June 20, 2026 · 13 min read
Credo Technology is the hidden infrastructure layer of AI data centers — making the high-speed chips and cables that connect GPUs to each other and to the network. FY2026 revenue hit $1.34B, up 3× from $437M the prior year. FY2027 guidance calls for 80%+ revenue growth. Here is whether the momentum is sustainable.
Credo went from a niche semiconductor IP licensor to a $1.34B revenue company in a single fiscal year. The acceleration is not gradual — it is an inflection.
FY ends April 30. Revenue grew from $59M (Q1 FY25) to $459M (Q4 FY26) — a 7.8× increase in 8 quarters.
Credo's competitive advantage is not just a product — it is a deep stack of proprietary intellectual property that took years to develop and is tightly integrated into hyperscaler hardware designs.
SerDes (Serializer/Deserializer) IP is the fundamental building block of high-speed chip-to-chip communication. Credo has spent over a decade developing proprietary SerDes cores for 112G and 224G per-lane speeds — the fastest available. This IP is licensed directly into hyperscaler custom ASICs (network interface cards, custom AI accelerators), meaning Credo's technology lives inside the silicon of its customers' own chips. Switching costs are enormous: re-designing an ASIC takes 18–24 months and hundreds of millions of dollars.
Traditional data center short-reach connectivity used either passive copper cables (cheap but limited to ~3 meters at 400G+) or active optical modules (long reach but expensive and power-hungry). ZeroFlap Active Electrical Cables integrate a retimer chip directly into the cable connector, extending reach to ~10 meters at 400G–800G at a cost 40–60% lower than optical equivalents. For AI clusters where thousands of GPUs sit in adjacent racks, this is an enormous cost saving. Hyperscalers deployed ZeroFlap as their default rack-to-rack interconnect in FY2026 — driving the revenue inflection.
As AI clusters scale from one building to campus-wide and multi-site deployments, fiber-optic interconnects become mandatory. Credo's optical DSP chips process the analog signals from coherent optical transceivers, enabling 400G–800G per-fiber transmission over distances of 1–80km. The company is also developing Linear-drive Pluggable Optics (LPO) — a lower-power optical transceiver standard that eliminates the need for a full coherent DSP for shorter reaches — and co-packaged optics (CPO) for future switch integration. These technologies represent a $3B+ incremental TAM beyond Credo's AEC business.
The AI infrastructure buildout is universally discussed as a compute story — GPUs, TPUs, accelerators. But the compute is only as fast as the interconnects that feed it data. In a modern AI training cluster:
This means the demand curve for Credo's products is structurally tied to — and actually grows faster than — GPU shipments. Every dollar of Nvidia, AMD, or custom AI accelerator CAPEX at a hyperscaler creates downstream demand for Credo's AEC cables, SerDes IP, and optical DSPs.
The 800G transition (currently underway) specifically benefits Credo because ZeroFlap AEC is validated for 800G while passive copper cables cannot reliably operate at those speeds. This forces hyperscalers to either use optical transceivers (expensive) or ZeroFlap (cost-effective) — making Credo a beneficiary of the speed upgrade cycle, not just aggregate volume.
Credo's revenue concentration is both its greatest strength and its most significant risk. The company has achieved extraordinary scale by going deep with a small number of hyperscaler customers rather than broad with many smaller ones.
Microsoft is Credo's defining customer relationship. The Azure AI infrastructure buildout — driven by the multi-billion-dollar OpenAI partnership and Microsoft's internal AI Copilot products — has made Microsoft the largest single buyer of AI data center networking hardware on earth. Credo supplies both AEC cables and SerDes IP for Microsoft's next-generation Azure networking ASICs. This concentration worked spectacularly when Microsoft accelerated its AI CAPEX in FY2026 — but it creates a single-point-of-failure risk if Microsoft's spending shifts, if they qualify a competing AEC supplier, or if they develop internal SerDes capability.
AWS and Google have been in earlier qualification stages with Credo's products. Management has indicated both are expanding their engagement with ZeroFlap AEC and optical DSPs. If these relationships scale to even 15–20% of revenue each by FY2028, they would meaningfully diversify the revenue base while adding growth on top of Microsoft's continued spend. The qualification timelines suggest meaningful AWS/Google revenue contribution starting in FY2027.
AI-focused cloud providers — CoreWeave, Lambda Labs, Together AI, Crusoe — have emerged as a fast-growing customer cohort. These companies are building hyperscaler-equivalent AI clusters but moving faster on new technology adoption. NeoCloud revenue was a minority contributor in FY2026 but is expected to grow as the buildout continues.
Credo operates in the high-speed connectivity segment of the semiconductor market, competing with much larger companies for specific product categories.
| Company | Annual Rev | YoY Growth | Gross Margin | Primary Strength |
|---|---|---|---|---|
| Credo Tech (CRDO) | $1.34B/yr | +207% YoY | ~63% | SerDes IP + ZeroFlap AEC + Optical DSPs; AI-native design |
| Marvell Technology | $6.8B/yr | +61% YoY | ~50% | Cloud custom silicon, PAM4 PHYs, InPhi optical (acquired) |
| Broadcom (AVGO) | $51B/yr | +44% YoY | ~76% | AI XPU custom chips (Google TPU, Meta MTIA); dominant networking |
| InPhi (acq. MRVL) | N/A | Acquired | N/A | Optical interconnect pioneer; absorbed into Marvell Cloud Networking |
Note: Marvell and Broadcom compete in specific sub-segments. Neither has a ZeroFlap AEC equivalent. Marvell (via InPhi) competes in optical DSPs; Broadcom competes in switch ASICs and custom AI chips where Credo's SerDes IP may be designed-in.
Modeling CRDO to FY2028 (fiscal year ending April 2028) gives a 2-year forward view that captures the optical ramp and potential customer diversification.
Current price ~$75. Base case implies modest downside; bull case implies ~47% upside. The bear case at $35 would require a major Microsoft-driven revenue decline.
Credo Technology is a genuinely exceptional business executing at a rare pace. Tripling revenue in a single fiscal year — and guiding for another 80%+ — places it among a handful of public semiconductors in history that have achieved this trajectory. The technology moat is real: SerDes IP has multi-year switching costs, ZeroFlap AEC is the cost-performance leader for short-reach AI cluster connectivity, and the optical DSP portfolio adds an entirely new growth vector.
The bull case requires believing that Microsoft continues to scale aggressively (likely), that AWS and Google contribute meaningfully by FY2028 (possible), and that no hyperscaler begins building internal AEC capability (uncertain over a 5-year horizon). If all three hold, CRDO at $75 looks inexpensive against $110+ potential in the bull case.
The bear case is real but binary: it essentially requires a Microsoft relationship disruption. At ~70% concentration, this is not a tail risk — it is a headline risk that deserves a dedicated position-sizing discipline. A CRDO position sized for 2–4% of a portfolio captures meaningful upside while limiting the damage if the Microsoft concentration thesis breaks.
Verdict: CRDO is a high-conviction AI infrastructure pick for investors who accept hyperscaler concentration risk. At $75, the stock is fairly valued for the base case and materially cheap for the bull case. Own it, size it prudently, and monitor Microsoft quarterly spend figures as the leading indicator.