Every optical transceiver in an AI data center starts life as a crystal.
Not a chip. A crystal. Someone grows a boule of indium phosphide, slices it into wafers, polishes them flat enough that a laser structure can be grown on top, and ships them to the epitaxy houses. Those epiwafers become lasers. Those lasers become 800G and 1.6T optical modules. Those modules are the only reason a GPU in one rack can talk to a GPU in another rack fast enough to matter.
The market has spent two years pricing the modules. It has spent about two days pricing the crystal.
On Thursday July 30, after the close, AXT reported a quarter that broke a three year pattern. Revenue of $47.6 million against roughly $34.8 million of consensus, up about 165 percent year on year. Gross profit of $21.4 million. Operating profit of $10.4 million. The stock rose 27 percent that session and traded up another 24 percent after hours.
I want to do three things. Walk through what the quarter said. Then explain the one mechanism that actually determines AXT's reported revenue, which is not demand and which almost nobody modelling this company has right. Then put the price into the context that a two day chart hides.
What the quarter said
Revenue has moved in a step, not a drift. Six quarters of history and the guided September quarter, which the company put at roughly $66 million on the call:
The margin line matters more than the revenue line, and it is the reason this is worth your time at all.
Fifteen months ago this business sold product below cost. Gross margin was negative 6.4 percent. It printed 44.9 percent in the June quarter. That is a fifty point swing in five quarters and none of it came from a price increase. It came from volume moving through a fixed cost base.
A crystal growing operation is close to a fixed cost business. The furnaces run whether they are full or not. The difference between running them at sixty percent and ninety five percent is most of the income statement. Which is why the bottom line flipped so violently:
One correction to the headline before we go further. Diluted EPS was $0.17, but core EPS is closer to $0.11. About a third of pre-tax income is treasury yield on the cash from the April raise, not operations. Interest on a bank balance is real money and it is not the business, and it will not scale with wafer volume.
A second one. Minority interests take 14.6 percent of everything from here. AXT does not own all of Tongmei, the Chinese subsidiary that grows the crystal. When you read consolidated profit, roughly a seventh of it belongs to somebody else.
And a third thing you cannot find in any filing. AXT's filed segment disclosure has exactly two buckets, Substrates and Raw materials and other. The product level split exists only on the earnings call, and it shows how concentrated this company has become:
Two thirds of AXT is now one product line. If you are reading the 10-Q you cannot see that.
The mechanism nobody is modelling
Here is the part that changes how you read every AXT quarter, including this one.
AXT's reported revenue is gated by Chinese export permits, not by customer demand.
On February 4, 2025, China's Ministry of Commerce issued Announcement No.10, placing indium phosphide substrates under dual-use export licensing as control code 3C004.a, alongside trimethylindium, triethylindium and the associated production technology. Every shipment of indium phosphide out of China to a customer outside China now requires a permit. Permits run roughly sixty business days per order.
Watch what that does to the revenue line:
That is not a demand curve. Demand for AI optical interconnect did not fall 39 percent in the fourth quarter of 2025 and then rise 70 percent in the first quarter of 2026. Permits did.
The company said so directly. Fourth quarter revenue came in at $23.0 million against guidance of $27 to $30 million, and the chief executive explained the miss in these words: "we didn't receive as many export permits in Q4 as we had hoped based on the average processing time we had seen up to that point in October."
So when a $47.6 million quarter lands against $34.8 million of consensus, the honest reading is that it is at least as much a permit release event as a demand event. Permits granted inside a quarter convert to revenue inside that quarter. The backlog they unlock is lumpy by construction.
Two consequences follow and they point in opposite directions.
The bearish one. A single administrative decision in Beijing can produce another fourth quarter style air pocket with zero change in end demand. If you own this stock you are underwriting a permit office.
The bullish one, which is subtler and which I think is underappreciated. If revenue is permit-gated rather than demand-gated, then reported revenue is a floor on demand, not a measure of it. The orders exist whether or not the paperwork cleared. And the air pockets are mechanical, which means they are buyable.
