the FY2022 peak of 36.9%. Quarterly revenue of $47.6m annualises to roughly $190m against the FY2022 record of $141.1m, but the company has only printed one quarter at that level and the second half of the year is unreported. On the company's own framing, demand is not the constraint — “although demand for InP wafers increased in 2025, revenue declined because of the export permit requirements” (FY2025 10-K) — and the Q2 2026 filing still describes a permit-blocked order backlog.

The exposure is concentrated and geographically odd. China is 66% of Q2 2026 revenue and Europe 19%; the United States, the largest single market for AI datacentre infrastructure, is 0.3%. Long-lived assets are $175.8m in China against $0.9m in North America. A US-listed company earns essentially all of its revenue outside North America from assets that are essentially all inside China.

The downside mechanism, as this business would feel it

Three things would have to happen for the current economics to hold: permits keep flowing, InP prices stay elevated, and the 2027–2029 capacity wave does not overshoot. The downside case is that the third arrives while the second unwinds. Sumitomo's 3.1× expansion completes by FY2028, Coherent's Sherman quadrupling by end-2027, Vital's Zhejiang plant in 2029 and Yunnan Germanium's 450,000-wafer line sooner — all landing into a demand curve that will have decelerated from its current rate. For AXT this transmits directly: with a largely fixed cost base and no device business to absorb the volume, a 20% price reversion on substrates removes roughly the whole of the gross profit improvement, and the fixed-cost leverage that produced 44.9% runs in reverse exactly as it did into Q1 2025's −6.4%. The company has been on both sides of that line inside eighteen months.

Durable versus borrowed

Leading indicators, and where they are published

• North America and Taiwan revenue in the quarterly 10-Q. The cleanest read on permit flow. North America returning above 5% of revenue would mean US GaAs and InP permits are being granted; it stayed at 0.3% in Q2 2026.
• MOFCOM announcements and their suspension dates. Announcement 2024 No. 46's Article 2 suspension expires 27 November 2026; the February 2025 indium controls were never suspended. Published by China's Ministry of Commerce.
• USGS Mineral Commodity Summaries, published each January. Gallium and germanium price and trade series — the outside check on whether the input-cost squeeze that forced AXT out of germanium wafers is easing.
• Gross margin against the fixed-cost line, in the quarterly release. Cost of revenue growing materially slower than revenue means volume is filling existing capacity; the two converging means price, not volume, was doing the work.
• Capital expenditure and construction in progress in the 10-Q. The $700m raised has not yet been spent. Capex of $8.6m in the first half of 2026 is not yet consistent with the capacity commitments made to Coherent and Lumentum.
• Customer deposits and contract liabilities on the balance sheet. Accrued liabilities rose from $14.8m to $27.1m in six months. Deposits received are cash today and shipment credits later; tracking the balance separates cash-in from revenue.

7. Risks, Unknowns & Questions for Deeper Work

Cyclicality is covered in Section 6. What follows is what cyclicality does not capture — risks that would compound rather than pass.

