AI Demand Fills Optical Communication Orders, but InP Substrate Bottleneck Constrains Near-Term Delivery
nashnova research
AI network buildout has sold out optical component makers' capacity for the next 18 months, but upstream indium phosphide substrate shortages are choking laser-chip delivery — full order books mean nothing if the material can't keep up.
Order books are full — so why can't they ship?
Optical component and system vendors have sold out roughly 18 months of production capacity, driven by AI network infrastructure demand.
The real bottleneck sits upstream: AXTI and Sumitomo supply indium phosphide (InP) substrates — the key wafer material for making laser chips — and capacity is severely constrained.
This means → delivery timelines for electro-absorption modulated lasers and continuous-wave lasers are capped by substrate supply. The sector's near-term revenue ceiling is set by materials, not by orders.
Among current players, who secured supply?
Cisco and Nokia retained strategic InP inventory, giving them relatively stronger near-term delivery capability.
Ciena holds most of the line-systems market share but faces the same substrate-supply constraint.
In plain terms = the competition right now is not about who has the most orders — it is about who locked in upstream material early. Those who didn't are sitting on contracts they cannot fulfill.
Near-package optics are going standard — who benefits first?
The industry expects near-package optics (NPO) — moving optical modules from the rack to right next to the chip package, shortening signal paths — to enter AI scale-network testing by 2028.
The Open CPX standards body is unifying optical-engine specifications. This means → hyperscalers no longer worry about single-vendor lock-in, lowering the procurement barrier.
Commercialization is already underway: Coherent plans to ramp its PhotonLink product in 2027; Lumentum is expected to sign agreements within months. Because first-generation NPO still relies on external high-power lasers, both companies benefit directly from rising laser-chip demand.
All-optical switching — why is only Google deploying?
All-optical circuit switching (OCS) — routing data paths with light directly, bypassing electrical conversion — shows sharply divergent adoption across hyperscalers. Only Google has completed a large-scale production deployment so far.
Google plans to hand volume production to Lumentum over the coming years, but other hyperscalers remain at limited test scale. More telling: at least one potential cloud customer canceled deployment even after the solution passed full validation.
This means → the barrier to replacing existing switching infrastructure is not just technical — organizational inertia plays a role. Until more hyperscalers follow, suppliers' near-term OCS revenue is locked to Google's single-customer purchasing cadence.
What to watch next?
Two milestones will determine whether this optical-communications cycle converts to real revenue: InP substrate capacity expansion timelines and laser-chip yield improvement schedules.
In plain terms = until upstream materials unlock, downstream orders are paper numbers; until yields improve, even expanded capacity may not translate into shipments.
This reflects a shift in AI infrastructure's real bottleneck — from "who wins the orders" to "who can actually deliver." Supply-chain depth now matters more than demand heat for near-term share prices.
市场有风险，内容仅供研究参考，不构成投资建议。