AXT’s miss is the wrong signal: InP capacity, not headline revenue, is the trade

The actionable point in this print is that AXT did not need a clean revenue beat to change the debate. What was priced in was a company still trapped by weak volumes and negative mix after Q2 FY2025 revenue of $18.0 million and gross margin of 8.0%. What actually surprised was that revenue of $28.0 million missed the $28.8 million estimate by -2.8%, while EPS of -$0.03 beat the -$0.11 estimate by +72.7%. That separation matters: if investors anchor on the top-line miss, they are valuing AXT as a low-visibility substrate vendor; if they anchor on the earnings surprise, they have to underwrite a business where incremental InP volume can move the loss profile quickly. Gary Fischer’s own bridge framed the quarter around the company’s accounts, not the Street basis: “For Q3 2025, we had a non-GAAP net loss of $1.2 million or $0.03 per share compared with a non-GAAP net loss of $6.4 million or $0.15 per share in the second quarter of 2025.” That wording earns attention because it commits to the scale of operating repair, not just the existence of a rebound.

That distinction between what was priced and what mattered becomes clearer in the financial trajectory. Revenue had been volatile rather than structurally compounding, with Q1 FY2025 at $19.4 million and Q2 FY2025 at $18.0 million before the Q3 rebound to $28.0 million. The Street expected a larger snapback, but the earnings model reacted better than the revenue line because gross margin recovered to 22.3% from 8.0%. This is the print’s variant perception: AXT’s Q3 was not a clean demand-beat story, but it was a better-than-modeled mix and absorption story. A top-line miss normally weakens conviction in a recovery; here, the simultaneous EPS beat says the revenue composition mattered more than the absolute dollar shortfall.

The margin recovery explains why the revenue miss should not be read mechanically. Q3 gross margin of 22.3% was still below Q3 FY2024’s 24.0%, but the more relevant comp is Q2 FY2025’s 8.0%, because the company had just emerged from a volume and utilization trough. The Q3 operating line followed the same pattern: non-GAAP operating loss improved to $384,000 from $6.1 million, while GAAP operating loss improved to $1.1 million from $6.7 million. Those are not revenue-only effects; they show fixed-cost leverage and expense discipline working at a revenue level that was still only $28.0 million. The risk to the thesis is also embedded here: if gross margin cannot hold near 22.3% when Q4 revenue normalizes into the company’s guide, then Q3 was licensing and mix, not a restored earnings base.

The product mix makes that risk more measurable than usual. Fischer said Q3 indium phosphide was $13.1 million, primarily from data center and PON applications, while gallium arsenide was $7.5 million. That means the quarter was not a broad substrate recovery led equally by all categories; it was an InP-led reset, with germanium substrates down to $640,000 and Morris Young adding that germanium sales declined by about $1 million. The Street’s revenue miss is therefore less informative than whether InP sustains a higher run rate. Timothy Bettles sharpened the constraint in Q&A: “But we estimate that current capacity is around about $20 million a quarter for indium phosphide with our current run rate and current capacity that we've got.” That sentence changes the frame from demand proof to capacity utilization: with Q3 InP at $13.1 million, the debate is how fast the company fills the available InP ceiling, not whether InP is still an optionality story.

That capacity framing is why the Q4 guide is more important than the Q3 revenue miss. Management guided Q4 revenue to $27 million to $30 million, above the Q3 Street-comparison actual of $28.0 million at the midpoint range’s center of gravity, but the quote carried a caveat that should not be ignored. Fischer said AXT had “the capability to achieve revenue in the range of $27 million to $30 million in Q4, subject to the caveats I just mentioned.” The word “capability” is doing work: management is not presenting Q4 as riskless backlog conversion, and the caveat language leaves room for shipment timing. Still, the range itself says the company does not view Q3 as an isolated spike. If Q4 lands inside that band while gross margin stays near Q3’s 22.3%, the market will have to treat Q3 as a base-forming quarter rather than a catch-up event.

The expense guide complicates that bullish read, but it does not break it. AXT cut non-GAAP operating expense to $6.7 million in Q3 from $7.6 million in Q2, while GAAP operating expense was $7.3 million against $8.2 million. For Q4, management expects OpEx to increase to approximately $9 million due to end-of-year adjustments and a return to a more normalized level. That means the Q3 EPS beat benefited from both gross margin recovery and temporarily lower expense intensity. The thesis, therefore, cannot be that Q4 EPS automatically improves again; it has to be that revenue scale and InP mix can absorb a higher OpEx base. If investors buy the stock only because EPS beat by +72.7%, they may be disappointed by Q4 operating expense. If they buy it because InP has moved from recovery to capacity utilization, the approximately $9 million OpEx guide becomes the hurdle to monitor rather than a reason to dismiss the print.

