AXTI surged 28.74% overnight, but it was still at the halving level

NEXT PICK · Market Insights

AXTI surged 28.74% overnight, but it was still at the halving level

Friday, July 31, 2026

A long-established small company specializing in semiconductor substrates surged 28.74% in a single day today, closing at $60.43, with trading volume expanding to three times the daily average over the past three months. A financial influencer on X mentioned it seven times in one day.

The trigger was straightforward

After the market closed on July 30, AXT delivered its best quarterly report in its history—quarterly revenue of $47.6 million, up 164% year-over-year, with non-GAAP earnings per share of $0.19, while the market was only waiting for $0.07.

Yet this financial report, which exploded so much that the market needed to remodel, only pulled the stock price back from $46.94 back to $60.43, still 23% below the 50-day moving average and nearly 58% below the May 26 all-time high of $143.16.

Good news has materialized, yet stock prices are still lying in ruins—this gap is what truly deserves to be examined today.

Latest price $60.43 ▲ +28.74%

Q2 gross margin 45.0% ▲ Same period last year: 8.2%

Near the May high -57.8% ▼ Fell back from $143.16

The first brick for light advancement and copper retreat is stuck at the very top

AI data centers are undergoing an architectural transformation: connections between cabinets and even chips are shifting from copper cables to optical. And for light to run, there must first be a substrate that can grow into a laser.

AXT is made with this brick. The company designs and mass-produces three types of compound and single-element semiconductor substrates—indium phosphide (InP), gallium arsenide (GaAs), and germanium (Ge)—and supports upstream raw materials such as 6N+/7N+ high-purity gallium, providing one-stop solutions.

Among them, InP is the absolute star of this market cycle. It is used in data center optical interconnects, high-speed data transmission, 5G communications, fiber lasers and detectors, passive optical networks, silicon photonics, and photonic integrated circuits—each term represents an expanding pipeline for AI capital expenditure.

The other two lines weren't idle either

Semi-insulated GaAs for Wi-Fi, IoT devices, satellite communications, and power amplifiers; Semi-conductive GaAs serves micro-LED displays and VCSEL arrays for 3D sensing; GE substrates are designed for satellite multi-junction solar cells and high-end infrared detection.

So, why can a seemingly ordinary wafer cause such valuation fluctuations in the market?

Because the growth difficulty of compound crystals is an order of magnitude higher than that of traditional monocrystalline silicon. The stoichiometric ratios of volatile elements like phosphorus and arsenic must be precisely controlled, and the thermal field distribution during single crystal growth must be absolutely stable—a technique monopolized by a handful of old factories.

AXT has long been deeply engaged in proprietary crystal growth technologies such as Vertical Gradient Solidification (VGF), accumulating extensive experience in parameters that cannot be reverse engineered to reduce crystal dislocation defects and improve yield for large wafers. New entrants can buy the equipment, but not these parameters.

Its main production capacity is located at its Chinese subsidiary, Beijing Tongmei Crystal Materials, with a sales network covering direct sales in the US, China, and Europe, as well as independent agents in Japan, Taiwan, and South Korea. This "Made in China + Global Distribution" structure is both its cost moat and its largest risk exposure.

The claim of a 42x price-to-sales ratio has been shattered by just one quarter

First, let's talk about a premise that must be corrected

The market rumor of a "price-to-sales ratio of 42.88 times" is anchored to a full-year revenue of $88.3 million for fiscal year 2025—that is the bottom of the industry cycle and is now no longer worth referencing.

Combining the last four quarters, rolling revenue is about $125 million, corresponding to a market value of $3.95 billion, with a price-to-sales ratio of about 31 times. Based on the $66 million annualized figure in Q3 guidance, the multiple quickly shrinks to around 15 times.

More importantly, the denominator is still running. Management clearly targets Q3 revenue of about $66 million and earnings per share of $0.30 to $0.32, while Wall Street consensus estimates only $38.7 million and $0.10—figures the company itself gives, 70% higher than analysts.

The story of cash flow has been completely rewritten. The tense narrative in the research report about "cash on hand plunging from 120.3 million to 41.8 million USD" remained stuck in the first quarter; The April additional placement raised $632 million, and by the end of Q2, the company held $749 million in cash and investments.

This amount accounts for nearly 20% of the current market value. It directly dismantles the short-selling logic of "burning money until supply chain breaks", with the cost of diluting old shareholders—some of the market value expansion is not driven by rising stock prices, but by issuing out capital shares.

