Skip to content
primarysourced Photonics sector GF
GFS
~8 min read · 1,895 words ·updated 2026-04-29 · confidence 38%

Confidence legend: ✓ verified-primary (LightCounting / Yole / Mordor / Markets-and-Markets press release; Tower / TSMC / GF official platform pages) · ◐ partial / aggregator · ⚠ inferred / estimate.

This page sizes the silicon-photonics market in which GFS Fotonix operates. The market has three structurally distinct sub-segments — pluggable transceivers, co-packaged optics (CPO), and on-board optics (OBO) — plus tail applications in LiDAR / sensing / quantum. Datacom (datacenter / hyperscaler) is the dominant revenue pool and the only one that grows at AI-capex-driven rates through 2030.

For the GFS thesis, the most important framing is: the foundry-wafer revenue capture per dollar of end-market spend is small (~10%) but the wafer-level differentiation defends pricing power. Fotonix’s monolithic 300mm CMOS-photonics integration is a meaningful technical advantage over Tower’s 200mm PH18 (with 300mm migration in progress) and over the various 200mm research-foundry alternatives.

1. Silicon Photonics Market Sizing — Bracketing the Forecasts

The silicon-photonics market is forecasted by multiple research firms with materially different methodologies. The key forecasts (as of April 2026):

SourceEndpointEndpoint ValueCAGRMethodology
Yole Intelligence (2025 SiPh Module report)2029$10.3B SiPh module market45% CAGRModule-level (transceiver / CPO / OBO modules); pluggable SiPh @ 52% = $5.3B
LightCounting (May 2024 report)2029$3B SiPh chip-level~25% CAGRChip-level only (not full module ASP)
Markets and Markets (2025–2030 forecast)2030$9.65B29.5% CAGRComponent-and-system synthesis
Mordor Intelligence (forecast)2030$10.36B27.21% CAGRComponent-and-system synthesis

The key reconciliation: Yole’s $10.3B by 2029 = full module ASP including DSP, lasers, packaging, etc. LightCounting’s $3B by 2029 = chip-level only. The forecasts are not contradictory — they measure different parts of the value chain. The foundry-wafer revenue is closer to LightCounting’s chip-level number divided by ~3 (since chip-level includes value-add beyond raw wafer), implying ~$1B foundry-wafer revenue at the 2029 horizon.

⚠ The 30% CAGR Markets-and-Markets number through 2030 implies steady growth; Yole’s 45% CAGR through 2029 implies sharper acceleration. The actual trajectory likely sits in between, with rapid acceleration through 2027 (during the AI-buildout peak) and tapering toward 25–30% by 2029–2030.

CPO sub-segment specifically

Co-packaged optics is the most-watched sub-segment:

  • 2024: ~$46M CPO market (per one Mordor-class forecast); essentially zero meaningful 2024 deployment.
  • 2025: First commercial CPO ramp (Broadcom Tomahawk-Ultra references; ~50,000 CPO switches shipped per DigiTimes — CPO Revolution analysis).
  • 2030: $0.75B (Mordor at 35.92% CAGR), $8.1B (one Mordor forecast at 137% CAGR), or up to $20B by 2036 (IDTechEx at 37% CAGR).
  • 2036: $20B+ (IDTechEx). ◐ IDTechEx — Co-Packaged Optics 2026-2036

The CPO forecasts are wider than for pluggables because the CPO transition is technically demanding and the timing of hyperscaler adoption is uncertain. The conservative case is that CPO ramps slowly through 2030 (more like the 35% CAGR / $0.75B path); the aggressive case is that CPO compresses pluggable demand sharply 2027–2030 (the 137% CAGR / $8B path).

For GFS Fotonix specifically, CPO is positive net-net because:

  • Each CPO switch contains 8–16 SiPh dies (8 ports × 2 lanes × dies).
  • CPO dies are typically sourced from a foundry tier (vs OEM-internal MEMS optics).
  • GF Fotonix’s monolithic CMOS-photonics integration is well-suited to CPO use cases (unifying driver + photonics on a single die).

2. Datacom Optics Market Segmentation

Pluggable Transceivers

The dominant volume segment today (2025–2026), targeting hyperscaler spine-leaf and inter-datacenter / DCI links:

  • 800G QSFP-DD800 / OSFP: Primary 2024–2025 hyperscaler deployments; mostly silicon-photonics-based at this data rate (vs VCSEL multimode, which doesn’t scale to 800G).
  • 1.6T OSFP: 2025–2027 deployment cycle. Almost entirely silicon-photonics-based. Multiple module makers (Innolight, Eoptolink, Coherent, etc.) selling into hyperscalers.
  • 3.2T: ⚠ on the roadmap for 2028–2030; technical feasibility of pluggable form-factor at 3.2T is debated; some industry consensus that 3.2T may be where CPO becomes mandatory.