Working capital is the other place where the quarter looks worse than the P&L, and it is the cleanest counterargument to the thesis. Cash and cash equivalents and investments decreased by $3.9 million to $31.2 million, despite the operating loss narrowing sharply. Fischer attributed the cash movement to receivables, saying accounts receivable increased by $11 million, while inventory decreased by approximately $2.4 million to $77.7 million. That mix is not fatal, but it does change the quality of the earnings beat. The P&L showed the model responding to volume; the balance sheet showed that revenue conversion has not yet translated into cash. For portfolio managers, this is where position sizing matters: the print supports a more constructive view on earnings power, but confirmation requires receivables to unwind rather than remain the shadow cost of the rebound.

The customer read-through is narrower, and more valuable, than a generic optical-cycle call. AXT’s InP revenue of $13.1 million was primarily tied to data center and PON applications, which makes the quarter directly relevant to Coherent and Lumentum as customers using InP substrates for compound semi wafers and EML laser fabrication. The magnitude is not the whole $28.0 million quarter; it is the $13.1 million InP line and the current capacity around about $20 million a quarter for InP. WIN Semiconductors and IQE also sit in the InP and GaAs substrate customer set, but the read-through is more mixed because GaAs was $7.5 million and did not drive the same capacity discussion. On the supplier side, Korea Zinc’s refined indium exposure is levered to whether AXT pushes InP toward that around about $20 million quarterly capacity, not to germanium, where revenue was $640,000 and sales declined by about $1 million. The second-order implication is that AXT’s Q3 is supportive for InP-linked optical supply chains, but it is not evidence of a full compound-substrate upcycle across every material.

That same selectivity matters in peer context. In the latest substrates peer table, AXTI shows $26.9 million of revenue, gross margin of 29.6%, and revenue YoY of +39.1%, which puts it near the upper end of growth versus larger Japanese materials names whose latest revenue bases dwarf AXT’s. Rogers, the only USD peer in the table, reported $200.5 million of revenue with gross margin of 32.2% and revenue YoY of +5.2%. The comparison argues against treating AXT as merely a subscale laggard; on the most recent peer snapshot, AXTI’s revenue base is far smaller, but its growth rate is the differentiated attribute. The caveat is that peer gross margin at Rogers is 32.2%, while AXT’s Q3 event margin was 22.3%, so AXT still has to prove that growth can become margin parity rather than only recovery from a trough.

The call delivery supports that cautious constructive view rather than an all-clear. In the tone history, Q3 FY2025 sentiment was 0.19 versus Q2 FY2025 at 0.22, while guidance_tone slipped to 0.28 from 0.32. That is not the tone of a management team declaring victory, even though the financials improved. Prepared_sentiment rose to 0.26 from 0.22, but qa_sentiment improved to only 0.15 from 0.03, which suggests the scripted narrative was more confident than the interactive discussion. The conflict is visible in the uncertainty index too: Q3 FY2025 uncertainty was 81.6, higher than Q2 FY2025 at 72.4. The numbers say management had better facts to discuss, but not yet lower uncertainty around conversion, timing, and sustainability.

That tonal split is useful because it prevents the thesis from becoming promotional. The call sounded more specific where it mattered, especially on InP and Q4 revenue capability, but the model still carries execution friction. Top customers generated approximately 45.2% of total revenue, and 2 customers were over the 10% level. Asia Pacific represented 87% of revenue, with Europe at 12% and North America at 1%. Those concentrations are not automatically negative, but they mean a small set of customers and geographies can decide whether Q4 reaches the $27 million to $30 million range. If the InP capacity story is right, customer concentration accelerates the earnings recovery. If any of those large customers pause shipments, concentration turns a capacity story back into a timing story.

The reason to lean constructive despite those frictions is that the surprise mix is exactly what should precede a re-rating in a small-cap substrate recovery. The market was looking for $28.8 million of revenue and got $28.0 million, so the headline screen missed. But the market was also looking for -$0.11 of EPS and got -$0.03, which is a more important signal for a company that had printed -$0.20 in Q1 FY2025 and -$0.16 in Q2 FY2025. The operating model has now shown it can approach breakeven at Q3 revenue levels, provided InP remains the mix anchor and OpEx does not rise faster than the Q4 normalization already flagged. This is why the print should be bought selectively on weakness caused by the revenue miss, not chased blindly on the EPS beat. The mispricing is not that AXT suddenly became a clean compound semi growth story; it is that the Street may still be valuing Q3 as a one-time rebound when the guide and InP capacity disclosure imply a more durable utilization path.

What to watch next is therefore precise. For Q4 FY2025, the first test is whether revenue lands in the $27 million to $30 million guide; a result below that range would break the capacity-utilization thesis because management explicitly framed it as achievable capability. The second test is gross margin versus Q3’s 22.3%; holding near that level with OpEx increasing to approximately $9 million would confirm that Q3 was not just temporary expense compression. The third test is InP against the $13.1 million Q3 level and the current capacity around about $20 million a quarter; progress toward that capacity would be the cleanest confirmation for Coherent, Lumentum, WIN Semiconductors, IQE, and Korea Zinc. Finally, monitor cash and receivables after cash fell by $3.9 million to $31.2 million and accounts receivable increased by $11 million; if working capital does not begin to normalize next quarter, the earnings recovery will deserve a lower multiple even if the revenue guide is met.