AXT’s miss is the wrong focal point: permits, China data-center InP, and a $128.4 million cash reset change the risk/reward

The point of this print is not that Q4 was good. It was not, relative to expectations. Revenue was $23.0 million against the Street’s $25.2 million, a -8.5% surprise, and EPS was -$0.05 versus the -$0.04 estimate, off a near-zero estimate base where the percent surprise is not meaningful. What was priced in was a cleaner sequential handoff from Q3’s $28.0 million revenue and 22.3% gross margin, with losses narrowing as utilization recovered. What actually surprised was the opposite in the reported quarter, because revenue fell to $23.0 million, gross margin slipped to 20.9%, and the company’s own GAAP EPS basis was -$0.08. The variant perception is that the miss is a backward-looking artifact of shipment timing and export-permit friction, while the forward data points in the call and the next reported quarter show the substrate cycle re-accelerating: Q1 FY2026 revenue was $26.9 million, gross margin was 29.6%, revenue grew +16.9% QoQ and +39.1% YoY, and diluted EPS improved to -$0.03.

That distinction matters because the selloff case rests on treating Q4’s $23.0 million revenue as a demand signal, while the call’s evidence points to constrained conversion rather than weak end demand. Gary Fischer framed the shipment gate explicitly when he said, “As of today, we have approximately $26 million in revenue that can be realized in Q1 across our substrate product lines and raw materials, for which we either have -- already have a permit to ship or for which an export permit is not required.” The wording matters because it ties Q1 to realizable revenue, not only orders, and it separates permitted shipments from permit-exempt shipments. That is a different risk profile from an ordinary backlog comment. It also explains why Q4’s $23.0 million did not deserve to be annualized: the company had already identified approximately $26 million that could be realized in Q1, and the quarterly history subsequently shows $26.9 million revenue in Q1 FY2026.

The financial trajectory therefore argues against valuing AXT on Q4 trough optics, because the revenue and margin series shows the business reacting sharply to utilization rather than grinding sideways. The company’s revenue fell from $25.1 million in Q4 FY2024 to $19.4 million in Q1 FY2025 and $18.0 million in Q2 FY2025, while gross margin went from 17.6% to -6.4% and then 8.0%. That is the damage case investors remember. But the recovery from that trough is already in the numbers: Q3 FY2025 revenue jumped to $28.0 million with 22.3% gross margin, Q4 FY2025 revenue slipped to $23.0 million with 20.9% gross margin, and Q1 FY2026 revenue rebounded to $26.9 million with 29.6% gross margin. The margin signal is the cleanest part of the print: at $26.9 million of quarterly revenue, gross margin reached 29.6%, which is above every quarter in the provided history except none, and materially above Q4’s 20.9%. If investors were expecting a structurally impaired substrate model after Q1 FY2025’s -6.4% gross margin, Q1 FY2026’s 29.6% is the number that breaks that assumption.

The reason Q4 still matters is that it tells us which parts of the model have not yet normalized, and that is where the EPS disappointment came from. On the company’s own reported basis, Gary Fischer said Q4 revenue was “$23.0 million compared with $28.0 million in the third quarter of 2025 and $25.1 million in the fourth quarter of 2024,” and that revenue base was not enough to absorb expenses. Non-GAAP operating expense was $7.8 million in Q4 versus $6.5 million in Q3 FY2025, while GAAP operating expense was $8.8 million versus $7.3 million in Q3. That is why non-GAAP operating loss widened to $2.6 million from $384,000, and GAAP operating loss widened to $3.8 million from $1.1 million. The surprise was not merely lower revenue; it was lower revenue coinciding with higher expense, a combination that turned a near-breakeven Q3 into a larger Q4 loss. But the comparison to Q4 FY2024 is also relevant: non-GAAP operating loss improved from $5.4 million to $2.6 million, GAAP operating loss improved from $6.2 million to $3.8 million, and non-GAAP operating expense fell from $9.8 million to $7.8 million. The company is still subscale, but the fixed-cost structure is not deteriorating versus last year.

