AXT: A Real Indium Phosphide Bottleneck, Priced Well Ahead of the Business

Company history and financial trajectory

Vertical history

The most useful way to reconstruct AXT is to start with structure, not branding. The company today describes itself as a materials science business that develops high-performance semiconductor wafer substrates, while keeping sales, administration, and customer support in Fremont, California. The manufacturing base, however, sits in China through Tongmei and a set of related subsidiaries and joint ventures. Tongmei itself was founded in 1998, and by 2020–2021 AXT had reorganized much of its Chinese structure underneath Tongmei in preparation for a planned STAR Market listing in Shanghai. That is the hinge point in the company's modern history: AXT shifted from a simple U.S.-listed designer with outsourced Chinese production into a layered cross-border structure designed to combine U.S. listing access with China manufacturing, China capital, and eventually, if approvals arrive, a separate China public-market minority base at Tongmei.

The restructuring mechanics matter because they explain both the present governance discount and the present operating advantage. In late 2020, AXT reorganized its China entities so that JinMei and BoYu and their subsidiaries were assigned to Tongmei, with minority holders in those businesses converted into Tongmei minority stakes. Private-equity investors also injected about $49 million into Tongmei, ultimately receiving a 7.28 percent redeemable noncontrolling interest once approvals were complete in January 2021. By March 31, 2026, Tongmei's noncontrolling interests and redeemable noncontrolling interests still totaled about 14.5 percent, and AXT remained the controlling stakeholder with a majority of Tongmei board seats. In plain language, AXTI shareholders do not own 100 percent of the business engine the stock is now being valued on. They own control over it, but not all of it.

That 2020–2021 reorganization left a second long tail: the STAR listing has still not happened. Tongmei submitted its IPO application to the Shanghai Stock Exchange in December 2021; the application was accepted in January 2022; the Shanghai Stock Exchange approved it in July 2022; the CSRC accepted it for review in August 2022; and it remains in process. The redeemable noncontrolling investors have rights to force AXT to redeem their Tongmei shares at original purchase price if the IPO fails required approval steps or is canceled. AXT's 2025 Form 10-K gives the aggregate redemption amount as about $49 million. That is manageable after the April 2026 raise, but it shows the STAR process is embedded in the capital structure, not merely a blue-sky option.

From there, AXT's last several years break into four clear stages. The first was the pre-reorganization period: a niche compound-substrate producer using China manufacturing and a raw-material network to compete on cost and supply resilience rather than on corporate simplicity. The second was the 2020–2022 expansion phase: management pushed Tongmei toward the STAR market, cleaned up the China structure, and benefited from stronger demand and healthier margins. Revenue reached $137.4 million in 2021 and $141.1 million in 2022; gross margin moved from 34.5 percent to 36.9 percent; GAAP net income attributable to AXT rose from $14.6 million to $15.8 million. That was the last period when AXT looked like a healthy but still ordinary specialty-materials company.

The third stage was the breakdown period in 2023–2025. The demand environment softened, AI optical demand had not yet become the dominant revenue engine, and Chinese export controls on gallium and germanium started to hit the operating model. China announced export controls on gallium and germanium in July 2023, effective August 1, 2023. Then, according to China's 2025 dual-use import and export licensing catalog, indium phosphide substrates as a category also moved under export-permit administration. AXT's own 2025 quarter-to-quarter record shows the practical effect. Revenue fell from $99.4 million in 2024 to $88.3 million in 2025; GAAP gross margin collapsed from 24.0 percent to 12.7 percent; GAAP net loss attributable to AXT widened from $11.6 million to $21.3 million. This was a cyclical downturn interacting with a policy chokepoint, not simply a normal semiconductor downcycle.

The fourth stage began in late 2025 and accelerated in 2026. Here the same permit regime that suppressed revenue also enhanced AXT's scarcity story. Investors started to pay for the idea that if even constrained shipments were improving, then unconstrained demand must be much larger. That logic gained force as Q3 2025 rebounded, as management discussed longer customer lead times and around-60-business-day permit windows, as Q4 2025 and Q1 2026 commentary kept pointing to InP as the growth engine, and as the company said backlog had moved from about $60 million to more than $100 million in a single quarter. Whatever one makes of the valuation, that was the period in which AXT stopped being a lagging compound-substrate name and became an AI optical bottleneck trade.

