AXT: a real shortage, priced near my Bull case

Yahoo Finance gives AXT a P/E ratio of 1,995. The number is arithmetically correct and tells you nothing, because trailing earnings of $0.04 a share still contain three quarters of the old, loss-making business. What matters is the third-quarter guide: at least $66 million of revenue and non-GAAP earnings of $0.30 to $0.32 a share for one quarter. At $79.79 the shares trade at about 43 times my Base earnings for 2027, and the company has to fill its own capacity targets to earn that.

At the October 7, 2026 close the shares trade 49.9% above my triangulated Base value of $53, a downside of 33.3%. The Bear value of $23 is 70.6% below the price and the Bull value of $85 is 6.6% above it, so the market is paying close to my best case. Confidence is grade D, which carries a required margin of safety (MOS) of 60%. That puts the maximum buy price at $21, and the shares would have to fall about 73% to reach it. The shortage is real. The price is a bet that it lasts.

Route: SF-15 Semiconductors and ecosystem, subtype FAB/IDM-like materials (a manufacturer of compound semiconductor wafer substrates), with SF-13 industrial manufacturing as a cross-check for utilization and capital spending. Economic classification: Quality Growth (5/12, at the boundary with Standard). Primary method: growth-adjusted EV/EBITDA on normalized 2027 EBITDA. Secondary method: ten-year free cash flow model. Cross-check: forward P/E. AXT's fiscal year is the calendar year. Values are present fair value estimates, not 12 month price targets.

What AXT is

AXT grows crystals and slices them into wafers, the base layer on which other companies build lasers and light detectors. Indium phosphide (InP) wafers go into the lasers and detectors inside AI data center optical transceivers. Gallium arsenide (GaAs) wafers serve vertical-cavity lasers (VCSELs), robotics and radio chips. In the second quarter InP revenue was $30.7 million, GaAs $6.6 million, germanium $0.3 million, and a group of consolidated raw material joint ventures added $10.0 million.

The operating base is in China. Of $176.7 million of long-lived assets, $175.8 million sit there, and China took $30.8 million of the $47.6 million of second-quarter revenue, 64.8%. Management said China was above 50% of revenue in the quarter and expects a 40% to 60% range. Exports from China need permits, which is why the third-quarter guide is worded as revenue that already has a permit or needs none.

The ownership structure matters for valuation. AXT's main operating subsidiary, Tongmei, has outside investors, and noncontrolling interests in Tongmei total about 14.5%. Those holders took 14.6% of second-quarter net income ($1.9 million of $13.0 million). I treat that as a claim on the operating business. Tongmei withdrew its Shanghai STAR Market listing application (the exchange accepted the withdrawal in July), which created redemption rights on about $49 million invested by private equity funds in 2021. Management says those investors currently want to stay in, and that AXT has the cash to redeem them in full.

The latest quarter

The inflection is genuine and management names the cause: more volume, a richer mix toward indium phosphide and, on the call, larger wafer diameters and tight supply. Gross margin went from 8.0% to 44.9% in a year. Management also said further gains depend on volume, product mix and fixed cost absorption, which is the polite way of saying that margin follows utilization in both directions. Q2 earnings per share of $0.19 non-GAAP was against a consensus of $0.07.

Why trailing earnings and headline cash flow cannot be used

Trailing earnings blend a loss-making 2025 with one very good quarter, and the company's quarterly profit includes interest on the $632.5 million it raised in April. Cash flow tells a different story from the income statement. Operating cash flow in the first half was -$0.9 million and free cash flow (operating cash flow minus capital spending) was -$9.5 million. That figure was helped by $12.0 million of customer advances, which rose from $0.1 million at the end of 2025. Excluding the advances and deducting $1.8 million of stock-based compensation (SBC, a real cost paid in shares), the hard figure is -$23.3 million. Another $6.4 million of equipment was paid for with notes receivable, a non-cash item.

Inventory rose to $96.3 million from $81.7 million at the end of 2025 and accounts receivable to $36.7 million from $26.8 million, so growth is absorbing cash before it produces any. I did not reopen the 2021 to 2025 filings for cash flow history. The reported revenue history is $137 million in 2021, $141 million in 2022, $76 million in 2023, $99 million in 2024 and $88 million in 2025, which shows how short the profitable record is.