One detail that softens the risk. In the first quarter of 2026, 61.5 percent of revenue went to China-destination customers, which requires no export permit at all. Only 38.5 percent was licence-exposed. Management guides a 40 to 60 percent China mix. The permit exposure is real and it is not the whole company, and the mix is a lever management can pull.
Three things about the trade regime that are not in the coverage
One. There is a 50 percent tariff on this product and it was filed under solar.
HTS 3818.00.00 is the classification for doped indium phosphide and doped gallium arsenide substrates. Effective January 1, 2025, the US Trade Representative moved that entire eight digit subheading from the 25 percent Section 301 list to the 50 percent list. The action was framed publicly as a solar measure covering polysilicon and wafers, but it was executed at the eight digit level rather than the statistical suffix level, so compound semiconductor substrates were swept in as collateral scope.
AXT's product is Chinese origin. Any Chinese origin doped indium phosphide or gallium arsenide substrate entering the United States has carried a 50 percent duty since the start of 2025.
That also explains a number which looks alarming without context. North America revenue was $1.708 million in fiscal 2025, down 77.5 percent. AXT states plainly that it cannot forecast when export permits will allow indium phosphide shipments to the United States, and that no gallium arsenide permit to the US has ever been approved. The American market is functionally closed to them right now.
Two. The November 2025 truce did not touch any of this. Announcement No.10 was deliberately left in force. Only Article 2 of the separate Announcement No.46 was suspended, and that suspension runs to November 27, 2026. The indium phosphide regime is legally distinct from the gallium and germanium regime and it has no expiry date.
Three. The 25 percent Section 232 semiconductor tariff does not apply. Effective January 15, 2026, its scope is logic integrated circuits in HTS 8471.50, 8471.80 and 8473.30. A search of the full Federal Register notice returns zero hits for 3818, wafer, substrate, indium or gallium arsenide. Substrates escape it entirely. That is a bear talking point that turns out to be wrong, and knowing which risks are real is most of the work.
The demand is corroborated, and this is the strongest part of the file
If revenue is permit-gated, you need independent evidence that the underlying orders are real. There is some, and it is unusually hard evidence. Customers are prepaying for capacity.
The Lumentum agreement runs to December 31, 2031 and carries a second $43.5 million tranche in 2028.
That is $91.2 million of customer cash committed in a year when the company will do somewhere near $190 million of revenue. Firms do not wire deposits of that size for capacity they are speculating about. Two of the three counterparties are among the largest optical component makers in the world and have every ability to source elsewhere if they thought they could.
This is the single most persuasive item in the bull case and it is more persuasive than the revenue print, because a prepayment is a decision made by a customer with its own money at risk.
Against it, management's stated 2027 exit target is roughly $130 million per quarter of indium phosphide capacity, about $520 million annualised. That number exists only on an earnings call and requires roughly $100 million of 2027 capital spending to actually happen. Treat it as an aspiration with a funding condition attached, not a forecast.
The price, which is the part the two day move hides
Here is where I have to correct something I wrote earlier this week.
The obvious way to describe this stock is that it rose 57 percent in two sessions and is therefore extended. That is what the tape looks like if you start on Wednesday. It is also wrong.
At $58 this stock is 59.4 percent below its May high and still 9.5 percent below the price at which the company itself placed equity in April. The 57 percent move is a bounce off a deep drawdown, not a breakout to new highs. Anyone who bought the April placement is underwater.
That is a materially different setup from the one the two day chart suggests, and it changes the answer to whether you are chasing. You are not buying a stock making new highs. You are buying one that halved and is trying to come back on a genuine change in the numbers.
What it does not change: insiders sold roughly 266,000 shares between $86.73 and $115.24 about six weeks before this print. They were not selling into the blowout. They were selling before it, at much higher prices. That is not a crime and it is not proof of anything. It is a fact worth having.