Financial vertical review

The financial record shows both why the market got interested and why caution is still necessary. The revenue and earnings table below captures the three big swings: profitable strength in 2021–2022, sharp deterioration in 2023–2025, and a visible but still incomplete recovery in early 2026. The important feature is the volatility of gross margin, not simply the level of sales. When the mix is favorable and capacity is utilized, AXT can look decent. When permits delay higher-value shipments or factories run under-absorbed, the P&L degrades quickly. That is classic operating leverage, but with a policy variable embedded inside it.

Amounts in USD millions except margins. Q1 2026 figures are quarterly, not annual. Operating cash flow and capex for 2023–2025 come from the 2025 Form 10-K; Q1 2026 operating cash flow comes from the Q1 2026 Form 10-Q.

What the table says in business terms is simple. AXT was a profitable mid-sized substrate supplier when demand and mix were healthy. Then it became a capital consumer. Operating cash flow was positive in 2023 only because working-capital movements and non-cash items helped offset losses; by 2024 and 2025, operating cash flow was negative $12.1 million and negative $12.8 million, and capex continued at a moderate but real level. In Q1 2026 alone, operating cash flow was negative $11.7 million as inventories and other operating assets built ahead of expected shipments. That is exactly what one would expect from a company with a real backlog that cannot yet convert cleanly to revenue. The accounting statement says demand exists. The cash-flow statement says conversion is still messy.

Balance-sheet analysis is unusual because it shifted abruptly from tight liquidity to excess capital. At March 31, 2026, before the April follow-on closed, AXT had $57.9 million of liquidity including cash and restricted cash, against $68.9 million of short-term loans and $6.8 million of long-term loans. That is not distress, but it is not a fortress either. After the April 2026 offering, the prospectus showed 65.4 million shares outstanding on a fully exercised basis, and the company raised roughly $632.5 million gross. The business therefore moved in one step from "permit-constrained small-cap with working-capital pressure" to "permit-constrained small-cap with an enormous cash buffer." That sharply reduces solvency risk. It does not solve return-on-capital risk.

Price and valuation history

AXTI's stock history over the last year is really three histories piled on top of each other. The first was the depressed micro-cap phase that culminated in the $1.92 close on 2025-08-01. The second was the discovery phase, when investors started to connect AXT's InP line to AI optical demand and when the stock moved from single digits to the low-$70s by late April 2026. The third was the speculative-blowoff-and-reset phase: from $71.07 on 2026-04-29 to $140.83 on 2026-05-22, then back to $57.21 by 2026-07-10. Moves of that size over that span nearly always mix fundamentals with momentum. AXT is no exception.

A rough market-cap reconstruction makes the scale even more obvious. AXT had about 45.4 million common shares outstanding at the end of 2024, 55.6 million at March 31, 2026 before the April offering settled, and 65.4 million on a fully diluted post-offering basis in the April prospectus. Applying the August 2025 low price to a roughly mid-40-million share base implies an equity value below $100 million. Applying the July 10, 2026 close to the post-offering share count gives about $3.74 billion. Even allowing for dilution, this has been an approximate 40-fold market-cap re-rating in under a year. That is far too large to describe as a routine "multiple expansion." It is a narrative regime change.

The market's valuation label for AXT has therefore changed twice. In 2023–2024 AXT often traded like a cyclical laggard with China risk and weak profitability. In late 2025 and early 2026 it traded like a turnaround. By May 2026 it was trading like a scarcity asset attached to AI infrastructure, with the market willing to ignore the fact that the company remained loss-making on a trailing basis. As of July 2026, the label is still "AI optical bottleneck," just with less euphoria than at the peak. That matters because valuation centers can shift permanently when business quality changes, but they can also overshoot when market preference changes faster than fundamentals. In AXT's case, both things happened at once.