Even better, the money for capacity expansion began to come from customers. Casella prepaid $22.3 million, Coherent upfront $25.4 million, and Lumentum's protocol locked $87 million in deposits—these are not order intentions, but cash deposits.

So, is the valuation really expensive? Looking only at the present, a 31x rolling price-to-sales ratio is still ridiculous for a semiconductor materials company; the industry norm is 3 to 8 times, and the historical range for peer IQE is only 1 to 3 times.

Looking at 2027—management aims for $130 million in quarterly InP revenue, annualized $520 million, by which time the price-to-sales ratio will fall back to around 7.6 times. The price the market pays today is precisely for this vintage that has yet to happen.

The turning point has been confirmed, but the technical side has not yet been voted on

The report sets AXT at a breakout (a period of comprehensive breakout), a judgment that was almost textbook-like in the Q2 report.

The revenue trajectory is a beautiful V: bottoming out at $18 million in Q2 2025, then continuously rebounding, reaching $26.9 million in Q1 2026, and soaring to $47.6 million in Q2, a 77% quarter-over-quarter increase.

The leap in gross margin was even more dramatic—from 8.2% in the same period last year, to 29.9% in Q1, and then to 45.0% in Q2. Nearly 37 percentage points of expansion over four quarters—this is what heavy-asset firms look like when operating leverage is released.

Profits have also doubled

Non-GAAP net profit was $11.9 million, marking the company's end of two consecutive years of losses. With order backlog exceeding $100 million and record-breaking shipments, it is still growing, indicating demand is outpacing capacity.

But the market's technical aspects have not caught up yet. The $60.43 stock price remains below the $78.57 50-day moving average and only 30% above the $46.47 200-day moving average.

The fundamentals have completed a phase transition, and the price structure is still stuck in the post-crash recovery period—this divergence often means the divergence is far from over.

It's not the strongest, but it's the only one betting its life to expand production

Compound substrates are a highly technology- and capital-intensive niche market, and AXT's competitors are no small players.

Sumitomo Electric Industries of Japan is the undisputed veteran leader in GaAs and InP substrates, with deep crystal growth and stable yield control, holding a high share in RF and optical communication substrates. However, the decision chains of major companies are long and manufacturing costs high. Faced with pulsed demand like AI optical interconnects, expansion responses are usually slow.

Japan's JX Nippon Mining & Metals specializes in purifying high-purity metals, which directly underpins the purity foundation of its compound wafers. However, its strategic focus is not entirely on semiconductor substrates, nor does it have AXT's focus on continuously doubling InP.

Germany's Freiberger Compound Materials is Europe's stealth champion, specializing in ultra-high physical quality GaAs and specific Ge substrates, with customers focused on price-less automotive radar, industrial infrared, and space photovoltaics. Its scale is much smaller than AXT's, with limited presence in the most explosive data center InP.

The UK's IQE plc is the world's largest compound epitaxial foundry, and strictly speaking, it is not a direct competitor—it buys bulk substrates from companies like AXT, completes epitaxial layers, and then sells them to chip manufacturers. IQE's historical price-to-sales ratio is usually only 1 to 3 times, which highlights the oddity of AXT's valuation and shows that the market is indeed paying an AI premium for the "lowest layer of material."

The real problem is the three names on the demand side

Lumentum secures capacity until the end of 2031, and Coherent and Casella are making early payments. When downstream market leaders are willing to use cash instead of orders to compete for capacity, the moat no longer needs to be argued.

So here's the question

A single substrate from import to mass production for 1.6T transceivers requires extremely long certification cycles and optoelectronic device pairing verification. Do you think customers would gamble on a yield disaster caused by revalidation just for a small price discount?

Of the $220 billion capital expenditure, the light portion

From a top-down perspective, the certainty of the demand pool is unusually high. Amazon has raised its 2026 capital expenditure to $220 billion, compared to the previous $200 billion threshold, and management has made it clear that even this is still insufficient to meet the computing power and bandwidth needs for 2026 and 2027.

In clusters of dozens to hundreds of GPUs, copper cables have hit the physical ceiling in terms of power consumption, latency, and transmission distance, and must be replaced with InP-based silicon photonic devices and photonic integrated circuits. This is a substitute driven by the laws of physics, not a matter of preference.