My normalized 2027 scenarios are EBITDA (earnings before interest, taxes, depreciation and amortization) of $106 million in the Bear case, $162 million in the Base and $201 million in the Bull. They rest on revenue of $330 million, $415 million and $470 million. A straight line between the capacity targets management gave on the call, roughly $60 million of quarterly InP revenue capacity at the end of 2026 and about $130 million at the end of 2027, plus about $20 million a quarter from other lines, implies about $495 million of 2027 revenue if every wafer is made and sold. The Bull case is 95% of that path, the Base 84% and the Bear 67%. That path is my estimate, not company guidance. For comparison, a third-party aggregator shows 2026 revenue of about $218 million, close to my $217 million (first half $74.5 million, third quarter $66 million, fourth quarter about $77 million).

Balance sheet and shares

The balance sheet changed in April. AXT sold 9,844,357 shares at $64.25, a gross $632.5 million, and bought $305.6 million of Treasury, agency and corporate securities. Cash, restricted cash and investments were $748.8 million on June 30. A screener that shows only cash and equivalents ($412.2 million) leaves out $336.7 million.

The July capacity reservation agreement brings two further customer deposits of $43.5 million each, applied as shipment credits against future wafer purchases. A deposit is cash in and a delivery obligation out, so I treat it as neutral and leave it out of net cash. Together with the $47.7 million of advance payments under the long-term supply agreements ($12.7 million received by June 30), it does fund part of the capacity build. The 10-Q puts remaining performance obligations under those agreements at $98.5 million through the first half of 2029.

The share count rose 18.5% in six months, from 55.337 million at December 31 to 65.570 million at June 30, and 65.573 million on August 3. For per share values I use 66.5 million economic shares, the company's own estimate for the third quarter. As a check, 65.573 million common shares plus 0.437 million from options (468,000 at an average $5.30, net of the treasury method at $79.79) plus 0.371 million unvested performance shares gives 66.38 million. I assume the 1.4 million unvested restricted stock awards are already in the outstanding count, and if they are not the count is 2% higher. The Series A preferred is non-convertible and non-voting, so it enters as a claim, not as shares.

Primary valuation: growth-adjusted EV/EBITDA

Enterprise value (EV) is market value plus debt minus cash. I apply a multiple to normalized 2027 EBITDA, subtract the 14.5% minority share of the operating business and add net cash of $628.7 million. The multiples are assumptions, not outputs of a peer table I could not build: 14x in the Bear case, 24x in the Base and 32x in the Bull. The Base multiple works out to about 16x my 2028 Base EBITDA, inside the 12x to 18x tier the framework uses for quality manufacturers, which fits growth that is still above 40% a year. The grid below tests it.

Each value is EBITDA times the multiple, times 85.5% for the minority adjustment, plus $628.7 million of net cash, divided by 66.5 million shares. The primary Base is $59.

Sensitivity: value per share

Only the corner cell, EBITDA of about $200 million at 28x, reaches today's price. That is the Bull EBITDA at a multiple close to the Bull multiple.

Secondary valuation: ten-year free cash flow model

The secondary method does not use a multiple. It discounts ten years of free cash flow from 2027 to 2036, deducts the minority share, adds the same net cash and divides by the same shares. Free cash flow is operating profit after 14% tax (China's rate is 15%, and the second-quarter effective rate was 13.9%), plus depreciation, minus capital spending, minus working capital at 25% of each revenue increase, minus SBC. Capital spending is $100 million in 2027, the plan an analyst cited on the call and management confirmed, then $90 million, $80 million and $70 million, and 6% of revenue after that. Interest income is left out because the cash is already in the bridge.

The secondary Base is $39. The Base case still has revenue growing from $217 million to $1.15 billion by 2036, and free cash flow in 2027 is slightly negative (-$12 million) because the capacity build and working capital take more than the profit. A beta of 1.9 would justify a higher discount rate than 11%, so the Base is not harsh on that point.

Sensitivity of the Base model: value per share

No cell is above $45. The model values the cash flow, not the story, and that is where the gap to the price comes from.