Business model and moat

Revenue structure and operating mechanics

AXT reports as one segment, but economically the business has two moving parts. The first is specialty substrates, which represented 67 percent of 2025 revenue and almost $19.3 million of Q1 2026 revenue. Inside that bucket sit InP, GaAs, and Ge. The second is raw materials and other products, which represented 33 percent of 2025 revenue and $7.6 million in Q1 2026. The raw-material layer is functional, not decorative, and it is one reason management keeps arguing that AXT is structurally better placed than peers to expand InP output. The company has built and invested around its critical materials rather than simply buying them on the open market and hoping they stay available.

That supply-chain architecture is distinctive. As of March 31, 2026, the Q1 2026 10-Q showed AXT's consolidated China holdings included Nanjing JinMei, ChaoYang JinMei, and Beijing BoYu at 85.5 percent effective ownership via Tongmei, plus ChaoYang XinMei at 58.5 percent and ChaoYang ShuoMei at 75 percent. It also held minority stakes accounted for under the equity method in JiYa, Xiaoyi XingAn, and ChaoYang KaiMei. The point is not to memorize the names. The point is that AXT has spent years trying to internalize and secure parts of its material chain. That lowers some supply risk and cost risk. It increases structure risk and governance complexity.

Source: AXT Q1 2026 Form 10-Q note on privately held raw-material companies and Tongmei ownership.

On cost structure, AXT is a classic materials manufacturer with meaningful fixed-cost absorption. That is why permits matter so much. When higher-value substrate shipments are delayed, fixed labor, facilities, depreciation, and process overhead do not disappear. They simply spread over less revenue and over a weaker mix. When shipments recover, gross margin snaps back quickly. That is exactly what happened from negative 6.4 percent gross margin in Q1 2025 to 29.6 percent in Q1 2026. The company therefore does have operating leverage. It just does not control the release valve. MOFCOM does.

What the moat really is

AXT's real moat is narrower than management's rhetoric, but it is real. The first part is process know-how in crystal growth and specialty-substrate manufacturing. Management says AXT designs and builds its own crystal-growth furnaces and is using an existing footprint plus brownfield repurposing to scale InP faster than peers. For a small company in a specialized materials niche, that matters: it means the company is more than a merchant reseller of someone else's engineered capacity. It owns part of the production learning curve.

The second part is customer qualification and application fit. Substrates are upstream materials, but they are not commodity ingots in the simple sense. Optical-data and laser applications demand consistency, defect control, and reliability. Management's emphasis on low-EPD wafers, 6-inch development, and nearly all leading optical customers suggests that AXT has already crossed a qualification threshold that newcomers cannot replicate overnight. If customers are indeed placing longer-term orders and giving more visibility because permits take time, that also implies they are not treating AXT as instantly replaceable.

The third part is the integrated supply chain. JinMei's move into refining high-purity indium is particularly important because indium is a direct ingredient in InP substrates. In the same Q1 2026 call, management called the raw-material layer a "crown jewel" of the growth strategy. That sounds promotional, but the substance is sounder than the phrase. In a market where China controls a dominant share of gallium production and where disruption in gallium and germanium exports has already had measurable economic implications, upstream access is part of the product itself, not a cosmetic edge.

The part that is not a moat is regulation. Investors sometimes talk as though the permit regime itself creates scarcity that benefits AXT. It does create scarcity. That choke point sits outside AXT's control, and it can block the company's ability to monetize the demand it already has. A true moat strengthens the company when conditions worsen. This variable can weaken the company while strengthening the stock narrative. Those are not the same thing.

Management and governance

Management credibility on operations has improved. Over the last several quarters, management's commentary around permits being the gating factor proved directionally right, and the company's Q1 2026 outcome plus Q2 2026 order visibility backed the claim that demand had materially improved. The issue is not whether management fabricated the demand story. The issue is whether management can spend a balance sheet transformed by more than $700 million of gross equity issuance since late 2025 in a way that produces acceptable returns for common shareholders. On that point, the record is not available yet. The funds are new.