That operating leverage is why management’s year-end capacity commentary deserves more weight than the Q4 miss, but only if investors discipline it against the numbers already printed. The company’s product mix in Q4 had indium phosphide at $8.0 million, gallium arsenide at $7.0 million, germanium substrates at $231,000, and consolidated raw material joint venture revenue at $7.6 million. InP is the strategic line because the call tied it directly to data-center applications, and Morris Young gave the strongest demand statement on the call: “Our revenue related to the data center market in China are expected to grow by more than 60% in Q1 over Q4, highlighting both increased investment in these Tier 1 data centers as well as the strong desire for Chinese domestic suppliers to secure local stores at every level of the AI infrastructure supply chain.” The key phrase is not the usual AI-supply-chain framing; it is “more than 60% in Q1 over Q4,” because that is a near-term conversion claim tied to a specific geography and end market. If Q4 InP revenue was $8.0 million, the data-center portion of that mix accelerating by more than 60% in Q1 is the piece that can move consolidated margin, since Q1’s 29.6% gross margin already shows better absorption.

The balance sheet reset gives management time to execute that ramp without forcing investors to underwrite a near-term liquidity event. Cash, cash equivalents and investments increased by $97.2 million to $128.4 million as of December 31, compared with $31.2 million at September 30. Accounts receivable decreased by $2.6 million in the quarter, while net inventory increased by approximately $4 million to $81.7 million. The market may treat the inventory build as a warning because Q4 revenue missed by -8.5%, and that concern is not baseless: inventory rising into a revenue shortfall can signal demand mismatch. But here the conflicting numbers cut the other way more than usual. Backlog moved from $49 million to above $60 million according to Richard Shannon’s question on the call, Q1 realizable revenue was approximately $26 million, and Q1 actually reported $26.9 million. Inventory at $81.7 million is therefore better read as ramp preparation and permit-buffering than as a pure demand warning, unless the next quarter fails to convert backlog into revenue.

The second-order read-through is most actionable for optical and compound-semiconductor names exposed to InP and GaAs procurement, because AXT’s order visibility implies customers are extending planning horizons rather than buying only spot wafers. Coherent is a direct customer for InP substrates and compound semi wafers, and Lumentum is a customer for InP substrates for EML laser fabrication; for both Coherent and Lumentum, the AXT signal is that China data-center-related InP demand is expected to grow by more than 60% in Q1 over Q4, while AXT’s backlog was described as moving from $49 million to above $60 million. WIN Semiconductors and IQE, both customers for InP and GaAs substrates, get the same substrate availability read-through from AXT’s $8.0 million InP and $7.0 million GaAs Q4 revenue mix, but without a customer concentration problem at AXT because the top 5 customers generated approximately 22.6% of total revenue and no customer exceeded the 10% level. Korea Zinc, as refined indium metal supplier for InP substrates, is the upstream beneficiary if AXT’s InP-led demand translates into purchasing, with the magnitude anchored to AXT’s $8.0 million Q4 InP revenue and management’s more than 60% Q1 China data-center growth expectation.

Relative to substrate peers, the market’s likely mistake is anchoring on AXT’s small revenue base instead of the growth and margin combination that now screens better than the recent Q4 headline. The peers table shows AXT’s latest reported quarter at $26.9 million revenue, 29.6% gross margin, and +39.1% revenue YoY. Rogers is much larger at $200.5 million revenue, with 32.2% gross margin and +5.2% revenue YoY, while 3445.T shows 32.3% gross margin and +8.7% revenue YoY. AXT is not comparable in scale to Rogers, and that scale gap is a real valuation constraint, but the operating question for a substrate small-cap is whether revenue growth can carry gross margin toward peer-like territory. AXT’s 29.6% latest gross margin is already close to Rogers’ 32.2% and 3445.T’s 32.3%, while its +39.1% revenue YoY is above Rogers’ +5.2% and 3445.T’s +8.7%. That comparative point supports the thesis that AXT’s problem is not demand quality, but confidence in permitting, conversion, and sustained utilization.

The call delivery backs that interpretation, though it also tells us not to ignore execution risk. The tone history moved in the right direction where it matters most: Q1 FY2026 sentiment was 0.32 versus 0.22 in Q4 FY2025, guidance_tone rose to 0.57 from 0.34, prepared_sentiment rose to 0.42 from 0.23, and qa_sentiment improved to 0.22 from 0.04. That is a meaningful change in management’s forward posture, not just a more upbeat script. But the same table shows tone_confidence slipped to 0.38 from 0.39 and uncertainty rose to 86.5 from 73.6, with qa_evasiveness moving to -9.4 from -74.6. The conflict is straightforward: management sounded materially more positive on guidance, but uncertainty also increased. That matches the business reality. The company has line of sight to permits, backlog, and capacity, but export authorization and customer timing still govern revenue conversion.