Cross-check and growth diagnostics

Two checks are not available. A historical multiple comparison would need profitable years, and AXT lost money in 2023, 2024 and 2025, so the history gets no weight. A peer comparison needs synchronized multiples for the few other indium phosphide suppliers, which I could not source. I use a forward P/E check instead: 2027 Base EPS (earnings per share) is $1.86 (operating profit plus about $21 million of net interest income, 14% tax, 14.5% minority share, 66.5 million shares), and I apply 20x in the Bear, 32x in the Base and 40x in the Bull. That gives $25, $59 and $91. For reference a third-party aggregator shows 2027 EPS consensus near $2.15, 16% above my Base EPS and close to my Bull EPS of $2.28. It is one source, it moved a lot over the summer, and I do not use it in the Base.

The growth valuation cross-check (SM-19) is diagnostic only. The price is 42.9 times my 2027 Base EPS (37.1 times the $2.15 consensus), and my Base EPS grows 42.5% in 2028, a PEG of about 1.0 that supports nothing on its own. Growth durability: structural demand with execution risk and a strong rebound element, since 2025 was a loss. Cyclical distortion: high, because gross margin sits at a shortage level and the comparison base is a trough. GAAP and non-GAAP are close ($0.17 against $0.19 in the second quarter) because SBC is only $1.8 million a half year. The framework does not allow SM-19 to set fair value.

The free cash flow growth quality check (SM-20) says the same from the cash side. Free cash flow in the first half was -$9.5 million, so a compound growth rate is not meaningful and I use a path to free cash flow instead. In my Base, free cash flow goes from -$12 million in 2027 to $71 million in 2028, $137 million in 2029 and $184 million in 2030. On 66.5 million shares that is -$0.18, $1.07, $2.06 and $2.76 a share. Against the economic market value of $5.31 billion the yield is 1.3% in 2028, 2.6% in 2029 and 3.5% in 2030. Distortion flags: low base high, cyclical rebound high, capital spending timing high, working capital high, dilution medium. Estimate reliability: low, because the guide is permit-dependent and there is almost no consensus history. Classification: structural with execution risk. No score is given, because a diagnostic score on a negative starting point would mislead.

Valuation triangulation

The primary and secondary Base values differ by 42.0%, above the 30% threshold, so the framework flags a valuation divergence. The consequences are a Confidence grade no higher than B (it is D anyway) and 5 extra points of required margin of safety. The gap is not noise. The two multiple-based methods price AXT as a business that keeps growing for years, and the cash flow model charges it for the capital spending and working capital that growth needs. I weight the multiple method highest because it is the market's language, and the cash flow model at 30% because it is the only method that sees capital spending. Using the primary method alone gives $59 and using the cash flow model alone gives $39, and neither reaches $79.79. The unrounded values are $23.46, $53.23 and $85.09.

What is the market already pricing in?

At $79.79 and 66.5 million economic shares, market value is about $5.31 billion. After removing $628.7 million of net cash and adding back the 14.5% minority share, the operating business is valued at about $5.47 billion. That is 33.8 times my 2027 Base EBITDA, 22.8 times my 2028 Base EBITDA and 20.7 times the annualized third-quarter guide of $264 million. At my Base multiple of 24x, the price requires EBITDA of about $228 million in 2027, 41% above my Base and 13% above my Bull. At the Bull multiple of 32x it needs about $171 million, 5.6% above my Base and below my Bull.

In cash flow terms, the price equals about 2.4 times my entire Base free cash flow path at an 11% discount rate. Alternatively, it equals my Base path at a discount rate of about 6.4%, or my Bull path at about 9%, which the model values at $81. A buyer today needs five things to be true: indium phosphide demand keeps outrunning supply for years, AXT fills a capacity target that doubles twice, export permits keep flowing, prices and margins hold near today's levels as competitors add capacity, and the new capacity turns into free cash flow after the 2027 spending. The market is not pricing a good outcome. It is pricing something close to my best case.

Bear, base and bull scenarios

I assign no probabilities to the scenarios. The Bear is not a collapse in demand: revenue stays above $330 million, but profit per wafer falls toward the levels AXT earned before 2026. The Bear and Bull are far apart, a factor of 3.6, and that spread is itself a reason for low Confidence. Only the Bull value is above the price, and it is above it by 6.6%.

