AXTI

Every high-speed optical transceiver that carries data between AI servers starts with a substrate wafer. The laser diodes inside these transceivers are grown epitaxially on indium phosphide (InP) wafers — thin discs of crystalline InP that provide the precise atomic lattice on which III-V semiconductor layers are deposited. Without InP wafers, you cannot make the lasers that drive optical networking. AXT makes those wafers.

The company operates from facilities in Fremont, California and Beijing, China (subsidiary Tongmei), manufacturing three types of compound semiconductor substrates: InP (the AI optical play), GaAs (used in RF amplifiers, LEDs, and solar cells), and germanium (used in satellite solar cells and fiber optic components). InP is the strategic product — it's the material platform for every high-speed optical transceiver laser, including the 800G and upcoming 1.6T modules being deployed in AI data centers. Q4 2025 revenue mix: InP $8M, GaAs $7M, Ge $231K, raw materials JV $7.6M — InP is ~35% of substrate revenue today but the dominant growth vector.

AXT has already added 25% more InP capacity since Q4 2025 and is on track to double capacity from Q4 2025 levels by year-end 2026, targeting an InP run rate of ~$35M/quarter. Phase 1 capex is ~$30M using brownfield space at Tongmei (existing clean room, building, power). Management is already planning a second doubling in 2027 requiring $100–150M for greenfield construction. The $93.9M net equity raise funds Phase 1; Phase 2 may be funded by a planned Tongmei IPO on Shanghai's STAR Market. AXT's vertical integration — subsidiary JinMei now refines high-quality indium, and subsidiary BoYu makes PBN crucibles for InP crystal growth — secures critical upstream supply.

The near-term constraint is China's MOFCOM export permit process for InP and GaAs shipments. Permits are required to ship substrates from Tongmei's China facilities to customers outside China. The process started at ~60 business days in Q3 2025 but has become highly variable and non-transparent, with geopolitics a likely factor. In Q4 2025, AXT received a couple of outright denials for the first time — but with instructions to resubmit with additional information, which management interprets as fixable rather than permanent blocks. Q1 2026 floor guidance is $26M in revenue already permitted or not requiring permits, with "significant upside" possible — management noted finished goods are staged in clean rooms ready to ship the moment permits arrive.

The fundamental story is compelling but the valuation is extreme. AXT generated $88.3M in FY2025 revenue with a $21.3M net loss, yet the stock has surged ~6,800% over the past twelve months to a market cap of ~$4.4B. This is ~50x trailing revenue for a loss-making company with $23M quarterly sales. The market is pricing in a dramatic scaling of InP demand. The transcript reveals the demand signal is real — InP backlog has reached $60M+ (up from $49M), customers are placing LTAs with forecasts beyond 2030, and customer demand "increases every week" per management — but execution depends on permit flow and capacity ramp.