Governance needs a discount. The company has a complicated China minority-interest structure, a future subsidiary IPO that could further dilute AXT's economic interest in Tongmei even if control remains, and a redeemable minority stake whose economics depend on Chinese approval processes. The offering prospectus also notes that as long as AXT's Series A preferred stock remains outstanding, the company and its subsidiaries cannot repurchase common shares unless accrued preferred dividends are paid. That makes a future cash-rich buyback response less flexible than casual investors may assume. Governance here is functional but imperfect, and imperfections matter more when the stock trades at narrative-driven multiples.

Industry and peers

Industry structure and policy cycle

AXT sits in a small but strategically important corner of the semiconductor chain. Its substrates are used where silicon no longer performs well enough: optical links, lasers, RF devices, selected wireless applications, and other high-performance optoelectronic functions. The company's own investor materials now place "data center connectivity" up front among end markets, which is consistent with the Q1 2026 product mix and with management's repeated emphasis on AI-oriented optical transmission. This is a photonics-materials exposure to bandwidth growth, not a broad silicon memory cycle.

The industry's most important cycle for AXT is now a combination of technology cycle and policy cycle. The technology side is favorable: rising bandwidth, scale-out AI architectures, and faster optical modules all support more InP content. The policy side is hostile: China announced export controls on gallium and germanium in July 2023, effective August 1, 2023, and China's current dual-use catalog requires licensing for controlled items, with AXT's own filings and calls showing that all three product families are now in a permit-governed environment. Management has said the permit process is opaque and around 60 business days for InP. That means the company can be "right" on demand and still miss a quarter.

Upstream geopolitics amplifies that vulnerability. The U.S. Geological Survey's 2026 gallium summary says China accounted for 99 percent of world primary low-purity gallium production. Its work on gallium and germanium supply disruptions frames China's export controls as a meaningful macroeconomic risk, not a symbolic trade action. Germanium remains a smaller direct product line for AXT, but the broader point stands: AXT's production geography overlaps with one of the most contested mineral and materials supply chains in the world. That gives the company access to capability. It also means investors are underwriting a geopolitical variable whether they want to or not.

Horizontal analysis

There is no clean public comparable for AXT. That is an analytical fact, not a dodge. Most public photonics names sell finished components or systems. Most public semiconductor-materials names either sit in different substrate families or are buried inside larger diversified groups. The right way to compare AXT is therefore to use a reference set rather than pretend a perfect peer group exists. The most relevant public references are downstream optical suppliers and photonics hardware names that benefit from the same AI optical build-out, because they reveal how the market is pricing the chain.

AXT price and market cap from market data and public quotations; peer prices, market caps, and P/E from the finance tool as of the U.S. trading close on 2026-07-10.

What this group portrait says is useful even without pretending these are direct substitutes. Coherent, Lumentum, AAOI, and IPG are being valued as beneficiaries of the same optical and AI capital-spending wave, but they sell downstream products closer to the customer budget line. AXT sits much further upstream and is much smaller, yet its market value has reached a level where it no longer looks cheap even relative to a chain already carrying rich sentiment. The difference is that downstream names at least convert demand into recognizable end products. AXT still has to convert demand into export permits first. That should earn it a discount, not parity with the chain's most speculative expectations.

Customers choose AXT today for three reasons that seem credible from the evidence available. They want qualified InP supply for high-speed optics, they want a supplier with tangible scaling plans and upstream material access, and in China they can buy without export permits at all, which management says is already creating fast domestic growth in InP-based laser demand. Customers will leave, or at least diversify away, for the opposite reasons: policy uncertainty, shipping delays, and geopolitical pressure to source outside China. The market currently prices the first set of reasons much more heavily than the second. That imbalance is the core valuation problem.

