AI supply-chain thesis — mapping bottlenecks, focus companies, and supply-chain exposure for investors.
**Bottleneck theme:** InP & Compound Semiconductor Substrates **Focus:** $AXTI — AXT, Inc. ## Why this matters The 800G / 1.6T optical transceiver story sits on top of one ingredient that almost nobody outside the industry has heard of: **indium phosphide (InP) wafer substrates**. Every EML and DFB laser inside an AI data-center transceiver is grown epitaxially on an InP wafer. Two companies make ~80-90% of merchant InP outside China: **AXT (AXTI)** and **Sumitomo Electric (5802.T)**. Hence "InP duopoly". When optical transceiver volume ramps from 400G → 800G → 1.6T → 3.2T, InP wafer volume scales *more than linearly* (each transceiver uses more lasers, and AI workloads ship ~3-4x as many transceivers per port). This is the original "Strait" trade structure: a small upstream node with no fungible substitutes, sold into a long-tail downstream of fast-growing customers. The kicker: the geopolitics of Chinese export controls on InP have made AXT — which manufactures *entirely in China through Tongmei* — the single most policy-exposed company in the entire AI supply chain. ## Q4 2025 → Q1 2026: the export-permit volatility - China's Ministry of Commerce imposed export controls on InP in **February 2025** (after GaAs in August 2023). Every shipment requires a permit. - After being granted the **first InP export permit in late June 2025**, AXT's InP revenue rebounded: - Q3 2025: $13.1M (3-year high) - Q4 2025: $8M (permit issuance slowed) - **Q1 2026: $10.5M** with permits coming in slightly better than guidance - **Q1 2026 total revenue: $26.9M** (+17% sequentially, +39% YoY). - Asia-Pacific 78%, Europe 21%, **North America just 1%** — explicitly throttled by U.S. tariffs (now 70% on Chinese wafer substrates) and Chinese export controls. - **InP backlog reached a new high of >$60M** at end of Q1 2026, up from $49M at end of Q3 2025. ## Capacity story: doubling InP by end of 2026 - AXT added ~25% incremental InP capacity from October 2025 onwards. - On track to **double Q4 2025 capacity by end of 2026**, supporting a **$35M quarterly InP revenue run-rate** (i.e. ~$140M annualized just from InP). - Capex requirement is modest (~$30M) because expansion uses brownfield space at the existing Tongmei facility. - **$632.5M capital raise completed** to support Tongmei InP expansion and R&D. - **Tongmei STAR Market IPO** still under CSRC review — a separate value-unlock event. ## Customer base broadening Per CEO Morris Young's commentary, AXT is broadening its customer base "to include tier-1 companies in the optical space to which we have previously had limited exposure," explicitly including tier-1 laser manufacturers and optical transceiver module makers both in China and globally. This is a meaningful shift from the prior posture as a niche, lab-stage InP supplier — AXT is now a mainline qualified supplier into the AI buildout. ## The "Strait of AXTI v2.0" thesis The original v1 thesis from late 2024 was: *AXTI is the bottleneck on the InP-laser supply chain at 800G*. The v2 thesis acknowledges: 1. **Demand is real and accelerating** — confirmed by $60M backlog, capacity-doubling trajectory, and the broadening customer list. 2. **Geopolitics has bifurcated the market** — Chinese export controls force ex-China customers to take whatever permits AXT gets. This is a *constraint on revenue but a structural moat on pricing*. 3. **AXT's competitor (Sumitomo) does not have AXT's cost structure** because it manufactures in Japan with non-vertical materials supply. AXT has BoYu (pBN crucibles) and JinMei (Ga, Ge, InP poly) consolidated joint ventures that supply its own wafer line — a vertical integration that prints cash through every cycle. 4. **The Tongmei IPO is a put on AXT's geopolitical risk** — Tongmei traded as a separate Shanghai-listed entity would re-rate AXT's stake at Chinese semiconductor multiples while leaving the U.S. parent as a holding-company structure. ## Position in the supply chain AXT (Tongmei) makes InP, GaAs, and Ge wafer substrates from raw materials produced in its consolidated JVs (BoYu, JinMei). InP wafers ship to laser manufacturers (Coherent, Lumentum, AAOI in-house, Sumitomo, Mitsubishi, Sivers, IQE epi customer) who fabricate DFB / EML / VCSEL lasers. Lasers go into 800G and 1.6T transceivers (Coherent, AAOI, Innolight, Eoptolink, Hisense), which feed switches (Arista, Cisco, Broadcom, Marvell), which connect into hyperscaler fabrics (Microsoft, Meta, Amazon, Google, Oracle). ## Risks - **Single-source manufacturing in China**: Every wafer comes from Tongmei in Beijing. A geopolitical shock — escalating export controls, U.S.-China decoupling, or Taiwan tensions — could force a multi-year revenue gap as customers re-source. - **Export permit cadence**: Revenue is essentially gated by the China Ministry of Commerce. Q4 2025 missed guidance because of permit issuance speed; Q1 2026 was slightly better. This is a structural overhang. - **U.S. revenue collapsed to 1%**: A combination of 70% U.S. tariffs and Chinese export controls. Re-shoring InP wafer production to Korea (Sumitomo Korea), Taiwan, or the U.S. is a 3-5 year project that, if pursued, takes share from AXT. - **Sumitomo competitive response**: Sumitomo Electric is the duopoly partner. If Sumitomo decides to expand InP capacity aggressively (it has the capital), the duopoly economics weaken. - **Customer-build-internal**: AAOI fabricates its own lasers and could in principle move further upstream. Coherent has its own InP capacity (II-VI legacy). The merchant InP TAM at the high end could compress over 2027-2028. - **Currency / China policy**: Tongmei IPO progress is a function of CSRC review timing. Delays push the value-unlock event further out. ## What we're watching - Quarterly InP export permit cadence (key swing factor) - $35M/quarter InP revenue run-rate by year-end 2026 - Tongmei STAR Market IPO approval timeline - Sumitomo capacity announcements - North America revenue % (currently 1%) — any normalization signals a thaw in tariffs / export controls - AXT customer concentration shift toward tier-1 transceiver OEMs (announced Q4 2025 call)
Browse all AI supply-chain theses · Macro trends · Industries · Product categories