Skip to main content

CPO & Silicon Photonics — Products

Updated 6/19/2026

Engine-synthesised product landscape for CPO & Silicon Photonics, ranked by trend signal across hiring, capital, orders, and discussion axes.

Last refresh: 2026-06-18.


Ayar Labs TeraPHY Optical I/O Chiplet — optical I/O chiplet

Trend: ↗ rising

Opportunity: Scale-up GPU domains beyond a single rack — copper SerDes runs out at multi-rack distances and pluggable optics burn too much power at chiplet edge; in-package optical I/O is the alternative to expand the coherent compute domain.

TeraPHY is the canonical 'optical I/O at the package' product in the topic. NVIDIA strategic check + Wiwynn 1024-GPU demo + monolithic GF process = strongest non-switch CPO play. Open question is whether merchant accelerators (vs NVIDIA's own integrated photonics) actually adopt it in volume.


NVIDIA Quantum-X Photonics CPO Switch — switch

Trend: → steady

Opportunity: Scale-up InfiniBand fabrics for AI GPU clusters need lower-power, higher-radix switches; CPO replaces pluggables to cut optical power at 100+ Tbps switch radix.

Quantum-X is the flagship CPO product in the topic — NVIDIA is the demand anchor, TSMC COUPE is the SiPh process, and Coherent/Lumentum are locked in as ELS suppliers. The architecture (multiple SiPh engines per switch ASIC + remote DFB lasers) is now the de-facto CPO reference design others must match.


NVIDIA Spectrum-X Photonics CPO Ethernet Switch — switch

Trend: · weak signal

Opportunity: Hyperscaler Ethernet scale-out fabrics need 1.6T-per-port radix without paying full pluggable-optics power budget; CPO is the bet to land it at 102.4T.

Spectrum-X CPO is the Ethernet sibling to Quantum-X and pulls the same supplier stack (Lumentum + Coherent ELS). H2 2026 shipment timing makes 2025-2026 the supplier qualification window — laser, fiber attach and packaging vendors that miss it lose two years.


Broadcom Bailly 51.2T CPO Switch (TH5) — switch

Trend: · weak signal

Opportunity: Pluggable optics power is the wall at 51.2T+ switch radix; Broadcom's 50% optical power reduction quantifies the CPO value prop hyperscalers buy on.

Broadcom is the merchant-silicon CPO leader competing with NVIDIA's vertically integrated stack. Bailly's public demo + concrete power-savings number is the case study hyperscalers cite when greenlighting CPO procurement.


Ayar Labs SuperNova Remote Laser Source — external laser source

Trend: · weak signal

Opportunity: Lasers cannot live inside the package next to a 700W GPU; remote multi-wavelength CW lasers (16λ, ~256 mW) are the architectural bet that lets in-package SiPh engines exist at all.

SuperNova is the ELS half of Ayar's stack — analogous to what Lumentum/Coherent supply NVIDIA, but vertically integrated. Validates that 'remote laser + WDM' is the CPO architecture everyone now follows.


External Laser Source (ELS) / CW DFB Laser Module for CPO — laser module

Trend: · weak signal

Opportunity: CPO removes the laser from the optical engine — that demand shifts entirely to standalone high-power CW DFB / multi-wavelength ELS modules, a brand new SKU category laser vendors are racing to own.

ELS is the most direct money-good adjacency to CPO: Coherent's InP CapEx step-up + dual-sourcing into NVIDIA = a structurally new product line. Whoever holds the InP DFB capacity wins the bottleneck for the next two CPO generations.


Celestica Mt. Diablo 1.6T Optical Interconnect Platform — 1.6T optical interconnect / switch platform

Trend: · weak signal

Opportunity: Hyperscalers want a 1.6T optical interconnect platform tuned to their own pod architecture rather than off-the-shelf — Mt. Diablo is the contract-design vehicle, with >$5K/port optical BOM signaling CPO-class content.

Celestica is positioned as the hyperscaler-custom CPO platform integrator (vs. merchant-switch path of Broadcom/NVIDIA). $5K/port + 1.6T + 2025 ship = real CPO program revenue starting to show in HPS numbers.


800G EML + Silicon Photonics Datacom Transceiver — pluggable transceiver

Trend: · weak signal

Opportunity: 800G is the volume optics SKU funding the SiPh capacity buildout that CPO ultimately needs — until CPO ships at scale, 800G EML/SiPh pluggables are paying for the fab and packaging investments.

Not a CPO product per se, but the cash-flow bridge: $250M+/quarter is what funds Coherent's InP+SiPh CapEx that then feeds CPO ELS / engines. MACOM is the embedded driver IC pick-and-shovel beneath it.


MACOM 200G PAM4 Driver (MAOM-006R0AA) for 1.6T Transceiver — driver IC

Trend: · weak signal

Opportunity: 1.6T = 8×200G PAM4 lanes; the driver IC at 200G/lane is a hard analog problem and a sole-source choke point for both pluggable and CPO transceivers.

MACOM is the embedded analog enabler — every 1.6T design, including CPO engines, needs a 200G PAM4 driver. The Tier-1 1.6T design-in is the leading indicator that MACOM rides the same CPO ramp as the optics primes.


Photonic Inference Accelerator — AI accelerator

Trend: · weak signal

Opportunity: Founder-side and practitioner-side demand for a deployable photonic compute path — the HN poster explicitly flags rack-ability and programmability as the gating problem, not the photonics.

Adjacent opportunity rather than core CPO: there is grassroots interest in photonic compute (not just photonic I/O), but no named product or fact yet. Worth watching as the demand pull for SiPh process volume beyond CPO interconnect.


Get this data as JSONLast updated: Jun 19, 2026