Newcomer in opticsCredo built its name in copper; it must win optical sockets against long-established transceiver and laser makers.
Integrating an acquisitionDustPhotonics' photonic chips must be folded into Credo modules and roadmaps toward 3.2T and co-packaged optics.
Physics limits
Lasers are the weak partSemiconductor lasers age and fail with heat far more often than silicon chips. With millions of optical links in a cluster, even rare failures happen daily.
Converting electricity to light costs energyEvery optical link must turn electrical bits into light and back at both ends, so optics use several times more power per link than short copper.
Silicon is poor at making lightSilicon cannot emit light efficiently, so photonic chips need separate laser chips attached, and coupling them adds loss and assembly cost.
How it works
3 parts
DSP
Cleaning the electrical signal
A signal-processing chip (Credo's Bluebird or Lark) reshapes the data from the switch so it can drive the optical parts at 200G per lane.
Photonics
Silicon chips that handle light
DustPhotonics' silicon photonics chips modulate laser light with the data and send it down fibre, built with chipmaking processes for cost and scale.
ZeroFlap
Telemetry that prevents drops
The module streams its health in-band to software on the switch, which can localise, fix or pre-empt link flaps and file precise service tickets.
Active copper cables, optical transceivers, SerDes
Credo makes the high-speed links that tie AI servers and switches together: copper cables with chips in their plugs, optical transceivers, and the signal-processing chips inside them. Revenue tripled to $1.3B in its fiscal 2026 on AI data-centre demand.
FY2026 revenue was $1.335B, up from $437M; Q4 alone matched the whole prior year.
Guides FY2027 growth above 80%, with over $600M from optical products.
Bought Israeli silicon-photonics firm DustPhotonics for about $750M plus shares, closed May 2026.