·Dashboard·Research·Work·Archive
/
← wiki
ticker stocksai-semiopticsinpsubstrateschina updated 2026-05-31

AXTI — AXT Inc. / Beijing Tongmei

Thesis

The market is debating whether AXT is a stranded China-supply story or a structurally advantaged InP merchant supplier in a multi-year shortage. The bear case treats Chinese substrate as permanently impaired and Sumitomo as a clean substitute. Both pieces are wrong. MOFCOM permits delay and re-route Chinese supply rather than removing it; Q4 2025 revenue was 99% non-US (APAC 81.5% / Europe 17.5%) and management explicitly says it is "getting permits pretty readily for U.S. customers based in other global regions." Sumitomo is not a neutral merchant supplier — it is also the world's #2 EML/CW laser maker, sells transceivers in direct competition with Lumentum and Coherent, and its 2026 +40% InP wafer expansion is paired with internal SEDI device-capacity language, not framed as merchant supply. The real addressable ex-China merchant pool is smaller than headline market share suggests, which is why Lumentum and Coherent are both building captive InP fabs rather than relying on Sumitomo. Result: in a market where qualified InP capacity takes years to add, AXT's qualified, non-vertically-conflicted merchant position retains strategic value despite permit risk.

Identity

  • Parent: AXT Inc., Delaware, HQ Fremont CA. NASDAQ: AXTI.
  • Operating subsidiary: Beijing Tongmei Xtal Technology — founded 1998, AXT holds ~85–86%, ~7.3% with 10 Chinese PE funds (~$48–49M invested for redeemable stake contingent on STAR IPO), ~7.6% with former BoYu minority holders, ~0.4% JinMei minority. STAR Market application (688756.SH) filed Jan 2022, SSE-approved Jul 2022, CSRC-accepted Aug 2022; IPO has not closed as of Q1 2026 — management says Tongmei "remains in process as part of a much more selective and smaller group of prospective listings."
  • Raw-material subsidiaries (under Tongmei): JinMei (gallium and now high-purity indium refining — vertical-integration milestone disclosed Q1 2026), BoYu (pyrolytic boron nitride crucibles).
  • Products: Compound-semiconductor substrates — InP, GaAs (semi-insulating + semi-conducting), Ge — produced by Vertical Gradient Freeze. Wafer sizes 2", 3", 4" in production; 6-inch InP under qualification, customer roadmap "within ~one year" per Q1 2026 call.

Industry landscape

The full InP supply-chain analysis — direct vs. indirect bandgap, why silicon physically cannot replace InP for telecom-wavelength lasers, the 6-inch transition economics, and the Coherent vs. Lumentum dynamic — sits in inp-sige-photonic-materials. The captive-vs-merchant analysis on Sumitomo (which underpins the AXT thesis) lives in that primer's "Sumitomo: captive vs. merchant InP" section. Read it as part of the broader landscape.

Headline structure relevant here: AXT/Tongmei holds estimated 60–70% of global InP substrate share; Sumitomo Electric (TSE: 5802) holds ~30% (with the wide range — some sources cite up to 60% — reflecting whether captive output is counted). JX Advanced Metals, Freiberger, InPACT, and IQE fill the remainder. Chinese entrants (Vital Materials, Zhuhai Dingtai Xinyuan) trail on telecom-grade quality.

