Is CPO or LPO the winning near-term category?

LPO is the 2025-2027 bridge (400G/800G pluggables at ~50% lower power). CPO is the 2027-2030 endgame for 1.6T and 3.2T switch ASICs. Startups typically pick one but design for the transition.

Do I need to own a fab?

No. GlobalFoundries Fotonix, TSMC COUPE, Tower, and IMEC iSiPP offer commercial photonics PDKs. Fabless is now the default model, mirroring the digital ASIC playbook.

Strategic acquisition by NVIDIA, Broadcom, Marvell, AMD, Intel, or Cisco is the base case. IPO for category leaders (Lightmatter, PsiQuantum have signaled IPO tracks). Hyperscalers occasionally acqui-hire for internal ASIC teams.

Silicon Photonics & Co-Packaged Optics Fundraising Guide

How silicon-photonics, co-packaged optics, and optical-interconnect startups raise capital in 2026 as hyperscaler AI clusters bottleneck on electrical I/O.

Raising Capital for Silicon Photonics & Optical Interconnect Startups

AI training clusters have outgrown copper. Every hyperscaler is racing to co-packaged optics (CPO), linear-drive pluggable optics (LPO), and photonic switches to break the power/bandwidth wall between GPUs. Ayar Labs, Lightmatter, Celestial AI, xLight, and PsiQuantum have collectively raised >$3B on the thesis that optical I/O is the next data-center substrate. NVIDIA's Quantum-X and Spectrum-X CPO switches, Broadcom's Bailly, and TSMC's COUPE roadmap turned photonics from research into scheduled 2025-2027 product cycles. Investors want proven wafer-scale integration, foundry partnerships (GlobalFoundries Fotonix, Tower, TSMC), and a signed hyperscaler or optical-transceiver OEM design win.

Why 2026 is different

Hyperscaler capex hit >$400B/yr, with the training-cluster bottleneck shifting from GPUs to interconnect bandwidth and power. NVIDIA announced Quantum-X CPO shipping 2025 and Spectrum-X CPO in 2026. Broadcom's Bailly CPO switch is in production sampling. Microsoft, Meta, and Google published multi-year CPO roadmaps. The linear-drive pluggable (LPO) category exploded as a bridge technology. TSMC's COUPE and GlobalFoundries' Fotonix opened commercial PDKs, letting fabless photonics startups tape out at foundry rather than build custom fabs.

Realistic capital stack

Seed: $5-25M for team + first tape-out. Series A: $30-100M for characterization + reference design. Series B: $100-300M for design wins + reliability qualification. Series C+: $200M-$1B for volume ramp. Reference: Ayar Labs ($155M C at ~$1B), Lightmatter ($400M D at $4.4B), Celestial AI ($175M C at $2.5B+), PsiQuantum ($1B+ raised, ~$6B), xLight ($40M A), Nubis, Nvidia-backed Enfabrica ($125M B).

Common failure modes

No foundry PDK path — bespoke fabs kill unit economics. Ignoring laser sourcing (external DFB vs. heterogeneous III-V integration) and reliability. No hyperscaler or transceiver-OEM design-in by Series B. Underestimating packaging, fiber attach, and thermal engineering (often >50% of BOM). Racing the incumbents (Intel Silicon Photonics, Cisco Acacia, Marvell Inphi) without a differentiated power or density story.

Frequently asked questions

Is CPO or LPO the winning near-term category?: LPO is the 2025-2027 bridge (400G/800G pluggables at ~50% lower power). CPO is the 2027-2030 endgame for 1.6T and 3.2T switch ASICs. Startups typically pick one but design for the transition.
Do I need to own a fab?: No. GlobalFoundries Fotonix, TSMC COUPE, Tower, and IMEC iSiPP offer commercial photonics PDKs. Fabless is now the default model, mirroring the digital ASIC playbook.
Realistic exit?: Strategic acquisition by NVIDIA, Broadcom, Marvell, AMD, Intel, or Cisco is the base case. IPO for category leaders (Lightmatter, PsiQuantum have signaled IPO tracks). Hyperscalers occasionally acqui-hire for internal ASIC teams.

Related fundraising verticals (40)

Investor directory · Fundraising library · Articles A–Z · Company funding database