AXT Inc

AXT is the Western merchant InP substrate bottleneck at the exact moment AI-datacenter optical interconnect demand inflects, and the Q2 print is the first hard evidence the thesis converts to economics: revenue +164% YoY, gross margin from 8% to 45% in four quarters, and a swing to real GAAP profitability — all funded by a fortress ~$665M net-cash balance sheet.

• Margin proof, not margin promise. Non-GAAP gross margin went 8.2% (Q2 2025) to 29.9% (Q1 2026) to 45.0% (Q2 2026). That is operating leverage on a fixed-cost substrate line, and management is guiding toward a 50%+ handle as volume and larger-diameter wafers mix in.
• The demand is contracted, not hoped for. Customer prepayments of $22.3M (Casella) and $25.4M (Coherent) plus a new long-term Lumentum agreement mean end customers are funding AXT's capacity to lock supply — the strongest possible signal in a bottleneck market.
• Capacity roadmap is capital-light in 2026. Reaching ~$60M/quarter exiting 2026 needs only ~$14M of capex because the gains come from productivity and yield, not new tools. That is a very high incremental return on invested capital in the near term.
• Balance sheet risk is essentially gone. The April 2026 raise took cash + investments to ~$749M against $84.2M of short-term loans. The 2027 $100M capex program is self-funded many times over; there is no financing-risk overhang on the ramp.
• Q3 guidance is built on permitted volume only. Guiding $66M on the basis of export permits already in hand (with additional permits as unbudgeted upside) makes the near-term number unusually derisked for a company in this situation.
• Backlog above $100M against a ~$47.6M quarter means the constraint is supply and permits, not demand — the favorable side of a shortage.

A ~$4.4B market cap on a company that did $47.6M of quarterly revenue and only just crossed into profitability implies flawless execution of a 2.7x capacity ramp — while the single largest determinant of whether shipments happen at all, Chinese export permits, is outside management's control and explicitly un-forecastable by its own CEO.

• Export-permit dependency is the whole story and it is unmanageable. Management states plainly it 'cannot predict the future timing of permits or success in obtaining them for specific orders.' A $100M+ backlog that cannot be converted on a schedule is not equivalent to revenue.
• Valuation leaves no room. ~$4.38B market cap on ~$125.5M TTM revenue is ~35x TTM sales, and ~17x even the annualized Q3 guide ($66M x 4 = $264M). The multiple already discounts the 2027 $130M/quarter capacity case being fully sold at high margin.
• Geographic concentration cuts both ways. Management expects China to be 40-60% of future revenue, and all three manufacturing sites are in China. The same geography that supplies the growth supplies the export-control, environmental-regulation, and geopolitical risk.
• Cash generation is unproven. The Q2 release contains no cash-flow statement, so the FCF profile through the ramp is unverified — and 2027 capex steps up ~7x to ~$100M while working capital (inventories already $96.3M, receivables $36.7M) absorbs cash as revenue scales.
• Dilution is real and recent. Weighted diluted shares went from 43.7M (Q2 2025) to 63.5M (Q2 2026) and the Q3 guide assumes ~66.5M — the per-share claim on the ramp has shrunk by roughly a third in a year.
• The margin gain is partly a mix and utilization artifact. A 45% gross margin at $47.6M of revenue reflects a favorable InP-heavy quarter; a permit delay that shifts mix back toward GaAs/industrial would reverse a large slice of it quickly, as FY2025's 13.1% trough showed.
• Volatility itself is a risk to the position: the stock traded to a $143.16 intraday high in late May 2026 and closed at $68.67 on 2026-08-03 — a roughly 50% drawdown inside ten weeks on no change to the underlying thesis.