Financials snapshot

Period Total Rev InP GaAs Ge Raw mat. Non-GAAP GM
Q1 2025 $19.4M — — — — GAAP −6.4%
Q2 2025 $18.0M $3.6M ~$6M ~$1M ~$6.7M 8.2%
Q3 2025 $28.0M $13.1M (+250% QoQ) $7.5M $0.64M $6.7M 22.4%
Q4 2025 $23.0M $8.0M $7.0M $0.23M $7.6M 21.5%
Q1 2026 $26.9M $13.6M (>50% of total) n/d n/d n/d GAAP GM 29.6%
  • Cash: $128.4M end-Q4 2025 (up $97.2M QoQ post-offering).
  • December 2025 raise: Closed 2025-12-30. 8,163,265 shares (incl. overallotment) at $12.25 = ~$100M gross / ~$93.9M net. Lead Northland Securities. Use of proceeds: Tongmei InP capacity expansion, R&D, GCP. (One secondary source cited a $632.5M raise — unverified, likely error.)
  • Pricing power (new this cycle): Q1 2026 call — "we are raising some of our prices" — first explicit pricing-power statement, citing indium cost pass-through and geographic standardization on lower-end GPON.
  • Q1 2026 InP backlog: >$100M record (vs. ~$60M end-Q4); Q2 2026 expected to be "the largest quarter for indium phosphide in AXT's history."

Capacity roadmap

  • Current run-rate baseline: ~$17–18M/quarter exiting 2025.
  • End-2026 target: $35M/qtr via brownfield expansion, ~$30–40M CapEx. Capacity already +25% since Oct 2025.
  • End-2027 / early-2028: $65–70M/qtr (next-door facility), ~$100M CapEx.
  • Greenfield optionality: $220–250M CapEx if demand persists.
  • 6-inch InP: In qualification for both In-doped and S-doped specs; customer adoption targeted ~12 months out.

Customer base

  • Named customers (per filings and trade press): Lumentum, Coherent, Broadcom, Intel, MACOM, AAOI.
  • Concentration is volatile by quarter due to permit timing:
    • Q3 2025: top 5 = 45.2% of revenue, 2 customers >10%
    • Q4 2025: top 5 = 22.6%, none >10% (concentration collapsed because permit-gated InP shipments stalled)
    • Q1 2026: top 5 = 32%, none >10%
  • New behavioral signal: Q1 2026 management says tier-1 hyperscalers and hardware OEMs are "encouraging their suppliers to enter into long-term supply agreements with AXT" — a structural shift from optics suppliers' historical preference to keep AXT as a backup behind Sumitomo/InPACT. This is the "qualification-becomes-strategic-dependence" inflection that the bear case dismisses.
  • End markets: datacom/AI optical transceivers (primary InP driver), telecom PON, EMLs/DFBs for 800G/1.6T, HBT for RF/PA, sensing, photovoltaics (Ge for space cells).

Geographic mix — direct evidence for the routing thesis

Q3 2025 Q4 2025 YoY (Q4)
Asia Pacific 87% 81.5% —
Europe 12% 17.5% —
North America 1% 1% (vs. ~11%)

The collapse from ~11% to 1% North America is the bear case's exhibit A — AXT has been functionally shut out of direct US shipments. But the same data is the bull case's strongest evidence the volume is going elsewhere, not vanishing. CTO Tim Bettles, Q1 2026 call: "we are getting permits pretty readily for U.S. customers based in other global regions. But that does not mean we are stopping any work on trying to obtain permits for the U.S." US-headquartered customers are receiving product through European and Japanese ship-to addresses. Domestic Chinese laser-customer revenue more than doubled Q1 and is expected to double again Q2 — no permit needed. MOFCOM has come back to AXT requesting more data on US-direct applications, which management calls "an encouraging sign that they're still looking at U.S.-based permits." No US-direct permit yet granted as of 2026-04-30.

MOFCOM permit dynamics

  • Trigger: China imposed export controls on indium and indium compounds 2025-02-04, framed as restricting military-applicable dual-use materials.
  • Permit cadence: First Tongmei InP permits granted 2025-06-11; additional grants August 2025. Stated processing time ~60 business days. No permit count disclosed by AXT.
  • Q3 2025: "Obtained export permits for a number of significant InP orders." InP rebounded to 3-year-high $13.1M.
  • Q4 2025: Missed guidance (revenue $23M vs. original $27–30M, revised down 2026-01-09 to $22.5–23.5M) because "fewer-than-expected export control permits…issued by China's Ministry of Commerce." CEO Morris Young: permit timing "can be fluid and doesn't necessarily align with our quarterly reporting."
  • Q1 2026: Permits "came in slightly better than guidance." Q2 "off to a solid start." CFO Gary Fischer: permit timing remains "the single most significant factor to our growth in Q2 and beyond" and "not predictable nor in our control."
  • No reciprocal US action found: AXT and Beijing Tongmei do not appear on any 2025–2026 BIS Entity List addition I could surface. No Tongmei-specific US controls.

