Twelve months ago, AXT Inc. traded at $1.13. On February 20, 2026 — the day after Q4 2025 earnings — it closed at $29.68. That’s a 25x return on a company that makes semiconductor substrates.
Not AI chips. Not GPUs. Substrates — the raw wafer material that other companies use to fabricate chips that go into fiber optic transceivers that connect AI servers to each other. AXT is three degrees of separation from the exciting part of AI, and it’s still one of the best-performing small caps of the past year.
The question now: is this the end of the run, or a pitstop?
What AXT Actually Sells
AXT (NASDAQ: AXTI) manufactures compound semiconductor wafer substrates — the starting material for chips that can’t be built on silicon. Its three main products:
Indium phosphide (InP): The key material for laser drivers and photodetectors inside fiber optic transceivers. Every 400G, 800G, and 1.6T optical transceiver connecting AI servers needs it. This is the growth story.
Gallium arsenide (GaAs): RF chips for smartphones, 5G infrastructure, and satellites.
Germanium: Satellite solar cells and specialized photodetectors.
InP is what this trade is about. When Google, Microsoft, Amazon, and Meta scale AI data centers, they need enormous volumes of optical interconnects to move data between servers at low latency. Those transceivers are built on InP substrates. There are only a handful of companies globally that can make high-volume InP wafers — AXT is one of them.
In Q4 2025, InP generated $8.0 million of AXT’s $23.0 million in total revenue, almost all from data center applications. That sounds small. The backlog doesn’t: InP orders reached a new all-time high of over $60 million, as customers place longer-term orders because they’re worried about supply. That’s roughly 2.5 quarters of Q4-level InP revenue sitting in the order book right now.
Q4 2025: A Miss, But Not the Reason You Think
AXT reported Q4 results on February 19, 2026:
- Revenue: $23.0M — missed $24.24M estimate; down from $28.0M in Q3 2025
- GAAP EPS: ($0.08) — missed ($0.04) consensus by 4 cents
- Non-GAAP EPS: ($0.05) — missed ($0.04) by 1 cent
- GAAP Gross Margin: 20.9% (up from 17.6% in Q4 2024)
- Cash: $128.4M (up from $31.2M — a December equity offering raised $93.9M at $12.25/share)
Why did revenue fall sequentially from $28M to $23M? China export permits.
AXT’s Beijing-based manufacturing subsidiary (Tongmei) needs Chinese government export permits to ship InP wafers to international customers. In Q4, permits came in slower than expected — and the shortfall dropped directly to the revenue line. Not a demand problem. Not a product problem. A bureaucratic bottleneck.
CEO Morris Young was direct on the earnings call: “We were disappointed that we didn’t receive as many export permits in Q4 as we had hoped.” He was equally direct about the fix: “We have received some permits to date in 2026 and believe we are in a strong position to achieve sequential revenue growth in Q1.”
Full Year 2025:
- Revenue: $88.3M (down 11% from $99.4M in FY2024)
- GAAP Gross Margin: 12.7% (down from 24.0% — crushed by fixed costs on lower volumes)
- GAAP Net Loss: $21.3M, ($0.49)/share (nearly doubled from ($0.27) in FY2024)
- Non-GAAP Net Loss: $18.0M, ($0.41)/share (swung from +$0.20 non-GAAP profit in FY2024)
The annual numbers are ugly. Revenue down 11%, margins collapsed from 24% to 12.7%, losses nearly doubled. Most of this traces back to the export permit roller coaster — Q3 was strong ($28M) when permits flowed; Q4 was weak ($23M) when they didn’t.
Q1 2026 Guidance: What the Market Is Actually Buying
The stock jumped 24.7% on earnings day despite a Q4 miss. Why? Q1 2026 guidance:
- Revenue: $26.0M — beat consensus $24.8M by 5%
- EPS: ($0.04) to ($0.02) — beat consensus ($0.05)
Sequential recovery as permits flow in. That’s the trade. Morris Young gave a specific number on the call: China data center InP revenue is expected to grow more than 60% in Q1 versus Q4. New Tier-1 optical transceiver and laser manufacturer customers — companies AXT hadn’t worked with at scale before — are placing orders. U.S. hyperscalers are now in active conversations about material requirements.
The $60M+ InP backlog on constrained supply is the most credible signal here. Customers don’t pre-book semiconductor materials 6-9 months out as a courtesy. They do it because they’re genuinely worried about supply shortages.
The Capacity Build: Doubling InP by End of 2026
The bull case in one line: AXT is doubling InP manufacturing capacity by the end of 2026, funded by $93.9M it raised in December at $12.25/share.
The plan:
- Already added 25% capacity since Q3 2025 — immediately after the capital raise closed
- Doubling from Q4 2025 levels by December 2026 — $30M in total capex
- Leveraging Tongmei’s vertically integrated Beijing campus for faster, lower-cost scale-up (raw materials, crystal growth, and recycling all on one site)
- Potential second doubling in 2027 (~$30M additional capex), subject to customer demand confirmation
- Co-packaged optics (CPO) — where lasers integrate directly onto chip packages — expected to be the next inflection, arriving in late 2027 and beyond
Management expects to return to profitability in 2026 as revenue scales and fixed costs get absorbed. The consensus forward EPS estimate for FY2026 is $0.47/share — a dramatic swing from ($0.49) in 2025. If revenues hit $140-150M annualized by H2 2026, the operating leverage math on a manufacturing business does work. Whether they get there is the question.