Current fundamentals and price action

Last four quarters and what the market is trading

The trailing operational story is cleaner than the stock story. Q2 2025 was weak, with revenue of $18.0 million and an 8.0 percent GAAP gross margin after AXT pre-announced that gallium-arsenide permits were being issued more slowly than expected and demand in China was weaker. Q3 2025 recovered to $28.0 million of revenue with a 22.3 percent GAAP gross margin. Q4 2025 fell back to $23.0 million after the company updated expectations in January 2026, saying it expected only $22.5–23.5 million because fewer InP export permits had been issued than previously expected. Q1 2026 then improved again to $26.9 million, with 29.6 percent GAAP gross margin and only a $1.6 million GAAP net loss attributable to AXT. The underlying business has therefore improved, but not in a straight line. It still lurches with permit timing.

What the market is trading right now is a blend of real fundamentals and a thematic extrapolation. The real part is that Q1 2026 included $13.6 million of InP revenue, that Q2 2026 had about $34 million of revenue already visible through permits in hand or permit-free domestic shipments, and that management expected profitability in Q2 even before additional permit upside. The thematic part is stronger: the market is treating AXT as a scarcity-value access point to the AI optical buildout. That is why a company with under $30 million of quarterly sales could sell more than $600 million of stock in one offering. That print says something more specific than "AXT had a decent quarter." It says "AXT may be one of the few public ways to buy a critical InP choke point."

That thematic premium would be easier to defend if public evidence of customer capture were firmer. Management says AXT is supplying nearly all leading optical customers and that its materials are used in multiple U.S. hyperscalers, but public filings and releases still do not name those hyperscalers or major end customers. For common-stock valuation, that missing bridge matters. The stock is being priced as though customer validation has already become public certainty. It has not. What is public is management commentary, order backlog, and capacity intent. That is meaningful, but it is not the same thing as disclosed multi-year supply agreements with named counterparties.

Price-action attribution mandate

The price action over the last twelve months is too extreme to wave away. The most defensible attribution is a stacked sequence. First came relief that AXT's permit-constrained business was not broken. Then came evidence that InP demand was materially better than the legacy business mix. Then came a scarcity narrative linking InP supply directly to hyperscaler and AI-optics spend. Then came momentum and thematic flow, which carried the stock far beyond what quarterly revenue alone could explain. The stock did not rise because revenue doubled. It rose because the market imagined what revenue could become if permits flowed, capacity doubled, and AXT's share of the optical chain proved more strategic than previously assumed.

The secondary-source confusion in the prompt is now resolved against the company's own record. The $22.5–23.5 million guidance figure was not 2026 quarterly guidance after Q1. It was the January 8, 2026 update for Q4 2025, explicitly tied to fewer-than-expected InP export permits. That matters because it shows the market's subsequent enthusiasm did not emerge from a company suddenly saying permits were no longer a problem. The company kept saying the opposite. It said permits were the key bottleneck, including in Q1 2026 commentary, and that Q2 visibility reflected what was already permitted rather than a normalized policy environment.

The offering itself deserves direct interpretation. AXT's April 2026 prospectus showed 64.1 million shares outstanding before the underwriter's option and 65.4 million if the option were fully exercised; it priced 8.56 million shares at $64.25 with the option for another 1.28 million, and the company later described total gross proceeds as $632.5 million. Total proceeds were earmarked broadly: support Tongmei's effort to expand InP capacity, fund R&D for new or improved products, and cover working capital and general corporate purposes. This reads less like a narrow project-finance story and more like a management team taking maximal advantage of a valuation window. The company also said the capital raise would support R&D investment in products like 6-inch InP. On balance, the raise signals both confidence in demand and pragmatism about valuation. Management did not wait for perfect visibility. It sold stock while the market was paying peak-theory prices.

Bull and bear divergence

The bullish case rests on four facts. Demand is visibly improving in the one product family that matters most, Q1 2026 margins improved sharply, AXT says backlog exceeds $100 million, and Q2 2026 already had around $34 million of revenue visibility before unissued permits. This is a business trying to release demand through a bottleneck, not one merely hoping for demand to show up. If the permit regime stabilizes rather than worsens, revenue can scale quickly because capacity expansion is already financed.

The bearish case rests on four other facts. First, the permit regime has already caused multiple revenue disappointments across GaAs and InP. Second, all three product families now sit inside that same export-control architecture. Third, the stock is still priced at a venture-like multiple of current sales despite continued losses and negative operating cash flow. Fourth, the business has just diluted shareholders massively twice in four months. When a policy-gated manufacturer with sub-$100 million trailing revenue is valued in the billions, the burden of proof shifts heavily onto execution.