Pluggable revenue per port: ~$1500 (800G) → ~$3000 (1.6T) at 2026 ASPs. Module-ASP / wafer-cost ratio: ~50:1, meaning the foundry captures ~2% of pluggable retail revenue.

Co-Packaged Optics (CPO)

Higher-density, lower-power-per-bit alternative. Optical engines are placed inside the switch ASIC package rather than at the front-panel cage.

CPO ASP per switch: ⚠ much higher than pluggable per-port equivalents at first deployment; expected to converge as volumes grow.

On-Board Optics (OBO)

A middle-ground architecture between pluggable and CPO. Lower deployment volume; mostly bypassed in favor of direct-to-CPO transition by leading hyperscalers. ⚠ less critical for the GFS thesis than pluggable + CPO.

LiDAR / Sensing / Quantum

Tail applications:

  • LiDAR for autonomous vehicles: Tower SiPho is primary supplier; GF less exposed.
  • Sensing: Mid-IR + visible spectrum SiPh applications; small revenue today.
  • Quantum: PsiQuantum (GF customer) leverages SiPh for photonic-quantum computing; volume scenarios are 2027+ at earliest.

The AMF acquisition (Nov 2025) explicitly cited “long-haul optical communications, computing, LiDAR, and sensing” applications as the rationale, indicating GF’s intent to pursue the full SiPh application surface.

3. Pluggable-vs-CPO Adoption Curve

The transition from pluggable to CPO is the most-debated topic in datacom optics. Two extreme scenarios:

Scenario A — “CPO replaces pluggable rapidly” (aggressive)

  • 2025: ~5% CPO penetration of new deployments.
  • 2027: ~25% CPO penetration.
  • 2029: ~50% CPO penetration.
  • 2030: ~70% CPO penetration.

Implication: pluggable transceiver TAM peaks in 2027 then contracts. Module makers (Innolight, Eoptolink, Coherent, etc.) face revenue compression unless they pivot to CPO modules. GFS Fotonix wins because CPO uses more SiPh dies per switch than pluggable does per port.

Scenario B — “CPO ramps slowly; pluggable persists” (conservative)

  • 2025: ~3% CPO penetration.
  • 2027: ~10% CPO penetration.
  • 2029: ~25% CPO penetration.
  • 2030: ~35% CPO penetration.
  • 2032: ~50% CPO penetration.

Implication: Pluggable TAM continues to grow through 2030+. CPO is incremental, not substitutional. GFS Fotonix wins because both pluggable and CPO use SiPh wafers.

Both scenarios are positive for GFS Fotonix at the foundry level. The downside scenario is “CPO ramps slowly AND pluggable demand softens” — i.e., AI-capex slowdown — which is the bear-case for the broader photonics ecosystem.

⚠ Industry consensus as of April 2026 sits closer to Scenario B (slow CPO ramp), driven by the technical-yield, thermal-management, and serviceability problems on first-generation CPO. Hyperscalers are deploying CPO at limited scale alongside continued pluggable volume.

4. GFS Fotonix Competitive Positioning

vs Tower Semiconductor PH18 (200mm)

Tower PH18 is the merchant SiPh foundry that has historically competed most directly with Fotonix:

  • Wafer size: 200mm (legacy) → 300mm (standard offering announced 2024-11-26, Tower press release). Tower operates two 200mm SiPh fabs in the US plus one 300mm SiPh fab in Japan.
  • Volume: Tower announced its capacity would double in 2025 with continued expansion in mid-2026. ◐ Tower CPO Foundry Tech announcement, 2025-11-12
  • Process differentiation: Best-in-class silicon waveguides + low-loss silicon nitride waveguides on the 300mm offering; industry-standard OSAT compatibility.
  • vs Fotonix: Tower lacks Fotonix’s monolithic CMOS-photonics integration on 300mm. Fotonix integrates SiGe drivers + photonic devices on the same die; Tower typically requires off-chip drivers.

vs TSMC Silicon Photonics (internal)

TSMC operates a SiPh process internally that supplies Nvidia, Marvell, and other top-tier AI customers under capacity-allocated arrangements. TSMC SiPh is not openly merchant; it is an allocated-capacity offering for top-tier strategic customers.

