GF photonics customers — per-customer profile

Updated: 2026-04-29 Status: ✓ Customer set verified via GF press releases, customer co-disclosures, and SemiAnalysis aggregation. Volume estimates flagged ⚠ where not publicly disclosed. Cross-references: Customer roadmaps · Competitors · fotonix pdk customers

1. Customer-set summary

GF’s photonics customer set partitions into:

Modulator-IP material partners — Lightwave Logic (Perkinamine polymer), NLM Photonics (Selerion silicon-organic hybrid), and the Marvell-owned Polariton plasmonic-organic hybrid (POH)
High-volume merchant transceiver / chiplet customers — Marvell (post-Inphi optical-transceiver portfolio + post-Polariton modulator IP), Ayar Labs (TeraPHY UCIe optical chiplet), Lightmatter (photonic-electronic AI accelerator), Ranovus (CPO chiplets and photonic IP)
Quantum / specialty integrators — PsiQuantum, Xanadu (low-loss waveguide library for photonic quantum computing)
Strategic captive-volume customers — NVIDIA, Broadcom, Cisco, Macom (named in GF’s launch ecosystem; specific Fotonix usage less publicly granular)
Inherited AMF customer book — undisclosed Asia-Pacific customer set from the Nov 2025 acquisition; NLM Photonics is the only confirmed transitioning customer

The two structurally most important customer types are (a) the modulator-IP material partners, because they bring novel performance unlocks that GF cannot replicate organically, and (b) the high-volume merchant customers, because they fund Fotonix capacity utilization.

2. Lightwave Logic — anchor EO-polymer integrator

Field	Value
Public engagement start	2024-05-13 (LWLG-AMF announcement, sub-1V drive at 200 Gbps PAM4); reframed as GF after Nov 17 2025 AMF acquisition
Commercial PDK live	2026-03-16 — GDSFactory PDK for Fotonix
Process	Fotonix 45SPCLO with slot-waveguide structure optimization + LWLG Perkinamine polymer BEOL fill
Deal type	Co-development; GF process modification + LWLG material supply
Performance targets	200 G/lane validation tape-out 2026; 400 G/lane next
Volume estimate	⚠ Not yet at volume; LWLG targets 2027 commercial revenue
Risk-to-GF if customer pulls	Low-medium: GF has parallel NLM Photonics SOH path (different chemistry); but losing LWLG would surrender first-mover positioning and slow EO-polymer Fotonix moat by 12-18 months
Source	LWLG-GF GDSFactory release (2026-03-16) ✓ ; LWLG-AMF release (2024-05-13) ✓

LWLG is GF’s most strategic photonics customer in the modulator-IP layer. The PDK is live on Fotonix (Malta NY) via GDSFactory, with slot-waveguide process modifications confirmed in progress. LWLG’s Tier-1 customer relationships (4 Fortune Global 500 customers in Stage 3 per LWLG disclosure; see platform overview) flow ultimately to GF capacity: every LWLG-modulator port shipping in volume implies a Fotonix wafer slot.

⚠ Royalty / NRE structure between GF and LWLG is not public.

3. NLM Photonics — first-gen on AMF, second-gen on AMF HP

Field	Value
Public engagement start	2026-03-16 (NLM SOH PIC sampling at GF)
Process (1st gen)	AMF GP (200mm Singapore, post-acquisition GF)
Process (2nd gen)	AMF HP; HP PDK 6.0 in development
Deal type	Foundry manufacturing services + PDK co-development + technology validation
Material	Selerion-HTX organic electro-optic polymer (r33 = 1000 pm/V class per June 2025 disclosure)
Performance targets	110+ GHz bandwidth, 40% smaller form factor than typical 1.6T PICs; 1.6T / 3.2T product class
Volume estimate	⚠ Pre-volume; sampling phase as of Apr 2026
Risk-to-GF if customer pulls	Low: parallel LWLG path; AMF capacity addressable by other customers
GF executive lead	Michael Reiha, VP Silicon Photonics Business
Source	NLM Photonics SOH PIC at GF (2026-03-16) ✓

NLM is the second-mover EO-polymer integrator on GF, structurally analogous to LWLG but on a different process line (AMF Singapore, not Fotonix Malta) and with different material chemistry (Selerion vs Perkinamine). The strategic value to GF: NLM validates the AMF acquisition’s commercial integration within four months of close, and establishes that GF has two parallel polymer-integration paths — reducing single-customer concentration risk.

