Datacenter optics TAM
The datacenter optical-transceiver TAM is the headline number for both halves of the InP EML merchant duopoly. Aggregator estimates from LightCounting, Cignal AI, and Dell’Oro converge on a range of $20–25B for 2026 and $40–60B for 2028 under the AI-driven baseline. These are estimator-house numbers — every TAM cited here carries a ⚠ flag and should be cross-referenced against management commentary on supply-tightness and ASP trajectory.
Aggregator-house TAM estimates
| Source | 2025 TAM ($B) | 2026 TAM ($B) | 2028 TAM ($B) | Methodology |
|---|---|---|---|---|
| LightCounting | $16.5 | $26 | $40–50 | Bottom-up by speed grade × form factor; CY2025 newsletter ⚠ |
| Cignal AI | ~$14–16 | ~$22–24 | ~$35–45 | Hyperscaler procurement-survey-driven ⚠ |
| Dell’Oro | ~$15–17 | ~$22–25 | ~$40–60 | Vendor-revenue rollup ⚠ |
| Cisco / Bill Gartner public commentary | n/a | n/a | ”≥$20B AI optics” by 2030 | Vendor-management commentary ⚠ |
Source: LightCounting July 2025 newsletter ⚠; LightCounting January 2025 AI cluster forecast ⚠.
The directional consensus across all three estimators is 30–35% CAGR 2025 → 2026 with 15–20% CAGR 2027–2030 as base growth normalizes after the AI-cycle peak. The 2028 estimate spread of $40–60B reflects sensitivity to (a) CPO transition timing, (b) ASP discipline holding, and (c) Chinese-supplier penetration of Western datacom.
TAM decomposition by speed grade
Under the LightCounting baseline (extrapolated to nearest round numbers; ⚠):
| Speed grade | 2025 share | 2026 share | 2028 share |
|---|---|---|---|
| 100G / 200G (legacy) | ~25% | ~15% | ~5% |
| 400G | ~35% | ~25% | ~10% |
| 800G | ~30% | ~40% | ~30% |
| 1.6T (incl. CPO 1.6T) | ~10% | ~20% | ~40% |
| 3.2T+ (CPO scale-up) | n/a | small | ~15% |
The mix-shift from 400G to 800G to 1.6T is ASP-positive — each speed-grade upgrade roughly doubles transceiver ASP and approximately triples InP-chip content (more lanes per module). Coherent and Lumentum benefit on both vectors.
Coherent-vs-Lumentum supply-share allocation
Within the merchant InP source-laser layer (the chip-volume that drives the duopoly economics), the Coherent-vs-Lumentum share allocation is the structural variable. Both companies are vertically integrated to varying degrees — they sell finished transceivers as well as merchant chips.
| Layer | Total TAM (‘26) | Coherent share | Lumentum share | Notes |
|---|---|---|---|---|
| InP EML chip merchant supply | ~$3–4B | 45–55% | 35–45% | Combined ≥80% |
| Datacom transceiver finished module (merchant-sold) | ~$22B | ~12–15% | ~6–8% | Innolight, Eoptolink dominate the module-vendor layer |
| NVDA-direct dedicated capacity | $4B equity-anchored | $2B (4-year ramp) | $2B (4-year ramp) | Bilateral |
| CPO source-laser TAM (‘28+) | $5–10B | 45–55% | 35–45% | NVDA bilateral preserves duopoly |
⚠ Aggregator estimates throughout. The chip-level merchant share is the most defensible (cross-validated in trade-press and management commentary on duopoly framing); module-level share is more dispersed because of the strong Chinese transceiver vendors.
The structural read: Coherent’s broader vertical-integration scope (InP + SiPh + transceiver assembly via Finisar heritage) means more of the TAM expansion flows through Coherent’s revenue line than through a pure-chip vendor. This is why Coherent’s Datacenter & Communications segment grew 33.6% YoY in Q2 FY2026 even as some specific InP-chip merchant share rotated to Lumentum’s San Jose new-fab capacity dedication.
TAM under the bull/base/bear scenarios
| Scenario | 2026 TAM ($B) | 2028 TAM ($B) | Coherent revenue capture |
|---|---|---|---|
| Bull | $26–28 | $55–65 | Datacenter & Communications segment $7–8B FY2028 |
| Base (LightCounting central) | $24–26 | $45–55 | $5–6B FY2028 |
| Bear (capex digest) | $20–22 | $30–40 | $4–5B FY2028 |
The bull case requires (a) AI capex acceleration, (b) ASP discipline through 1.6T transition, (c) successful CPO ramp at NVDA — see CPO market. The bear case requires hyperscaler digestion + a third-source emergence — see InP EML duopoly.
⚠ Why aggregator TAM estimates are imperfect
Several structural reasons to flag aggregator TAM with care:
- Module-vs-chip double-counting — module ASP includes chip ASP plus module value-add. Some aggregator estimates count both layers; some count only modules. The headline $40–60B figure for 2028 typically counts modules at endpoint — chip-layer TAM is roughly 15–20% of module TAM.
- Captive-supply attribution — NTT, Sumitomo, and Chinese fabs supply some merchant share that doesn’t always land in aggregator-house module-share counts. Coherent + Lumentum’s “≥80% of merchant supply” framing excludes captive supply.
- NVDA-CPO accounting ambiguity — when CPO ramps, the source-laser layer is sold to NVDA as discrete components, and the assembled CPO engine is sold by NVDA’s switch business as part of the switch system. TAM apportionment between transceivers and switches is under reconciliation in the aggregator-house frameworks for 2028+.
- Chinese supply penetration uncertainty — Innolight and Eoptolink ship merchant 800G transceivers to AWS and others; their 1.6T qualification status shifts share allocations year-to-year. Trade-restriction policy is the binding variable.
For investment-grade work, trust the directional shape (≥30% CAGR through 2026, decelerating to 15–20% by 2028, with 1.6T mix-shift driving ASPs) rather than precise dollar-figures. Quarter-on-quarter Coherent revenue + management ASP commentary is the more reliable telltale than aggregator TAM revisions.
Cross-link
- AI capex cycle — top-down hyperscaler demand
- InP EML duopoly — supply-share mechanics
- CPO market — 2028+ architecture transition
- 05_financials segment revenue mix — Coherent reportable-segment view
- 05_financials margins and pricing — ASP trajectory
- LITE — datacenter optics TAM — duopoly-partner view
Sources
- LightCounting July 2025 cloud datacenter optics forecast ⚠
- LightCounting January 2025 AI cluster optics forecast ⚠
- Coherent Q2 FY2026 press release ✓
- Coherent Q1 FY2026 press release ✓
- LightCounting Optical Brochure — AI clusters, datacenters, optical networks ⚠
- Network World — OFC 2026 AI optics commentary ◐