Valuation, risks and conclusion

Historical valuation and peer framing

Trailing valuation is the simplest place to start because it already makes the point. At about $3.74 billion of equity value and roughly $95.9 million of trailing-twelve-month revenue, AXT trades around 39 times sales. Using annualized Q1 2026 revenue produces a lower but still extreme figure of roughly 35 times sales. The company has no useful trailing P/E because earnings remain negative. On owner earnings, the picture is even harsher. In 2025, operating cash flow was negative $12.8 million and capex was about $6.0 million, implying deeply negative owner earnings. That makes any headline forward-EPS valuation fragile by construction. For this company, valuation has to be anchored in scenario analysis, not comfort multiples.

Peer comparison does not rescue the stock. The reference-set names most exposed to the optical build-out are themselves richly priced, but they are also much more obviously tied to finished photonics products. AXT's premium case depends on the market granting a scarcity multiple to an upstream supplier while overlooking policy gating, minority-interest leakage inside Tongmei, and the absence of named public customer wins. The market may keep doing that for a while. It is still hard to defend on disciplined valuation grounds.

Absolute valuation

The table below uses a practical framework rather than false precision. Because June 30, 2026 cash has not yet been reported, I use a simplifying assumption that post-offering net cash is worth about $9–10 per share after considering pre-offering debt, recognizing that actual net cash could vary depending on interim capex deployment and working-capital needs. The operating scenarios then value the business at EV/Sales multiples aligned with each outcome. This is valuation-scenario analysis within a research framework, not investment advice.

Margin of safety, risks, and tracking indicators

On margin of safety, the verdict is direct: none. The current price clears conservative value entirely and sits above even a reasonably generous optimistic case. The most fragile assumption in the base case is permit conversion, not market demand. Cut the base case's "steadier shipments" assumption to 70 percent and the valuation quickly drops toward the high-$20s or low-$30s per share. If earnings were flat for the next three years because permits kept distorting shipment timing, the expected annualized return from today's price would be poor and likely below long-term Treasury yields. This is the textbook "good asset, bad price" setup.

The most dangerous permanent-loss risks are specific. First is policy deterioration: Beijing can tighten permits, slow approvals further, or use the process selectively. Probability medium; impact high; observable indicator is permit-guided revenue repeatedly missing actual demand commentary. Transmission path: revenue deferral, margin compression, and multiple collapse. Second is execution with dilution: AXT has raised more capital than its current revenue base historically justified, and if that capital does not produce high-return capacity, investors will re-rate the stock as overcapitalized rather than strategically funded. Probability medium; impact high; indicator is capex rising faster than realized InP revenue. Third is customer concentration hiding inside unnamed accounts: management says no one customer exceeded 10 percent of Q1 revenue, but the real dependency may sit in a small group of optical applications rather than in a single billing line. Probability medium; impact medium-high; indicator is backlog growth without corresponding conversion breadth. Fourth is geopolitical escalation from the U.S. side, which could affect customers, investors, or supply-chain permissions even if Chinese permits improve. Probability low-medium; impact very high; indicator is new U.S. restrictions on China semiconductor materials or cross-border capital-market activity.

The near-term positive catalysts are clear enough. Another quarter like the Q1-to-Q2 2026 setup, where permitted revenue visibility rises again and profitability materializes, would support the fundamental story. So would evidence that InP capacity is actually nearing the stated $35 million quarterly run rate. A constructive update on the Tongmei STAR process could also help sentiment, though it would raise fresh questions about future minority dilution. Negative catalysts are equally visible: another permit-related shortfall, slower-than-expected use of proceeds, or any hint that customers are actively dual-sourcing away from China.

Q2 2026 visible-revenue floor, backlog, and capacity commentary come from the Q1 2026 call; margin and revenue thresholds come from recent reported results; next earnings date comes from company IR.