Tongmei’s pending STAR Market IPO in Shanghai is pure optionality on top. China is aggressively scaling AI infrastructure, and a domestic listing for an InP substrate supplier would access Chinese investors who value that supply chain position very differently than U.S. markets currently do.
Valuation: What You’re Paying
| Metric | Value |
|---|---|
| Stock Price (Feb 20 close) | $29.68 |
| 52-Week Range | $1.13 – $30.80 |
| Market Cap | ~$1.64B |
| TTM Revenue | $88.3M |
| Price/Sales (TTM) | ~18.6x |
| GAAP EPS (TTM) | ($0.49) |
| Forward EPS (FY2026 est.) | ($0.12) GAAP |
| 50-Day Moving Average | $19.48 |
| 200-Day Moving Average | $10.73 |
| Beta | 1.96 |
| Short Float | 8.61% |
18.6x trailing sales. For a company with shrinking revenue and GAAP losses that nearly doubled. The 200-day moving average is $10.73 — the stock is trading at 177% above its own 200-day MA. For context, Ichor Holdings (ICHR) — another semiconductor small cap that ran 285% over 12 months — recently traded at similar premiums to analyst targets and faced the same scrutiny: does the business justify the price, or is this pure momentum?
What Analysts Think
- UBS: $35 — new coverage, most bullish on the street
- Wedbush: $28 Outperform — raised from $8.50 (a major re-rate)
- B. Riley: $21 Neutral — raised from $18, but downgraded from Buy in January 2026
- Northland Securities: $20
- Needham: Hold — downgraded from Buy in January 2026
- Craig-Hallum: Buy (reiterated)
Consensus: Hold | Average target: ~$22.80
At $29.68, the stock trades 30% above the average analyst target. Only UBS ($35) has a target above current price. Wedbush’s $28 Outperform is already below where the stock trades. Both Needham and B. Riley — who were Buy-rated when AXTI was cheap — downgraded as it ran. Analysts getting cautious at 52-week highs is standard behavior, but it means you’re buying a stock where even the bulls don’t see meaningful upside from current levels.
The Bear Case
Export permits are a binary risk you can’t model. Every quarterly result depends on Chinese government bureaucracy that AXT cannot control. Q4 2025’s miss was entirely due to permit timing — not demand, not product quality, not competition. If U.S.-China trade tensions escalate further, AXT’s revenue can drop off a cliff in any given quarter regardless of underlying demand. This is the primary risk and it’s geopolitical — which means unquantifiable.
Insiders sold near the offering price. CEO Morris Young sold 35,000 shares at $12.05 on December 4, 2025. Director David Chang sold 25,000 shares at $14.84 on December 9. The company raised capital at $12.25 on December 30. The stock is now at $29-30. Either both insiders badly misjudged the demand trajectory coming in Q1, or they were reducing exposure to a stock they believed was running ahead of fundamentals. There’s no clean interpretation here.
The annual financials are moving in the wrong direction. Revenue fell 11%. GAAP gross margin collapsed from 24% to 12.7%. Losses nearly doubled. The 2026 recovery thesis is entirely forward-looking — Q4 missed estimates, and the reported track record doesn’t support buying at 18x sales.
The Bull Case
InP has no substitutes at the high end. Silicon photonics handles some optical interconnect use cases, but EML-based transceivers at 400G/800G/1.6T and the emerging CPO architecture require InP substrates. This isn’t a commodity — it’s a narrow, specialized supply chain where AXT has a vertically integrated cost advantage and years of process experience.
The $60M backlog on constrained capacity is a real signal. That’s 2.5 quarters of Q4-level InP revenue in the order book, placed by customers worried about supply. When doubled capacity comes online through 2026, that backlog converts to revenue.
The AI data center capex cycle has a multi-year tail. Microsoft committed $80B in data center capex for 2026. Google, Amazon, and Meta have similar commitments. Every hyperscale data center needs optical interconnects, and optical interconnects need InP. This isn’t a one-quarter story. Fastly (FSLY) and Radware (RDWR) are playing the same AI infrastructure cycle at the software and security layer — AXT is playing it at the materials layer, which is arguably harder to displace.
Verdict
AXT is a real company supplying a real material to a real supply-constrained market. The InP-for-AI-data-centers thesis is legitimate — the $60M backlog, new Tier-1 customer wins, and 60%+ China data center InP growth expected in Q1 are not noise. The capacity expansion is funded and underway. The Tongmei STAR Market IPO is optionality most U.S. investors aren’t pricing in.
But at $29-30, you’re paying for perfect execution. The average analyst target is $22.80. The stock trades 30% above that. Management sold shares at $12 two months ago. Q4 missed estimates. The primary risk is a geopolitical binary that nobody can model.
If you own it: The $35 UBS target is reachable if Q1 and Q2 execute well. At 177% above the 200-day MA, have an exit plan before May earnings. Don’t let a great trade become a round trip.
If you’re considering buying new: Don’t chase the 52-week high. Better entry points near the analyst consensus of $22-25, or the 50-day MA at $19, will likely emerge if the market rotates or export permits disappoint even once. The thesis doesn’t expire; the entry price matters.
The key variable: Watch for any management commentary on permit flow before the May Q1 earnings call. That’s your leading indicator — export permits flowing means the 60%+ Q1 China growth is real; permit delays mean another Q4-style miss is coming.
Disclaimer: This article is for informational purposes only and does not constitute financial or investment advice. The author holds no position in AXTI at the time of publication. Small-cap stocks carry significant risk — always conduct your own due diligence before investing.