The bear case

Most-likely target of the FundaAI counter-thesis: B. Riley Securities note dated 2026-03-18, "incrementally cautious" (analyst Dave Kang, last public PT $21 from 2026-02-20). Riley remained Neutral (was already Neutral; the firm's Buys are on Lumentum and Viavi). Core arguments from OFC 2026 channel checks:

  1. One major company "denied the existence of an InP bottleneck" at OFC 2026 — directly attacks the supply-shortage thesis underpinning AXT's >2,000% rally. (Identity not disclosed in the public summary; likely Coherent or Lumentum given their capacity ramps.)
  2. Another major company secured a seven-year supply agreement with Sumitomo Electric — implying customer displacement / long-term lockout.
  3. Adjacent overhangs: 502k+ insider shares sold over three months, board sought to expand authorized share count from 70M to 120M, ~$100M Dec 2025 follow-on diluted holders.

A separate and earlier articulation of the same skeptical view: Vikram Sekar (viksnewsletter), "Is Optics Really the Next Memory?" 2025-12-01. Doesn't name AXT but argues the supply tightness is real and self-correcting through industry expansion ("one company being sold out doesn't make a supercycle"). This is the cleanest version of the supply-self-corrects argument the bear case rests on.

A counter-piece already in market: Global Semi Research Substack (2026-04-18) rebuts the "low-barrier industry" framing and cites 20–25% industry yields (Chinese newcomers <10%), proprietary process knowledge, and customer certification cycles as durable moats.

The counter-bear case

From FundaAI, 2026-04-27 (verbatim)

We think the latest AXT report is too bearish. The key mistake is treating China-origin supply as permanently removed from the global market, rather than delayed, re-routed, and repriced. MOFCOM licensing is destination- and end-use-specific. Exporting substrate into European epi or laser capacity — for example through UK or broader European nodes — is very different from shipping directly into the U.S. supply chain.

The second issue is manufacturing efficiency. The WOLF lesson is not that compound-semiconductor substrates are bad businesses; it is that U.S. compound-semiconductor manufacturing is extremely difficult to ramp efficiently. AXT/Tongmei's China manufacturing base is not just a geopolitical liability — it is also a cost, yield, and execution advantage. In a market where qualified InP capacity takes years to add, that matters.

Customer diversification also should not be confused with displacement. Lumentum and others will qualify non-China sources for risk management, but in a multi-year shortage they still need every qualified wafer available. The report's own supply-demand model implies that the industry cannot self-correct quickly. That is precisely why qualified substrate supply should have value.

In short, the bear case overstates permanent impairment and understates routing optionality, backlog conversion, and China manufacturing efficiency.

Sumitomo is not a neutral merchant supplier

Another important point is that Sumitomo is not a neutral merchant substrate supplier. Sumitomo is itself an InP optical-device and laser player, which means Lumentum, Coherent, AAOI and other downstream laser suppliers cannot strategically rely on Sumitomo as their primary long-term source of critical InP substrate. In a tight market, Sumitomo's nameplate capacity is not the same as externally available merchant capacity; part of that capacity is likely to be captive, prioritized for internal downstream products, or commercially conflicted for competitors.

This makes the report's substitution argument too simplistic. Yes, customers will qualify Sumitomo and other ex-China sources for risk management, but qualification does not equal strategic dependence. For Lumentum and Coherent, relying heavily on Sumitomo would mean relying on a potential competitor for a bottleneck material. That is precisely why AXT/Tongmei remains strategically relevant: despite China export-permit risk, it offers qualified merchant InP supply that is not vertically competing with the same laser customers.