Cross-synthesis summary

Looking across the whole arc, AXT has proven one thing decisively: it knows how to build and operate a specialized compound-substrate manufacturing and supply-chain ecosystem in China. It also seems to have proven that this ecosystem matters more in an AI optical cycle than the market previously recognized. Where the market is probably wrong is the ease of monetizing the opportunity, not its existence. The stock price implies that backlog, permits, capacity, and customer qualification are all likely to move in the same favorable direction together. The company's actual record says they do not move together. Demand can surge while permits drag. Capital can be abundant while owner earnings remain negative. A scarcity narrative can be true while the equity is still overpriced.

AXT therefore looks like a company whose business prospects improved faster than its operating evidence, and whose stock improved faster than both. If permit cadence normalizes and management executes the InP ramp, the company could grow into a meaningfully larger revenue and earnings base over the next three to five years. If permit friction persists, the market will eventually stop valuing backlog as if it were ordinary backlog and start valuing it as contingent backlog. That is a very different number. At today's price, investors are being asked to pay for the cleaner version of the story before the business has earned that cleanliness. I would change my view only if one of two things happened: either the stock fell into a valuation range that offered a clear margin of safety, or AXT reported several consecutive quarters proving that permits, capacity, and cash conversion are all improving together. Until then, the asymmetry is poor.

Bull and bear reasons

Bull reasons

• InP revenue was already $13.6 million in Q1 2026 and management said it was driven primarily by data-center applications, which confirms the optical-AI thesis has reached reported results.
• Management said InP backlog exceeded $100 million and that about $34 million of Q2 2026 revenue was already visible from existing permits or permit-free domestic demand, giving the demand case more substance than a generic AI slogan.
• AXT's integrated supply chain and self-built crystal-growth capability provide a plausible edge in scaling InP faster than less integrated rivals.
• The April 2026 offering removed near-term balance-sheet stress and gives the company resources to fund capacity and product development without relying on debt.

Bear reasons

• Every AXT product family now depends on Chinese export-permit issuance, and the company's own 2025–2026 quarter pattern shows permits have repeatedly changed realized revenue.
• At about $3.74 billion of market value versus roughly $95.9 million of trailing revenue, the stock still trades around 39 times sales despite negative owner earnings.
• AXTI's two recent equity raises massively diluted existing holders and suggest management was willing, and eager, to sell into a very rich valuation window.
• The Tongmei structure leaves AXT with control but not full economics, and the still-pending STAR process adds contingent minority-redemption and future-dilution complexity.

Pre-mortem

A plausible three-year down-50-percent script is this: by 2027, InP demand remains healthy, but permit approvals stay irregular and customers increasingly dual-source outside China for strategic reasons. AXT doubles nominal capacity, yet annual revenue only reaches around $120 million because shipment timing never normalizes. Gross margin stalls near the high-20s instead of the high-30s, owner earnings stay weak, and the market cuts the multiple from a scarcity premium toward 6x sales plus cash. On today's share count, that can take the stock into the high-$20s.

A second script is capital-allocation failure. The company spends heavily from the April 2026 raise, but 6-inch product development takes longer than planned, domestic Chinese demand grows faster than export demand, and Western investors decide that the cleanest way to play AI optics is downstream, not through a permit-gated substrate supplier. AXT then looks less like the bottleneck and more like an overcapitalized upstream option. Even without an operating collapse, the re-rating unwinds and the stock halves through multiple compression alone.

Final research conclusion

AXT is attached to a real strategic bottleneck. That is the good news, and it is not trivial. The company has real InP demand, a plausibly valuable supply-chain position, and enough fresh capital to attempt a step-function expansion. The trouble is that the stock already prices that future far more confidently than the business has earned. AXT is still a permit-gated, structurally complex, cross-border manufacturer whose revenue conversion depends on Chinese policy timing. That deserves a discount to a clean AI infrastructure supplier. The market has given it a premium instead.