• The asset base and the regulator share a jurisdiction. $175.8m of long-lived assets, all manufacturing, 1,048 of 1,072 core employees and the entire crystal-growth capability sit in China, while the customers who set the price are in Europe, Taiwan, Japan and the United States. Beijing does not need to expropriate anything to determine AXT's revenue; it needs only to slow permit issuance, which it has already demonstrated. AXT's shareholders own a claim on cash flows whose transmission out of China runs through export licences, a 10% withholding tax on subsidiary dividends and SAFE approval.
• Engineered substrates attack precisely the stage now earning the rent. Fraunhofer has demonstrated 150mm InP-on-GaAs targeted at below 20% of today's InP price for large volumes. This would leave InP device demand entirely intact while removing the substrate profit pool — the one part of the chain AXT occupies. It is at research stage with no qualification evidence, which makes it a five-to-ten-year risk rather than a two-year one, and therefore easy to underweight.
• The capacity reservations concentrate counterparty and delivery risk simultaneously. Roughly $135m of the funding for this build comes from three customers, two of whom (Coherent and Lumentum) are also potential competitors — Coherent operates its own 6-inch InP platform and Lumentum a captive InP wafer fab, both expanding. The contracts require AXT to deliver reserved capacity; failure to meet the Coherent capacity commitment for six successive months permits Coherent to terminate and reclaim its unapplied prepayment, and the Lumentum deposit is refundable on AXT's material breach. AXT has committed to a capacity build it has not yet made the capital expenditure to deliver.
• Dilution is the funding model, and it has no stated limit. Shares outstanding rose 44% in eighteen months, authorised shares were raised to 120m, the December 2025 raise priced at one-fifth of the April 2026 raise, and the company has never returned capital. A shareholder underwriting the InP build is underwriting it repeatedly and, on the record, at whatever price the window offers.
• Governance is moving toward less shareholder friction, not more. The board carried a Nasdaq audit-committee deficiency from July 2025 to January 2026, cured by re-designating an 85-year-old former director as independent once his consulting payments aged out of the three-year look-back. Having failed to reach a majority quorum in May 2026, the board unilaterally amended the bylaws on 26 July 2026 to cut the quorum to 33⅔%. The board is classified, directors are elected by plurality, and the founder is chairman and CEO. Two genuinely new independents arrived only in mid-2026. Insiders and officers held 5.97% before the April dilution.
• The 2026 proxy does not mention Tongmei. The words “Tongmei” and “STAR Market” appear zero times in the DEF 14A filed 31 March 2026, despite Tongmei being the recipient of all offering proceeds, the counterparty to two of the three supply agreements, and the subject of a US$49m redeemable minority stake whose IPO was withdrawn ten weeks later. The proxy's own CEO pay-ratio figure of $920,841 also fails to reconcile to either year in its Summary Compensation Table.

What the sources could not answer

These are findings, not omissions. Each would materially change the reading of the business.

• Volume versus price in the 165% growth. AXT discloses no wafer volumes, no ASPs, no capacity utilisation and no backlog, in any year examined. Whether Q2 2026 represents units filling capacity or scarcity pricing on similar units is undeterminable from the filings, and it is the difference between a structural re-rating and a policy windfall.
• Substrate margin versus raw-material margin. AXT disclosed this split in FY2016 — “substrate gross margin increased to 34.7% … and raw materials gross margin increased to 23.0%” — and stopped. It has not appeared since. Without it, the consolidated margin cannot be attributed to the growing line or the shrinking one.
• The reserved-inventory contribution to gross margin. A $32.0m reserve against $96.3m of carried inventory is large enough that the share of Q2 2026 gross profit arising from previously written-down material is a first-order question. It is not disclosed.
• Management's own account of any of it. No earnings-call transcripts, shareholder letters or investor-day materials exist in the source set — the folders are empty. Two standard checks cannot be run: what management has said over time, and what they do with their own money, since the Form 4 insider-trading record was also not collected. Both should be pulled from EDGAR before any thesis is formed.
• Whether the April 2026 over-allotment was exercised, and the terms of the Series A preferred, which the proxy does not describe. Minor, but both affect share count.

8. Investor Takeaways

• What this business really is: a Chinese crystal-growth operation with a US listing, selling the base wafer that lasers, RF chips and satellite solar cells are built on, plus equity in the Chinese companies that refine its feedstock.
• The economic engine: a largely fixed cost base against a wafer whose price AXT does not disclose. Utilisation is the whole margin — the same plant produced −6.4% and 44.9% gross margin five quarters apart.
• The main growth lever: InP into AI datacentre optics, gated by Chinese export permits. Demand rose in 2025 while revenue fell, because permits, not orders, were the constraint.
• What could break the story: the 2027–2029 capacity wave arriving as export scarcity unwinds, with cheaper engineered substrates behind it — and a funding model that has met every need by issuing shares.
• What to monitor: North America as a share of revenue, MOFCOM permit policy, capital expenditure against the capacity promised to Coherent and Lumentum, and whether operating cash flow ever turns positive.