The 2025–2026 High-Speed Optical Module Battle: Customer Map, Capacity Race, and Hidden Supply Chain Bottlenecks
Updated: Oct 7
As AI compute demand keeps exploding, the focus of the 2025 optical communications market has shifted from "who has the technology" to "who has the capacity" and "who can secure the core materials."
Based on the latest supply chain research, I've mapped out the current 800G/1.6T optical module landscape. This article focuses on three areas: the supplier map of the four major North American cloud providers (CSPs), capacity and pricing trends, and the most overlooked issue right now — upstream material bottlenecks.
1. The Optical Module Supplier Map of the North American Giants (Vendor Mapping)
The market currently shows a clear "two leaders" pattern: Innolight and Eoptolink together hold almost all of the most important share. Below is an analysis of each optical module vendor's penetration at the four major North American CSPs and NVIDIA:
Optical Module Vendor | Key Customers and Position | Strategic Highlights and Risks |
Innolight | Dominant across the board. • Google: Primary core supplier (60%–70%) • NVIDIA: One of two core suppliers • Meta/Amazon/Microsoft: Significant share at each | Covers every top-tier customer; the most secure position. Most aggressive capacity outlook (16 million 800G units projected for 2026). |
Eoptolink | Fastest growing. • Amazon: Primary core supplier (~60%) • Meta: One of three main suppliers (~30%) • NVIDIA: Has entered the 800G supply chain | Breaking into the NVIDIA supply chain is its biggest highlight. Expected to enter Google's LPO supply chain directly in the future. |
Finisar (Coherent) | Seasoned veteran. • Microsoft: No. 1 supplier • NVIDIA: One of two core suppliers • Google: About 20% | Diversified customer base, and controls key upstream chips and materials (e.g., Faraday rotators, EMLs) — strong vertical integration. |
TFC Communication (TFC) | Core components / optical engines. • NVIDIA: Indirect entry into 1.6T via Mellanox | Deeply tied to the NVIDIA ecosystem; the optical engine model lets it avoid the cutthroat competition at the module level. |
Source Photonics (Source) | Meta/Microsoft. • Meta: Lower-ranked but already qualified | Second tier; constrained by material supply, so capacity expansion lags the leaders. |
CIG (Cambridge Industries Group) | Cisco ecosystem. • Cisco: Joint design, stable relationship | Working to enter Meta's supply chain, which requires meeting a million-unit capacity threshold. |
Insights: Customer strategies differ markedly. Google and Amazon favor a "dominant core supplier" strategy (high concentration); Microsoft has the most fragmented supplier base; NVIDIA maintains a stable structure of 3–4 core suppliers to secure delivery.
2. 800G and 1.6T: The Tug-of-War Between Demand and Capacity
2025–2026 is a critical window for the speed transition, with the market in an "800G ramping, 1.6T racing ahead" phase.
1. Shipment Forecast
800G optical modules: 2025 shipments are expected to reach 35 million units. 2026 is conservatively estimated to exceed 43–45 million units.
1.6T optical modules: 2025 demand is about 10 million units (with severely insufficient capacity), expected to grow to 12–13 million units in 2026.
2. Price Trends
800G: FR (longer reach) currently sells for about $420–$450 and DR (shorter reach) for about $360–$380. Prices are not expected to start declining until 2H 2026, when capacity is released.
1.6T: Due to shortages, prices remain elevated ($1,300–$1,500), with no room for cuts in the near term.
3. The Deeper Bottleneck: Upstream Core Chips and Materials
This is what I believe deserves the most attention right now. The reason optical module capacity can't ramp isn't assembly and test — it's the upstream EML chips and the inconspicuous Faraday rotator.
In the table below, I've reorganized the original technical comparison into a "supply chain bottleneck analysis" to make it easier to see why silicon photonics (SiPh) is an inevitable trend:
Core Component | EML chip (electro-absorption modulated laser) | CW laser chip (continuous-wave laser) | DSP chip (digital signal processor) |
Current Status | 🔴 Severe shortage (the bottleneck) | 🟢 Supply relatively loose | 🟡 Bottleneck easing |
Cause of Shortage | Global gap of 20%–30%. Low 200G EML yields, slow capacity ramp, long equipment lead times. | Overall capacity can support the market; only high-power (100mW) parts face a structural shortage. | Capacity allocation at foundries such as TSMC has improved. |
Price Trend | Surging. 100G up from $5 to $7; 200G as high as $20–$25 per unit. | Stable. 100mW parts about $6 each; 70mW about $3.5 each. | High but declining. 800G DSPs have fallen to the $42–$80 range. |
Key Players | Lumentum (>40%), Broadcom, Coherent | 源杰 (leader), Lumentum, Sumitomo | Broadcom, Marvell |
Technology Impact | High cost and shortages are pushing customers toward silicon photonics solutions. | The core light source for SiPh solutions; will replace EMLs at scale. | LPO tries to eliminate the DSP, but the mainstream still needs one. |
The Hidden Killer: The Faraday Rotator (Core Material of the Isolator)
This is a very niche but critical bottleneck.
Status: Coherent holds 50% of global capacity and prioritizes internal supply.
Gap: The market shortage rate reaches 50%, with prices soaring from $120 to $175.
Impact: It directly causes optical isolator shortages, which in turn hold back optical module shipments.
4. Trend Summary: The Opportunity in Silicon Photonics and OCS
Silicon photonics (SiPh) adoption accelerates: Precisely because traditional EML solutions are too expensive and in short supply (200G EML capacity in 2026 falls short of 50 million units), SiPh solutions (SiPh + CW laser) will become the lifeline. SiPh is expected to reach a 60% share in the 1.6T era — a major tailwind for CW light source suppliers.
OCS (optical circuit switching) architecture: The 3D Torus architecture unique to Google's TPU clusters is driving OCS demand (estimated at 20,000–30,000 units per year). This is tied not to data rate but to port count, and will directly benefit passive component suppliers (such as Coherent, Fujikura, and CASTECH).
Comments