In other words, Sumitomo capacity should not be treated as fully fungible replacement capacity for AXT. The real addressable merchant supply is smaller than headline ex-China capacity suggests, which makes AXT harder to displace than the report implies.

What the data adds (2026-05-03)

Three pieces of evidence not in the original FundaAI text materially strengthen the counter-thesis:

  1. The routing thesis is empirically validated. Q4 2025 geographic mix (APAC 81.5% / Europe 17.5% / NA 1%) and the CTO's "we are getting permits pretty readily for U.S. customers based in other global regions" statement are direct evidence that "delayed, re-routed, repriced" is not a hopeful hypothesis — it is what Q1 backlog conversion looks like. >$100M backlog at end-Q1 2026 means the volume hasn't been displaced; it has accumulated.
  2. Captive-vs-merchant math on Sumitomo (full work in inp-sige-photonic-materials#sumitomo-captive-vs-merchant-inp): Sumitomo's ~80M annual laser-unit capacity (20–30% EML, balance CW per PhotonCap), plus APDs / coherent PIC die, implies meaningful internal wafer-area consumption. The +40% InP wafer expansion announced for 2026 is paired in Sumitomo Electric's own Nov 2025 growth-strategy deck with downstream EML/CW capacity language — not framed as merchant supply. Bounded estimate: ~30–60% of Sumitomo InP wafer output is captive, low confidence absent paywalled Yole data. Coherent and Lumentum building captive InP fabs (Coherent's 6-inch Sherman + Järfälla, Lumentum's San Jose expansion + new US laser plant March 2026) is the directly observable consequence — the customer-competitor friction is real and operative, not theoretical.
  3. Behavioral shift in customer posture. Q1 2026 commentary that hyperscalers and OEMs are now actively pushing their optics suppliers to sign long-term agreements with AXT is the inflection from "AXT is our backup" to "AXT must be in our supply mix." This contradicts the bear's "displacement to Sumitomo" framing in real time.

Honest caveats to the counter-thesis

  • Sumitomo is still a real merchant supplier. Japan-based fab is outside MOFCOM's licensing regime. If AXT shipments are disrupted at the margin, Sumitomo captures the price-inelastic spot demand at higher prices — a tailwind for Sumitomo's blended ASPs even at lower volume.
  • The captive-share bound is wide (30–60%). Without the Yole monitor or direct Sumitomo Electric IR clarification, the load-bearing number is uncalibrated. The qualitative case (capacity-expansion intent + downstream-product overlap + observable customer captive builds) is strong; the quantitative case is rough.
  • Permit timing remains structural risk. AXT has effectively no control over MOFCOM cadence. A meaningful tightening (e.g., a US-China escalation) could revisit the Q4 2025 miss pattern at any quarter. The bull case requires permits to drift toward looser, not tighter.
  • 6-inch InP qualification timing is the next supply-side overhang. Coherent's 6-inch fabs are already in production with reportedly higher yields than mature 3-inch lines. AXT's 6-inch is "within ~one year" away from customer adoption. If Coherent's 6-inch ramps and customer qualification before AXT's 6-inch in 2026–2027, the cost-per-die gap on volume products could compress AXT's pricing power even in tight supply.

Catalysts and risks

Near-term catalysts (next 1–3 quarters)

  • Q2 2026 print: management called it "the largest quarter for indium phosphide in AXT's history" — backlog conversion check.
  • US-direct MOFCOM permit grant — would re-open NA revenue line and remove the most-cited bear overhang.
  • Tongmei STAR Market IPO close — value-realization event for AXT shareholders if it happens.
  • 6-inch InP customer qualification milestones.
  • Long-term supply-agreement announcements with hyperscaler-anchored customers.

Risks

  • MOFCOM tightening (US-China cycle).
  • Coherent / Lumentum captive InP ramp eroding merchant TAM faster than expected.
  • Sumitomo +40% expansion turning out to be more merchant-targeted than current language implies.
  • Insider selling cadence and any further dilution if next CapEx phase is equity-funded.
  • The valuation: P/S 50–60x leaves no margin for execution stumbles or permit-cycle tightening.