What $58 is asking for
Enterprise value at $58.16, adjusted properly for the cash and long-term investments that screeners miss, puts AXT around 25.5 times trailing sales. That is above Lumentum at 21.7 and far above Coherent at 7.5.
On the September quarter run rate of $66 million annualised, the same enterprise value is about 12.1 times sales. At the $46.94 close it is 9.3 times.
So the valuation question reduces to one thing. Does the guided quarter happen, and does it hold. At 25 times trailing you are paying for a business that no longer exists. At 12 times forward you are paying a reasonable multiple for something growing triple digits, if the growth is real and if it recurs.
Given that revenue is permit-gated, whether it recurs is not a question about demand. It is a question about paperwork.
What I am doing
I am starting a position of 15,000 shares. That is a starter, not a full weight.
The plan is to accumulate on weakness, and the permit mechanism is precisely why that plan fits this stock. If reported revenue is gated by an administrative process running on a sixty business day cycle, this company will produce quarters that miss badly for reasons that have nothing to do with its customers. Those are the entries. A business whose air pockets are mechanical rather than fundamental is a business you want to buy on the air pockets.
What I will not do is put full size on before the September quarter tells us whether $66 million was a permit release or a new baseline.
What Part 2 has to answer
Does the guided $66 million land, and what was the permit count behind it. Everything else is downstream of that.
The capacity build. The April raise was $632.5 million against a company doing $27 million a quarter at the time. I want the furnace count, the cost per unit of capacity, the qualification timeline, and when incremental capacity actually produces revenue.
Six inch indium phosphide, which is the cost curve. Where is yield running, what is the customer qualification timeline, and is AXT genuinely ahead of Sumitomo and JX or merely first to announce.
The Tongmei listing on the Shanghai STAR Market. A separate Chinese listing of the subsidiary that owns the crystal growing is the largest structural question a US shareholder faces here. How much would be sold, and what is left for us.
The 2028 supply wave. Sumitomo is targeting 3.1 times its 2024 indium phosphide capacity. San'an has a six and a half billion yuan six inch line. SMART Photonics is bringing an Eindhoven fab to full operation. If all of that lands, the pricing window that produced 45 percent gross margins closes. I want the supply and demand curves side by side with dates on them.
Customer concentration. Three counterparties just prepaid $91.2 million. That is excellent evidence of demand and poor evidence of diversification.
Part 2 carries the model. This one is the map.
New position disclosed above, opened around the time of publication. Not investment advice. Every figure is sourced below and dated.
References
AXT, Inc. Announces Second Quarter 2026 Financial Results, 30 July 2026 (Business Wire; SEC 8-K EX-99.1, accession 0001437749-26-025061)
AXT, Inc. Announces First Quarter 2026 Financial Results, 30 April 2026
AXT, Inc. Form 10-K fiscal 2025, SEC CIK 0001051627, accession 0001437749-26-008612, Note 14 geographical information
AXT, Inc. Form 10-Q first quarter 2026, accession 0001437749-26-017054
US Trade Representative, Section 301 Federal Register notice modifying subheadings, effective 1 January 2025, Annex B, HTS 3818.00.00
Federal Register, Section 232 semiconductor proclamation implementation, 20 January 2026, US note 39
MOFCOM and General Administration of Customs Announcement No.10 of 2025, 4 February 2025, control codes 3C004.a through 3C004.c and 3E004
Semiconductor Today, AXT's Q4/2025 revenue constrained by delay in China export permits, 9 March 2026
Semiconductor Today, AXT's revenue grows 17% in Q1 after greater-than-expected export permits, 5 May 2026
Capacity reservation agreements, AXT 8-K filings and press releases, Lumentum, Coherent and Nanjing Casela
Form 4 filings, accessions 0001437749-26-019464 and 0001437749-26-019347
Price and volume data, 29 and 30 July 2026, NASDAQ consolidated tape
Thanks for the great work 💯