How the buy price is set

A Confidence D score carries a base MOS of 50% or more, and I start at 50%. I add 5 points for export permits and geopolitical exposure, which is a binary risk to shipments, 5 points for cyclicality and operating leverage, and 5 points for the valuation divergence, and subtract 5 points for the net cash balance sheet. The required MOS is 60%, the top of what the framework allows before a stock becomes a plain no-buy. Applied to the $53 Base, the maximum buy price is $21, and a stronger 70% margin marks a strong-buy reference near $16. There is no customer concentration add-on, because no customer exceeds 10% of revenue. The shares would need to fall about 73% from $79.79 to reach $21.

Dividend and capital return

AXT pays no common dividend, so the income check does not apply. The Series A preferred stock accrues a 5% cumulative dividend, which is $2.9 million unpaid and included in the $8.3 million preferred claim above. There is no active buyback and the company says it does not plan one.

Reasons to own AXTI

• A genuine bottleneck. Management says demand outpaces supply however fast capacity is added, backlog is above $100 million and orders are being rationed to what can be built.
• Visible contracts and cash. Long-term supply agreements carry $98.5 million of remaining obligations through the first half of 2029, and customers are paying deposits to reserve capacity.
• Operating leverage is proven in one quarter. Gross margin went from 8.0% to 44.9% in a year, and the third-quarter guide implies EPS of $0.30 to $0.32.
• A funded balance sheet. About $629 million of net cash after the April raise covers the 2027 capital spending plan several times over.
• Low customer concentration. The top five customers are 30% of revenue and none exceeds 10%.
• Technology headroom. Management says its 6-inch indium phosphide capability is progressing and larger diameters support margins.

What could go wrong?

• Thesis breaker: export permits and restrictions. China needs a permit for shipments abroad, and the CFO said AXT cannot predict the timing or success of permits for any order. The guide of $66 million is only the revenue that already has a permit or needs none.
• Pricing and margin normalization. Gross margin of 45% is a shortage margin. Management says it is roughly tripling its own capacity in 2026, and capacity added by other suppliers can bring prices down.
• China concentration. About 65% of second-quarter revenue shipped to China, nearly all long-lived assets are there, the company holds its China business through partly owned subsidiaries, and pledged cash backs local bank loans.
• Cash conversion. First-half free cash flow was -$9.5 million, inventory is $96.3 million, and 2027 capital spending of about $100 million comes before any proof of free cash flow.
• Dilution. Shares rose 18.5% in six months, authorized shares were raised from 70 million to 120 million in June, and the company has just shown it will sell stock at a high price.
• Tongmei minority holders. Redemption rights on about $49 million exist after the withdrawn listing, and a Hong Kong listing is expected to take about a year.
• Litigation. A securities class action covering March 2021 to April 2024 was dismissed with leave to amend on July 7, 2026, and a third amended complaint was filed on August 4.

Management execution

Dr. Morris Young is chief executive and Gary Fischer chief financial officer. In 2026 they raised $632.5 million at $64.25 a share, brought InP capacity forward faster than first planned, signed long-term supply agreements with prepayments, and guided far above consensus. The offsets are a record of losses from 2023 to 2025, a securities lawsuit that is still alive, a Shanghai listing that was withdrawn after a long process and a Hong Kong attempt that will take another year. Management also declines to give the backlog number, saying the order book depends on what it chooses to accept, which makes the most-quoted demand metric a management decision. Assessment: 11/20, adequate to good. The next test is whether margins and cash conversion hold as the capacity comes online.

Stock Analyza scorecard

A score of 5 is the lowest Quality Growth score, one point above Standard, and I say so openly. Compounder status is withheld because return on invested capital has not been above the cost of capital through a cycle, free cash flow is negative and the revenue record from 2021 to 2025 is flat to down. The reinvestment runway gets 1 of 2 because capacity is pre-sold, but I cannot measure the return on the new capacity yet. Business quality is 60/100: moat and pricing power 13/20, return versus cost of capital 6/20, balance sheet 13/15, earnings and cash quality 6/15, growth and runway 14/15, management and capital allocation 6/10, dilution and governance 2/5. The valuation score is 11/100: discount to Base 0/40, protection against the Bear case 0/20, agreement between methods 3/15, historical valuation 2/10, reverse valuation 3/10, data and model quality 3/5.