  • For the merchant-fabless photonics design ecosystem (LWLG, NLM, Lightmatter, PsiQuantum, Ranovus, Ayar Labs), TSMC is largely inaccessible as a foundry partner.
  • GFS Fotonix is the merchant-friendly alternative; this is structurally why the AI-photonics design-house ecosystem is concentrating its design-wins on GFS.

vs Intel Silicon Photonics → Jabil

Intel’s SiPh business was divested to Jabil in 2024. The post-divestiture entity continues to supply a residual customer base but is not a major share-taking competitor for new design wins. ◐ [Industry consensus]

vs IMEC / AIM Photonics

IMEC (Belgium) and AIM Photonics (Albany NY) operate research-and-prototyping SiPh foundries. They are critical for early-stage prototyping but are NOT volume-production alternatives. The transition path is typically: IMEC / AIM Photonics prototype → GFS Fotonix or Tower volume production.

Net competitive position post-AMF (Nov 2025)

After the AMF acquisition:

  • GFS becomes the largest pure-play SiPh foundry by revenue (TrendForce).
  • GFS gains a “China-free” supply-chain narrative — important for hyperscaler customers post-BIS export controls; Singapore is geographically diversified from both Taiwan and China.
  • AMF’s customer base consolidates into GFS — closes off AMF as a Plan B foundry option for any photonics design-house that previously had AMF + GF as dual-source. POET Technologies in particular had AMF in some triangulation paths; that is now GFS.

Customer roster (Fotonix tier)

The disclosed Fotonix customer base includes (per SemiAnalysis — GFS Fotonix):

  • Hyperscaler-aligned: Marvell (DSP + optics), Nvidia (some volumes), Cisco, Broadcom.
  • Merchant photonics design-houses: Ayar Labs (TeraPHY), Lightmatter (photonic AI accelerators), PsiQuantum (photonic-quantum), Ranovus, MACOM (some volumes).
  • EDA / design partners: Ansys, Cadence, Synopsys.
  • Modulator / IP vendors: Lightwave Logic (March 2026 GDSFactory PDK availability for validation tape-outs), NLM Photonics.
  • Acquired AMF customer base: 15 years of pre-existing AMF customer relationships across Telecom / Data Center / LiDAR / Sensors verticals.

The customer-roster breadth is meaningful for two reasons:

  1. Diversification: No single customer concentration risk like in the smart-mobile or auto segments.
  2. Strategic optionality: Multiple paths to upside if any one design-house’s design-wins materialize. The PsiQuantum quantum path is wholly different from the Marvell hyperscaler path; both are GF revenue streams.

5. Reading the Silicon-Photonics Market Signals

For an analyst tracking GFS’s photonics-foundry position, the highest-value signals to monitor:

  • Yole / LightCounting / Dell’Oro quarterly photonics-market updates — independent measurement of SiPh module / chip volumes.
  • Hyperscaler CPO ramp commentary (Broadcom Tomahawk-Ultra, NVIDIA Spectrum-X CPO, Microsoft / Google / AWS deployment cadence).
  • Module-maker capacity updates (Innolight, Eoptolink, Coherent, Lumentum) — direct downstream demand signal.
  • GFS quarterly earnings commentary on Fotonix design-win momentum — currently soft (no segment break-out); track for any disclosure-policy change.
  • Ayar Labs volume-shipment commentary — Ayar Labs is the closest to commercial-volume merchant-optical-engine customer of Fotonix; their ramp is a leading indicator for merchant-channel monetization.
  • Lightwave Logic + GFS commercial-PDK transition — March 2026 validation-tape-out announcement is a precursor to (potential) commercial PDK release. Track LWLG 6-K + GF press release cadence.
  • Tower Semiconductor SiPh capacity announcements — competitive-dynamic signal; if Tower keeps expanding faster than GFS Fotonix, the share-leader position is contestable.

The single most-important leading indicator: the first GFS-disclosed silicon-photonics revenue or wafer-volume metric. Currently no such disclosure exists; the first such disclosure (likely in a 20-F segment-mix update or investor-day presentation) would be a thesis-validating event.

Cross-section pointers

  • ./ai_capex_cycle — Hyperscaler-capex tailwind that drives this market’s growth.
  • ./foundry_industry_dynamics — Pure-play foundry market structure context for the SiPh sub-segment.
  • ./tam_sam — Bottom-up GFS Fotonix SAM with sensitivity analysis.
  • overview — Fotonix process / 45CLO / 9WG technical detail; the supply-side view.
  • overview — Customer roster (Marvell, Ayar Labs, LWLG, NLM, etc.); the customer-by-customer detail referenced above.