In the serviceable market, a significant portion of this capital expenditure will be converted into 400G, 800G, and even 1.6T optical modules, as well as procurement of active optical cables and optical backplanes within data centers. The share of optical interconnects in hardware costs in hyperscale data centers is rapidly surpassing 10% to 15%.

Now, for AXT itself: the company plans to reach about $60 million in quarterly InP capacity by the end of 2026, and double it to $130 million by the end of 2027. Based on $30.7 million in InP revenue in Q2, this is more than four times the growth in three years.

The only constraint is not demand, but whether the furnace can be built and yield achieved. The fact that order backlog exceeds $100 million and continues to grow indicates a bottleneck on the supply side.

Five reasons that have already fallen into the report

The most powerful aspect of the bullish logic is that it no longer relies on stories—every data item in Q2 has turned assumptions into facts:

▲ Bull Case

Five questions that financial reports cannot answer

But this beautiful report precisely does not answer the questions that truly determine long-term valuation:

▼ Bear Case

Three time windows that determine the direction of the second half of the journey

The upcoming stock price rhythm will most likely be dominated by these three factors: whoever lands first gains pricing power:

•	Q3 earnings: Verify whether the $66 million revenue guidance and $0.30 to $0.32 earnings per share can be realized, while also monitoring whether the upside from export approvals will be released
•	End of 2026: Will the first doubling of InP quarterly capacity climb to about $60 million be completed on schedule? The direction is bullish but execution risks are not low
•	End of 2027: Quarterly InP revenue of $130 million with a final target of over 40% gross margin, which is the final anchor for current valuation
•	Lumentum received the first $43.5 million deposit (within 20 working days after signing the agreement), with the second $43.5 million scheduled for 2028
•	Tongmei is making substantial progress in listing on the Hong Kong Stock Exchange, with a timeframe of about one year. If successful, it opens up financing channels; if delayed, it will consume patience

Twelve months of heaven and hell

For the 2024-2025 fiscal year, AXT's books are a disaster record. Revenue for fiscal year 2024 was $99.4 million, dropping 11.2% to $88.3 million in fiscal year 2025, with quarterly revenue dropping to a freezing point of $18 million in Q2 2025.

Losses worsened simultaneously

Net loss for fiscal year 2024 was $11.6 million, widening to $21.3 million in fiscal year 2025, with a loss per share of $0.49. The vulnerability of the heavy asset model during industry downturns was exposed without any concealment.

Then the narrative flips in the second half of 2025. The market began repricing the architecture upgrade of AI optical interconnects, with funds frantically chasing AXT's scarce position on InP. The stock price recorded an epic increase over twelve months, surging to a historic high of $143.16 on May 26.

The collapse came just as swiftly. From $97.18 on June 12 to around $48 by mid-July, nearly half of the price has fallen in a month. What triggered this was not bad news, but the overall pullback in the semiconductor sector, profit-taking, insider reductions, and the fatal shortcoming of "no profits at the time."

The market is heading in the wrong direction, but out of rhythm. It correctly judged the AI cluster's craving for the underlying optical communication layer and indeed received Q2 performance verification; But the price it set in May was a one-time discount to the 2027 target for the present.

Now that the stock price has regained its upper upper limit above the 200-day moving average but remains below the 50-day moving average, the debate over "how far to expect to outperform fundamentals" is clearly not over.

⚠️ Risk Notice • China accounts for over half of revenue, export license timing is unpredictable, and policy changes immediately impact deliveries • The rolling price-to-sales ratio is about 31 times, far above the normal range of 3 to 8 times for the semiconductor materials industry • If the capacity doubling plan encounters yield bottlenecks or is postponed, the 2027 target falls short will trigger a valuation revaluation • Insiders sold a net $29.2 million in three months with zero purchases, signaling a negative management signal • Falling from $143 to $47 in two months, extreme volatility is extremely unfriendly for position management • If the optical communication technology route shifts to thin-film lithium niobate solutions, there is a risk of large-scale inventory impairment

🟡 Neutral The fundamental turning point has been confirmed, but valuations still bet on execution in 2027

💬 Discussion Don't chase highs; wait for a pullback to confirm. Live buy and sell positions are posted on AXTI via the official WeChat account.

Data source

•	Source: NextPick real-time snapshot + company Q2 2026 earnings and earnings call + SEC 8-K announcement + cross-verification by mainstream financial media (data as of 2026-07-31 US East Coast close)