4. Marvell — post-Inphi SiPh + post-Polariton POH integration target

Field	Value
Engagement type	Productized PDK + custom-foundry agreements; volume-strategic
Current product set	Inphi-lineage 100/200/400 G PAM4 SiPh transceivers + Marvell-architected silicon-photonics light engine for AI fabric
Post-Polariton path	Plasmonic-organic-hybrid (POH) modulator integration following 2026-04-22 Polariton acquisition; 12-18 month integration timeline per Marvell
DSP node	TSMC 5/3/2 nm (separate die — not GF)
Photonic die node	GF Fotonix (with planned process modifications for POH)
SiGe BiCMOS	Historically separate die; increasingly monolithic on Fotonix
Volume estimate	⚠ Substantial — Marvell is one of the highest-volume Fotonix customers, but specific wafer commitments not public
Risk-to-GF if customer pulls	High: Marvell is largest single transceiver-volume buyer in the photonics customer set; Marvell shifting to Tower PH18 / TSMC SiPh would materially alter Fotonix utilization
Source	Marvell Optical DSP product page ✓ ; Marvell-Polariton acquisition (2026-04-22) ✓ ; SemiAnalysis Fotonix ◐

Marvell is GF’s most consequential single customer in the photonics layer. The Polariton acquisition (April 2026) put Marvell in control of:

Optical DSP (TSMC 5/3/2 nm) — Ara, Ara X, Spica, Petra family
Silicon-photonics modulator + photodetector (GF Fotonix) — Inphi-lineage and forward
Plasmonic-organic-hybrid modulator IP (Polariton, Zurich) — 1.0 THz BW class research; 3.2T+ commercialization target

The implications for GF:

Marvell now owns the modulator IP, making the Polariton-Fotonix integration a commercial-priority project for both companies
The process modifications needed for POH (plasmonic metal slot waveguides + EO polymer fill) are similar to (but distinct from) the LWLG slot-waveguide modifications; GF’s R&D bandwidth has to support both
Marvell’s preference between TSMC’s hybrid SiPh + CoWoS path vs GF’s monolithic Fotonix path is the load-bearing variable for Fotonix volume scaling

Cross-reference: the Marvell KB (polariton acquisition) treats this as accelerating commercialization 12-18 months vs licensing/partnership; the Marvell KB also notes (foundry relationships) that GF is currently “<5% of Marvell wafer spend” — a number that will need updating as Polariton-Fotonix integration matures.

5. Ayar Labs — TeraPHY UCIe optical chiplet

Field	Value
Product	TeraPHY optical I/O chiplet, UCIe-compliant; SuperNova 16-wavelength external comb laser
Throughput	4 Tbps current; 8 Tbps next-gen UCIe construct (announced 2025-03-31)
Process	Fotonix (likely 9WG variant ⚠ to confirm specifics)
Sampling status	”Thousands of engineering samples” shipping per Contrary Research ◐
Volume timeline	Commercial offerings 2026-2028
Strategic backers	NVIDIA, Intel, Microsoft, AMD-investment per public disclosures
Risk-to-GF if customer pulls	Low-medium: TeraPHY is the highest-profile chiplet customer; loss would signal chiplet-architecture migration to TSMC SiPh / Tower PH18
Source	Ayar Labs UCIe announcement (2025-03-31) ✓

Ayar Labs has the strongest indirect hyperscaler exposure of any GF photonics customer — Microsoft, NVIDIA, and Intel are all backers / customers. Every TeraPHY shipment = a Fotonix wafer slot. The volume timeline (2026-2028) is the most actionable forward indicator for Fotonix utilization growth.

6. Lightmatter — photonic-electronic AI accelerator

Field	Value
Product set	Mars / Envise / Passage photonic-electronic AI inference accelerators
Process	9WG (volume); 45CLO products in development per SemiAnalysis ◐
Volume estimate	⚠ Currently in production volume on 9WG
Strategic position	Direct AI-silicon competitor to NVIDIA / Cerebras / Groq / TPU; differentiated by photonic interconnect architecture
Risk-to-GF if customer pulls	Low: replaceable wafer demand from many sources
Source	SemiAnalysis Fotonix article ◐

Lightmatter is the highest-profile dedicated photonic-AI customer at GF. The 9WG → 45CLO transition is the technology-roadmap path that brings them closer to LWLG / NLM polymer-integration access.

7. Ranovus, PsiQuantum, Xanadu, Enosemi

Customer	Use case	Status
Ranovus	CPO chiplets + photonic IP licensing	Productized customer
PsiQuantum	Photonic quantum computer; fault-tolerant cluster states	R&D → pilot production on Fotonix low-loss waveguide library
Xanadu	Continuous-variable photonic quantum computer	R&D → pilot production on Fotonix
Enosemi	Silicon-validated electronic-photonic design IP for Fotonix	IP licensing partner; design enablement, not direct tape-out customer

These customers are individually low-volume but collectively significant — they validate Fotonix’s reach beyond datacom into chiplet-architecture, AI accelerators, and photonic quantum computing.

