Bull & Bear Memo

AXT, Inc.

company
AXT, Inc. (Nasdaq: AXTI)
sources used
FY2016–FY2025 10-Ks (chiefly FY2025, filed 2026-03-17); 10-Qs through Q2 FY2026 (filed 2026-08-13); 8-Ks of 2026-06-17 (Casela), 2026-07-02 (Coherent), 2026-07-08 (Tongmei STAR withdrawal), 2026-07-22; proxies through 2026-03-31.
latest period
Q2 2026 (quarter ended June 30, 2026)
missing or stale data
No earnings-call transcripts, shareholder letters, investor presentations or Form 4 data in the archive, so there is no management-voice evidence; the Q2 2026 earnings press release (Exhibit 99.1) was not captured, so no company guidance. AXT discloses one operating segment and two product lines only — there is no InP-specific revenue, no unit volumes, no ASPs and no utilization rate anywhere in the filings. Substrates vs. raw materials is the finest revenue cut available.
Section 01

Business in one line

AXT grows single-crystal ingots of indium phosphide, gallium arsenide and germanium in China using proprietary vertical gradient freeze technology, slices them into wafer substrates, and sells them to epitaxial-layer houses that sell on to chip fabs. Consolidated Chinese subsidiaries also make purified gallium, pBN crucibles and InP base material, consuming some internally and selling the rest. In FY2025 substrates were 67% of $88.3 million of revenue, raw materials 33%.

Type · Cyclical  policy-gated upcycle

For a cyclical the bear case lives in the cycle and in what gets bought at the top, not in a competitor taking share.

Sections 02 & 03

Bull case vs. Bear case

Paired view puts each assumption beside the risk that attacks it. Click any card to expand.

Peter Lynch pitchThe simple reason this stock could work is that AI data centers move data with light, light needs indium phosphide lasers, and AXT is one of a handful of firms that can grow low-defect InP crystals — and after eighteen months locked behind China's export-permit gate, the permits started flowing and the business went from losing money to a 44.9% gross margin in five quarters.

Munger invertThe most likely way I lose money is that the 44.9% margin quarter turns out to be the peak of a policy-driven catch-up — in a business whose revenue gate is held by a government with a stated interest in building the domestic industry its customers compete with.

Why the market might be missing it: AXT reports one operating segment and never discloses InP revenue, volumes or pricing, so the mix shift was invisible in the reported lines until it arrived all at once in consolidated gross margin — and the commitments underwriting 2027 sit in 8-Ks, not the P&L. But the stock was sold at $12.25 in December 2025 and $64.25 four months later, so the argument is not that the market missed the cycle. It is about how long the margin lasts and what the capital earns.

Section 04

Signals to monitor

Each signal tracks a named assumption or risk. A signal that tracks nothing in this memo is noise.

Section 05

External challenge notes

Omitted — external search was not requested for this run. Ask for “external challenge” to add competitor, regulatory, litigation and short-seller findings here, kept separate from company-source evidence.

Company filings used 10 · annual
  • FY2025 10-K, filed 2026-03-17 — business, competition, export controls, FY2023–25 P&L, liquidity
  • FY2024 10-K, filed 2025-03-14
  • FY2016–FY2023 10-Ks — cycle history, margin range, restructuring precedent
Interim reports 31 · quarterly
  • Q2 FY2026 10-Q, filed 2026-08-13 — latest period; revenue, margin, balance sheet, Coherent and Tongmei disclosure
  • Q1 FY2026 10-Q, filed 2026-05-14
  • Q3 FY2025 10-Q, filed 2025-11-13 — quarterly revenue bridge
Material events 8-K
  • 2026-06-17 — Casela long-term InP supply agreement, RMB 173.0m for calendar 2027
  • 2026-07-02 — Coherent Master Development and Supply Agreement, 6-inch InP, $22,288,500 prepayment
  • 2026-07-08 — Tongmei STAR Market withdrawal accepted; redemption right triggered
  • 2026-07-22 — Board expanded to six; Jia-Bin Duh elected
Not available in the source set gaps
  • Earnings-call transcripts — folder empty
  • Shareholder letters, investor-day and conference presentations — folders empty
  • Insider trading (Form 4) — CSVs header-only
  • Q2 2026 earnings press release (Exhibit 99.1) — not captured, so no management guidance
  • InP-specific revenue, unit volumes, ASPs, utilization — never disclosed by the company
Section 06

Bottom line

Why it could work. AI optical interconnect demand is pulling hard on indium phosphide, AXT is one of few qualified low-defect InP suppliers, and it now holds $748.8 million and two multi-year customer commitments with which to turn that demand into capacity.
What must go right. The Q2 volume step-up proves to be end demand rather than a cleared permit backlog, gross margin holds well above FY2025's 12.7% as capacity ramps, and the $694 million raised earns a return in a business that lost money in each of the last three years.
How it breaks. China's MOFCOM — under a Foreign Trade Law that explicitly contemplates restricting exports to build domestic industries — throttles InP permits by destination as Chinese customers scale, capping revenue at whatever is permitted and stranding capacity built for export.
What would change my mind. North American revenue: $161 thousand last quarter, so a return toward the 8% it represented in FY2024 would mean the gate is genuinely open, while four more quarters near zero alongside a rising China mix would mean this is a Chinese domestic business with a US listing.