Valuation snapshot

  • Price (2026-05-01 close): ~$95.97; 2026-05-02 intraday ATH $99.00.
  • 52-week range: ~$1.30 low → $99.00 high. Stock is up roughly 6,000% TTM.
  • Shares out: 55–65M (sources disagree post-Dec 2025 dilution; 10-Q will reconcile).
  • Market cap: ~$5.5B midpoint. EV ~$5.4B (cash ~$128M, minimal debt).
  • P/S TTM: ~55–60x on ~$95M TTM revenue. On the implied 2026 exit run-rate of $35M/qtr ≈ $110–120M actual full-year, P/S still ~45–50x. Multiple has expanded from <2x in early 2025.
  • Short interest: ~14.15% of float (~6.16M shares short), days-to-cover ~1 day on heavy volume — squeeze dynamics likely contributing to volatility.

The valuation is the cleanest argument the bear can make. A P/S in the 50s on a substrate company with permit-capped near-term revenue requires either (a) the merchant-supply chokepoint thesis to hold for several years, or (b) a 6-inch / margin-mix repricing on top of volume. The counter-thesis above defends (a). It does not defend the multiple.

Decision log

  • 2026-04-27 — Logged FundaAI counter-bear-case (above). Frames AXTI as still strategically relevant despite China export-permit risk on three grounds: (1) MOFCOM licensing is routing problem not permanent removal; (2) Tongmei's China cost/yield advantage vs. U.S. compound-semi ramps (WOLF analogue); (3) Sumitomo capacity is partially captive and commercially conflicted for downstream laser customers, so headline ex-China supply overstates merchant alternatives.
  • 2026-05-03 — Full deep-dive synthesis. Three pieces of evidence not in the original FundaAI text strengthen the case: (a) Q4 2025 geographic mix (NA 1% / APAC 81.5%) and Q1 2026 CTO routing quote empirically validate "delayed, re-routed, repriced"; (b) Sumitomo captive consumption bounded at 30–60% of InP wafer output (low confidence), with Coherent and Lumentum captive InP fab builds as observable confirmation of the customer-competitor friction; (c) Q1 2026 commentary on hyperscalers/OEMs encouraging long-term agreements with AXT marks the inflection from second-source posture to strategic-dependence posture. Open quantitative gap: Sumitomo captive share would tighten with paywalled Yole "Photonics GaAs and InP Compound Semiconductor Market Monitor" data — flag for inv-q follow-up. Verdict: counter-thesis is defensible on the supply-shortage and routing arguments; the valuation is not defended by this work. Position-sizing decision should be calibrated to permit-cycle tail risk and 6-inch competitive ramp, not headline backlog growth.

Topics

  • inp-sige-photonic-materials — InP supply chain primer; Coherent vs. Lumentum dynamics; 6-inch transition; Sumitomo captive-vs-merchant analysis
  • ai-infrastructure

Sources

Source updates (auto-maintained)

Drop/z Misc (May 23, 26) - Global Positioning in Stocks Contrarians beware BofA

BofA's April 2026 global fund-flow data shows long-only funds sold semis at -$50.6bn — the largest sector outflow globally — while rotating into Materials and Energy; AXTI appears in the stock positioning charts as part of this screen universe.

Relevant to your thesis: The broad semi de-risking provides macro context for AXTI's share-price volatility independent of fundamentals, but does not address the InP supply-demand or permit dynamics that drive the bull/bear debate.

Source: dropfile://z Misc/Global Positioning in Stocks Contrarians beware BofA.pdf

Pages that link here (4)
Sumitomo Electric: A hidden AI play or automotive conglomerate premium?Optical Networking & PhotonicsFoundry & Semiconductor MaterialsSemiconductors (cross-cutting / geopolitics)
Open in Obsidian