From today's price, the more useful question is what is left for new buyers after a once-in-a-generation re-rating, not whether AXT is interesting. My answer: not enough. AXT could still become a bigger and better company than the pre-2025 market ever imagined. But the current stock already assumes too much of that upside, while leaving little protection against the one variable management does not control. The better setup would be a much lower entry price plus evidence that permit cadence is becoming less erratic. Until then, this is a name to respect operationally and avoid valuation-wise.

【Company-profile scores】

• Fundamental quality: medium
• Growth: high
• Moat: medium
• Financial soundness: strong
• Management credibility: medium
• Valuation attractiveness: low
• Risk level: high
• Suitable investor type: event-driven

【Investment rating】

• Rating: Avoid
• One-line thesis: Real InP demand exists, but a permit-gated, loss-making supplier at roughly 35–40x sales offers no valuation cushion.
• Three price signals:
  • 【Ideal Buy Price】22–24 USD Basis: about 20% below a conservative intrinsic-value band centered on roughly $28–30 per share, which assumes only modest permit normalization and about $120 million of medium-term revenue.
  • Acceptable hold price: 34–46 USD
  • Clearly overvalued price: 57–63 USD
• Current-price classification: clearly overvalued.
• Whether to wait for a better price: yes. A buy would require both a pullback into the low-$20s and at least two quarters showing permit-backed revenue conversion, not just backlog accumulation. The opportunity cost of waiting is that a sudden permit surge could trigger another squeeze.
• Target holding horizon: 1–3 years
• Expected annualized return: conservative about -21%; base about -11%; optimistic about -3%, using three-year convergence toward scenario values from the current price.
• Max-loss risk: about 60% to 70% if permit issuance worsens and valuation compresses toward a cash-plus-conservative-sales framework.
• Reassessment-trigger signals:
  • if revenue visible from permits in hand falls back below $25 million for a quarter;
  • if InP revenue drops below Q1 2026's $13.6 million for two consecutive quarters;
  • if GAAP gross margin falls below 20% for two consecutive quarters;
  • if Tongmei's STAR process is canceled or the $49 million redemption feature becomes likely to be exercised;
  • if cash from the April 2026 raise declines sharply without a clear step-up in realized capacity or shipments.

【Valuation Range】

• current: 57.21 (close as of 2026-07-10)
• bear (conservative · ideal buy zone): [22, 24]
• base (fair · acceptable hold zone): [34, 46]
• bull (optimistic · above the clearly-overvalued line): [57, 63]

Research uncertainties

The biggest uncertainty in this report is customer specificity. Management's commentary strongly implies meaningful exposure to major optical and hyperscaler demand, but public filings and releases do not name those customers. The second uncertainty is the exact split of current demand between U.S. exports and China domestic sales; management gave directional commentary, but not a full geography-by-product bridge. The third is cash usage after the April 2026 offering, because June 30, 2026 financials were not yet reported as of the research date. The fourth is the STAR timeline, which remains open-ended despite years of procedural progress. The fifth is that some useful product-level details, such as the Q1 2026 InP/GaAs/Ge revenue split and backlog wording, were available through a republished earnings-call transcript rather than through the primary press release itself; I have used them, but with caution.

Sources

The core primary sources were AXT's 2025 Form 10-K, its Q1 2026 Form 10-Q, the April 2026 prospectus supplement, company earnings releases for Q2 2025 through Q1 2026, and company investor-relations materials on upcoming earnings events. For market data and price history, I used the finance tool plus public quotation and historical-price sources. For the policy layer, I used official Chinese licensing materials and U.S. Geological Survey publications on gallium and germanium supply chains. Where a datapoint came from a republished call transcript or trade-press summary rather than a filing, I have either said so directly or used it only where it matched the direction of management's primary-source commentary.

Other tickers mentioned

• COHR.US: downstream photonics and optical-components reference point for how the market prices AI-optics exposure.
• LITE.US: optical-networking reference used to frame valuation sentiment across the same demand chain.
• AAOI.US: optical-transceiver reference used for the AI-data-center optics narrative and momentum comparison.
• IPGP.US: photonics-equipment reference used to show how public markets value adjacent optical infrastructure exposure.

This report is based on public information and does not constitute investment advice. Markets carry risk; invest with caution.