Credo Technology Group Holding Q4 FY26 earnings call CRDO

Management framed the entire thesis around reliability rather than raw bandwidth: as AI clusters scale from tens of thousands to hundreds of thousands of GPUs, they are increasingly constrained by the reliability and efficiency of the connectivity fabric, not compute. Credo's zero-flap AECs deliver up to 1,000x greater reliability than commodity laser-based optical modules while consuming much less power. The company positions its vertically integrated stack — SerDes, silicon, systems, firmware and telemetry software — as a foundational network-architecture partnership, with telemetry-first software that autonomously detects and mitigates link instability before it impacts the cluster.

Credo calls FY27 an inflection point for optics, guiding optical DSPs, SiPho PICs and ZeroFlap optics to each exceed $100M and total more than $600M, ramping harder in H2. On the composition of the more-than-80% full-year revenue growth, Fleming said roughly half of the absolute-dollar growth comes from the optical portfolio and roughly half from the existing copper portfolio (predominantly AECs, plus retimers). Discrete optical DSPs and SiPho PICs carry 2-digit ASPs while ZeroFlap optics carry 3-digit ASPs, making ZF Optics the expected largest single optical contributor.

The Dust Photonics acquisition closed the week before the call, adding differentiated silicon-photonics PIC technology spanning 800-gig and 1.6T with a roadmap to 3.2 terabits per second and beyond. The architecture enables simplified optical designs with substantially fewer lasers, improving reliability, power, cost and easing supply-chain laser constraints. Integrating SiPho PIC with the ZeroFlap optics platform lets Credo control a larger portion of the optical stack, enabling richer telemetry and diagnostics, and provides a direct path to CPO/NPO architectures with initial revenue expected in FY28.

The AEC business remains the core growth engine, with AECs the preferred in-rack and multi-rack solution up to 7 meters, adopted across hyperscalers and Neo clouds at both 100-gig and emerging 200-gig per lane. The PCIe Gen 6 AEC family is on track with strengthening design activity. In retimers, the Blue Heron 200-gig-per-lane device supports Ethernet, UALink and ESUN, purpose-built for scale-out and emerging scale-up networks, alongside growing PCIe Gen 6 retimer traction. Management sees AECs as a long-term grower but slower in percentage terms than optical, with room to penetrate further at 5 of 6 hyperscalers (only xAI described as fully deployed).

Active LED cables (ALC) replace traditional lasers with MicroLED to extend AEC-class reliability and power profile to row-scale optical reach up to 30 meters, with a better form factor; management sees an AEC-like path to large revenue quickly. The OmniConnect family extends inward toward silicon; the first gearbox, Weaver, targets memory bandwidth/density for next-generation inference. Using first customer Positron — whose product carries 2 terabytes of LPDDR memory, described as more than 10x any announced inference engine — management noted Weaver revenue contribution per GPU can be between $2,000 and $3,000. Both ALC and OmniConnect are expected to ramp to production in FY28.

Management expects supply-chain tightness to persist through the next year or longer. For ZF Optics, Credo owns essentially the entire bill of materials and has secured capacity commitments by leaning into investment, supporting an aggressive H2 ramp and even more than double-to-triple that the following year. On silicon, 12-nanometer is the workhorse for 100-gig-per-lane AECs; optical DSPs are transitioning to 7-nanometer; a 5-nanometer program is in flight for significant volume; and all 200-gig-per-lane products use 3-nanometer. Despite broad discussion of 3nm capacity constraints, management expressed confidence based on close, multi-year supply-partner relationships.

Q4 revenue was driven by four domestic customers each at or above 10% of revenue (34%, 27%, 16% and 10%), with the top three unchanged from Q3 and the fourth new to the 10% tier for FY26. Management expects 3-4 customers above 10% in coming quarters and increasing diversification in FY27. The emergence of the Neo cloud ecosystem — beyond the 5-6 hyperscalers — is cited as one of the most encouraging trends; these fast-moving operators prioritize network performance, reliability and time-to-deployment, and collectively could reach on the order of 20% of revenue over the coming years.

Credo ended Q4 with $1.4B in cash against approaching $200M per quarter of operating cash flow. Fleming stated no current plans to raise capital and no plans for a share buyback. The earlier ATM was framed as maximizing strategic flexibility, and the company has now done three acquisitions. The Dust Photonics deal took about $750M of net cash out the door in Q1 FY27; management remains comfortable with the cash position exiting Q1 and open to opportunistic future acquisitions, but nothing is imminent.