That tension is clearest in the year-end target, where management put a large number on the table without removing the permit caveat. Morris Young said capacity is expected “to increase it to about $35 million a quarter by the end of the year,” and an analyst immediately tested whether that meant possible shipments of $30 million, $35 million in orders if capacity and permits align. The capacity target matters because it creates a bridge from Q4’s $23.0 million revenue and Q1’s $26.9 million revenue to a higher quarterly run-rate. The margin target is even more important: Gary Fischer said, “I think getting somewhere at $40 million a quarter in aggregate, not just indium phosphide, but we should be getting hopefully somewhere close to 35% gross margin.” The word “hopefully” is the hedge, and investors should respect it. But the numerical relationship is compelling: latest revenue was $26.9 million with 29.6% gross margin, and management is pointing to roughly $40 million aggregate quarterly revenue for gross margin somewhere close to 35%. The upside case is not multiple expansion on hope; it is a utilization case where incremental revenue narrows losses because fixed costs are already visible around Q4 GAAP operating expense of $8.8 million and Q1 expected OpEx of approximately $9.0 million.

The bear case is that AXT has given investors a moving target before, and Q4 itself is the evidence. Revenue fell -17.6% QoQ from Q3 FY2025 to Q4 FY2025, and diluted EPS moved from -$0.04 to -$0.08 on the company’s GAAP history, even though gross margin only moved from 22.3% to 20.9%. That tells us the model is not yet self-funding at low-$20-million quarterly revenue, especially when OpEx moves from $7.3 million GAAP in Q3 to $8.8 million GAAP in Q4. The customer concentration data also cuts both ways. No customer above the 10% level lowers single-customer risk, but the top 5 at approximately 22.6% of revenue means the ramp must be broad enough across customers, products, and permits rather than solved by one hyperscale-style buyer. And while Q1 FY2026 revenue of $26.9 million validates the near-term guide, it is still below Q3 FY2025 revenue of $28.0 million, so investors should not declare a linear recovery until the company exceeds that Q3 level while keeping gross margin near Q1’s 29.6%.

The actionable conclusion is to look through the Q4 miss, but not to give management credit for the full year-end ramp until conversion shows up in revenue and margin together. What was priced in was a Q4 beat or at least a clean hold near the Street’s $25.2 million revenue estimate and -$0.04 EPS estimate. What surprised was a $23.0 million revenue print, a -8.5% top-line shortfall, and a larger loss, offset by Q1 realizable revenue of approximately $26 million, backlog moving from $49 million to above $60 million, and a balance sheet rising to $128.4 million of cash, cash equivalents and investments. The market may be missing that AXT’s latest reported $26.9 million revenue and 29.6% gross margin already validate the first step of management’s argument. The remaining debate is whether the company can turn backlog and permits into a $35 million quarterly revenue run-rate by the end of the year, and whether roughly $40 million aggregate quarterly revenue can carry gross margin somewhere close to 35%.

What to watch next is brutally concrete. For the next quarter, the thesis is confirmed if revenue holds above the approximately $26 million realizable-revenue marker cited on the call and does not fall back toward Q4’s $23.0 million, while gross margin stays near Q1 FY2026’s 29.6% rather than Q4’s 20.9%. It is strengthened if backlog remains above $60 million after shipments, because that would show the Q1 conversion did not drain the pipeline, and if Q1 China data-center revenue actually reflects the more than 60% growth over Q4 that Morris Young flagged. It is weakened if OpEx around approximately $9.0 million in Q1 combines with revenue below $26 million, because that would re-open the Q4 loss dynamic, or if inventory above $81.7 million rises again without revenue growth. By the end-of-year checkpoint, the stock needs evidence that capacity is approaching about $35 million a quarter and that gross margin is tracking toward somewhere close to 35% at roughly $40 million aggregate quarterly revenue; failure to exceed the prior $28.0 million quarterly revenue peak before then would break the operating-leverage thesis.