Confidence score

The filing data is solid, although the quarter is unaudited. The grade is held down by one quarter of profit, permit-dependent revenue, a 42% gap between the two main methods and a Bear to Bull spread of 3.6 times. It scores confidence in the valuation, not the quality of the company.

What I would watch from here

Today's readings are green on guided revenue ($66 million), non-GAAP gross margin (45.0%, exactly on the line), economic shares (about 66.5 million), inventory (up 6.8% in the quarter against revenue up 76.8%) and customer concentration. They are yellow on free cash flow (-$9.5 million in the first half), on China at 64.8% of revenue and on the Tongmei redemption right. Nothing is red today. The next review point is the third-quarter report, scheduled for October 29, 2026.

The three most important thesis breakers

1. Export permits or new restrictions cut shipments of indium phosphide wafers.
2. Indium phosphide prices and gross margin fall as industry capacity arrives, taking gross margin below 38%.
3. The capacity build absorbs the cash: free cash flow stays negative after 2027 spending, or AXT sells more stock to fund it.

Adversarial review

I recomputed the net cash bridge, share bridge, scenario values, both sensitivity grids, the triangulation, the margin of safety and the yields independently, and checked reported inputs against the June 30 10-Q, the second-quarter release and call, and the market price. The table shows where each input comes from and how far it can be trusted.

The weakest points of the model are the multiples (assumptions with no peer table behind them), the working capital and capital spending assumptions, and the permit-dependent revenue path. The customer deposits are treated as neutral timing, and if they are more than that the 2027 cash flow is somewhat better than my Base shows. The 14.5% minority deduction cuts both ways: it lowers value per share, but without it the Base primary and secondary values would be $68 and $44.

The strongest counter-thesis, for the bulls: indium phosphide is a multi-year bottleneck in AI optics, AXT becomes by far the largest producer by 2027, 6-inch wafers lift margins, and revenue reaches $1.5 billion or more with gross margin near 50%. At a 9% discount rate my Bull path gives $81, so this case is not absurd. The response is that it needs every item on the list in the section above to hold at once, and today's price already requires it, so the Base case offers no margin of safety.

Audit flags: the Base needs a capacity path that I derived from two numbers given on a call, and it does not include the deposits or the $98.5 million of obligations as separate cash flows. The first-half operating cash flow could be distorted by the timing of customer advances. The consensus figures come from third-party datasets that moved sharply in the last two months. I could not source peer multiples, so the primary multiple is labeled an assumption and tested. The most generous reasonable combination in the table, the Bull path at a 9% discount rate, only reaches $81. The publication gate status is PASS WITH WARNING because the two main methods diverge by more than 30%, the revenue path depends on export permits, and the quarterly data is unaudited.

Final verdict: reduce at $79.79

AXT is a small crystal grower that sits on one of the tightest bottlenecks in AI optics, and the second quarter shows what that is worth. The shares, though, trade at 33.8 times my 2027 Base EBITDA and just under my Bull value, and the cash flow has not turned positive yet. In April the company sold 9.84 million new shares at $64.25. A buyer today pays 24% more than that, for the same business with one more quarter of results.

Verdict: REDUCE at $79.79. Triangulated Base value $53, downside 33.3%, Confidence D (9/25). Required margin of safety 60%, maximum buy price $21.

The main risk to this verdict is that the shortage lasts much longer than I assume, so that capacity fills as planned, prices hold and revenue passes $1.5 billion with margins intact. New money belongs on the watchlist until the price approaches $21, or until the third-quarter report on October 29 shows free cash flow turning positive and margins holding.

Sources

Primary sources include AXT's Form 10-Q for the quarter ended June 30, 2026, the second-quarter 2026 release of July 30, 2026 and the call transcript. The October 7, 2026 close of $79.79 comes from Yahoo Finance. Consensus estimates come from public aggregators (StockAnalysis, MarketBeat) and are labeled as estimates.

The reference market price is the October 7, 2026 regular session close at 4:00 PM EDT. Balance sheet and cash flow figures are as of June 30, 2026, and the common share count is as of August 3, 2026. The economic share count, normalized EBITDA, scenario values, capacity path and stress tests are analytical estimates, not company guidance.