2026-05-15 — views $CRDO · Credo Technology · HiWire AECs · SerDes IP · Optical DSP
Credo (CRDO) — AECs displaced optical at 800G. The 1.6T fight is the next bottleneck.
CRDO ships Active Electrical Cables (AECs) that replaced expensive optical interconnects at 800G — Amazon is the concentrated customer. The open question: does the AEC cost curve hold at 1.6T as optical regains ground?
Credo Technology (NASDAQ: CRDO) is the other public pure-play in AI cluster interconnect IP — and while Astera Labs sits inside the server (PCIe), Credo sits between racks (Ethernet, 800G+). The company’s flagship product, the HiWire Active Electrical Cable (AEC), did something rare for a connector category: it took meaningful share from optical interconnects at 800G in hyperscale AI buildouts. That was the story of 2024 and 2025. The story of 2026 is whether that share holds at 1.6T.
Three product lines, but one moves the needle
| Line | Function | Status |
|---|---|---|
| HiWire AECs | Active Electrical Cables — copper + embedded SerDes DSP | Volume driver; Amazon is the disclosed concentrated customer |
| SerDes IP | High-speed serializer/deserializer IP licensed to other chipmakers | Steady revenue, low growth |
| Optical DSPs | Signal processing for optical transceivers | Growing as 1.6T ramps |
AECs are the headline. The other two are real businesses but the bull case is AECs and the bear case is AECs — so that’s where the analysis lives.
Why AECs beat optical at 800G
Inside a hyperscale rack at 800G per port, you have three connection technology choices:
- Direct Attach Copper (DAC) — cheapest, but reach is limited (~3m at 800G) and signal integrity is marginal at the edges of the spec.
- Active Optical Cable (AOC) — fiber + integrated optics — long reach, but expensive and power-hungry per port.
- Active Electrical Cable (AEC) — Credo’s HiWire — copper cable with a small DSP embedded at each end that cleans up the signal. Cheaper than AOC, longer reach than DAC, lower power than AOC.
At 800G, the AEC sweet spot is real: 2–7m intra-rack and inter-rack reach at 30–50% the cost of an AOC equivalent. Amazon validated the architecture early; that scale-up to volume is what built Credo’s 2024–2025 revenue ramp.
The 1.6T fight — why this is the interesting moment
At 1.6T per port (the next-gen Ethernet speed already in design for 2026–2027 racks), the physics shift:
- Copper signal integrity is harder to maintain at 1.6T over the same cable lengths — the DSP work an AEC has to do gets harder and pricier.
- Optical DSPs improve, partly because Credo itself sells them — meaning Credo’s optical DSP business might offset weakness in HiWire if the AEC sweet spot shrinks.
- New optical formats (LPO — Linear-Drive Pluggable Optics, NPO — Near-Package Optics) attack the AEC cost advantage from the optical side.
The bull case for CRDO: AECs hold their cost advantage at 1.6T because the DSP economics scale with Moore’s Law while optics economics don’t. The bear case: at 1.6T, optics regain enough ground that AEC growth flattens, leaving Credo with a customer-concentrated business and slower secular growth.
The Amazon concentration risk
Credo discloses customer concentration in 10-Q filings — Amazon has historically been the largest customer at meaningful share of revenue. The right framing is not “this is a red flag” (early hyperscale customers are how an interconnect vendor gets volume), but “diversification is the milestone to watch.” When Google or Microsoft enters the disclosed top-customer set at comparable share, the concentration narrative ends.
Practitioner note
For builders ordering network infrastructure:
- At 800G, AECs are the default cost-optimized choice for 2-7m intra-rack and short-reach inter-rack connections. The math works; the volume is shipping.
- At 1.6T, run the cost-and-power model yourself. Don’t assume the 800G AEC advantage carries forward. LPO and NPO are credible alternatives and the cost curves may cross before your buildout ships.
- The SerDes IP is the quiet leverage. If you’re a custom-silicon team building an Ethernet-attached accelerator, Credo’s SerDes IP is one of three real options (alongside Synopsys and Cadence). The licensing relationship matters more than the AEC sales channel for that buyer profile.
- CRDO + ALAB is the standard pair, not an either/or. The complement is structural — ALAB owns the PCIe domain, CRDO owns the Ethernet domain. A hyperscaler running both is the modal customer, not a bake-off.
The under-considered angle: the AEC is a “stealth IP” play disguised as a cable. The cable looks like a connector category, but the value is in the DSP silicon at each end. That positioning insulates Credo from cable manufacturing margin pressure — the DSP is the irreplaceable part, and Credo controls it. Whether that holds at 1.6T is the 2026–2027 thesis test.