⚠ Specific volume / revenue contribution from each is not public.

8. NVIDIA, Broadcom, Cisco, Macom — strategic captive-volume customers

These four are named in GF’s 2022 Fotonix launch press release as ecosystem partners. Public detail is sparser than for the modulator-IP and chiplet customers above.

8.1 NVIDIA

Engagement: Optical interconnects for HPC/AI networking via post-Mellanox photonic IP roadmap
Fotonix usage: One of multiple SiPh foundry options NVIDIA evaluates; specific wafer commitments not public
Strategic risk: NVIDIA is also a backer of Ayar Labs (chiplet path); NVIDIA’s direct-to-Fotonix engagement details ⚠ not publicly granular

8.2 Broadcom

Engagement: Tomahawk-class CPO is the highest-volume CPO product line in 2024-2025
Fotonix usage: Broadcom uses internal SiPh IP + multiple foundry options; specific Fotonix vs internal-fab vs TSMC routing not fully public
Strategic risk: Broadcom’s internal IP gives it foundry-routing flexibility; less captive than Marvell

8.3 Cisco

Engagement: Acacia + Luxtera lineage SiPh IP integration
Fotonix usage: Cisco’s photonic IP is internal-use-dominated; merchant pluggable share less Fotonix-dependent
Strategic risk: Cisco’s volume is bounded by Cisco’s own switch/router product line

8.4 Macom

Engagement: Datacom modulator drivers; could be either direct Fotonix customer or CMOS supplier integrated into Fotonix-fabbed photonic engines
Fotonix usage: ⚠ Specific role not fully primary-sourced
Strategic risk: Macom’s diversification across multiple foundries reduces concentration risk

For the strategic captive-volume customer set, the SemiAnalysis source ◐ is the best aggregator citation.

9. Hyperscaler indirect exposure

GF does not sell directly to hyperscalers (Microsoft, Google, Meta, Amazon, Oracle); the hyperscalers buy transceivers and CPO from Marvell / Broadcom / Coherent / Lumentum / Ayar Labs, who buy photonic dies from GF. The chain:

Microsoft → Ayar Labs UCIe optical chiplets → GF Fotonix wafers
Microsoft / Meta / Google → Marvell optical transceivers → GF Fotonix photonic die
Oracle / Amazon / Meta → Broadcom CPO → GF Fotonix wafers (partial)
All hyperscalers → Coherent / Lumentum (less GF-dependent — these competitors run more vertically integrated stacks)

This indirect-exposure structure means hyperscaler capex commits flow into Fotonix utilization with a 12-24 month lag — which aligns with the AI-cluster CPO roadmap forecast (1.6T volume 2026-2027, 3.2T 2028+).

10. Customer-set risk concentration

Risk type	Severity	Comment
Single-customer (Marvell) revenue concentration	High	Marvell is the largest single Fotonix customer; loss to TSMC SiPh would materially impact utilization
Modulator-IP-partner deselection (LWLG or NLM)	Medium	Two parallel paths; loss of either is recoverable
Chiplet-architecture migration (Ayar Labs class)	Medium	If chiplet wins broadly, GF still wins as the chiplet’s photonic die fab — but only if customer doesn’t migrate to Tower / TSMC
Hyperscaler capex slowdown	Low-medium	AI-cluster spend has only one direction in 2025-2028; cyclical risk in 2029+
Geopolitical (Taiwan disruption boosts US-domestic preference)	Low (positive for GF)	GF’s US + Singapore footprint is competitively advantaged in disruption scenarios

11. Open audit items

⚠ Wafer-volume per customer — GF does not publicly disclose; modeling work needed using customer revenue × estimated photonic-die share × estimated wafer cost.
⚠ Marvell-Fotonix POH process customization timeline — Polariton acquisition Apr 22 2026; integration timeline not yet disclosed.
⚠ NLM-Fotonix vs NLM-AMF — second-gen path explicit on AMF HP; whether NLM also runs on Fotonix Malta unconfirmed.
⚠ AMF inherited customer book — names and volume not publicly itemized.
⚠ Specific Fotonix node usage by Lightmatter, NVIDIA, Broadcom, Cisco — 9WG vs 45SPCLO breakdown not fully public.
⚠ Royalty / NRE / license structures with LWLG and NLM — not in public disclosures.

12. Cross-references

Customer roadmaps — timeline integration
Customers — diversified — non-photonics customer concentration (cross-context)
Competitors — what happens if a customer migrates away
fotonix pdk customers — technology-side customer detail
amf singapore acquisition — NLM-on-AMF context