Credo Technology
Fabless semiconductor + IP. Sells AECs (copper cables with integrated retimer/DSP), standalone ICs (optical DSPs, line-card PHYs, PCIe retimers), SerDes chiplets, and licenses SerDes IP. ~97% product/engineering, small high-margin IP-licensing tail. Outsources wafer fab; capital-light.
The thesis on this name
State of AI Compute
AEC ramp tripled FY26 revenue at improving 68% GM and the connectivity tax sits under every fabric (architecture-agnostic) — high-potential IF optics scales before co-packaged optics commoditizes copper, but verify flagged the 3-5yr CPO substitution clock, ~22x sales, customer concentration, AND a self-guided H1-FY27 'mid-single-digit sequential' decelerati…
Earnings, margins, COGS & capex
FY26 (ended May 2 2026) revenue tripled to $1.335B (+205.7% YoY) on an AEC-led AI-data-center ramp, at 68.0% GAAP gross margin and 47.8% non-GAAP operating margin; non-GAAP net income rose ~5x to $661.5M (fact). The model is capital-light and cash-generative — >40% FCF margin, ~$1.4B net cash, no debt (fact). Growth is decelerating QoQ (Q3 +51.9% → Q4 +7.4%) as the base scales, with FY27 guided to >80% growth and an optical-product inflection in H2 (fact/guidance). The central tension: a ~68%-margin AEC franchise concentrated in one hyperscaler, racing to scale optics before co-packaged optics commoditizes in-rack copper (analysis).
Income statement — where each revenue dollar goes
% of revenueOf every $1 of revenue, ~32¢ is cost of goods and ~35¢ operating expense, leaving ~33¢ of operating profit (~50¢ net).
Revenue trend
Margins
up — vs ~67% FY25 (fact)
up — operating leverage on 3x revenue (fact)
up (fact)
up — ~5x net income growth (fact)
up — $177.5M FCF on $437M rev (fact)
COGS structure
COGS is dominated by outsourced wafer fabrication, packaging/OSAT, and — for AECs — cable assembly and connector/component content (fabless, so no owned-fab depreciation). Gross margin sits ~68%; product mix is the main swing factor — higher-margin IC/optical-DSP and IP-licensing revenue lifts GM, while cable/hardware-heavy AEC content is comparatively lower-margin. Yield, advanced-node wafer pricing, and one-time production mask sets also move COGS (fact/analysis).
Capex
Minimal — fabless model means capex is well under 1% of revenue, mostly lab/test equipment and lumpy production mask sets (~$26.5M in Q3 FY26 for masks). The real 'investment' is R&D opex and M&A: the ~$1.3B DustPhotonics acquisition ($750M cash + ~0.92M shares upfront, up to ~3.21M contingent shares) funds the optical/silicon-photonics buildout (fact).
Latest earnings
Beat — prior quarter (Q3 FY26) non-GAAP EPS $1.07 beat ~$0.94 consensus (+~14%) and revenue topped by ~5%; Q4 also beat forecasts per the earnings call (fact).
Q1 FY27: revenue $465-475M, non-GAAP GM 67.0-69.0%, non-GAAP opex $86-90M. FY27: >80% YoY revenue growth, mid-single-digit sequential H1 then H2 inflection, non-GAAP GM ~consistent with FY26 (~68%), non-GAAP net margin ~50%; optics >$600M (fact/guidance).
- FY26 non-GAAP net income
- $661.5M (~5x YoY)
- FY26 non-GAAP diluted EPS
- $3.46 (GAAP $2.51)
- Cash + ST investments
- ~$1.4B, no debt
- Customer concentration
- 1 customer ~67% of FY25 rev; top-10 ~90%
Growth drivers
- AEC (HiWire) ramp for AI scale-up/scale-out — the engine that tripled FY26 revenue (fact).
- Optical inflection — ZeroFlap optics + SiPho PICs + optical DSP each guided >$100M (>$600M total) in FY27 (guidance).
- PCIe Gen6 retimer design wins converting to revenue in FY27 (fact/guidance).
- New multi-$B vectors: Active Line Cards and OmniConnect memory/scale-up (estimate/guidance).
- Customer diversification across hyperscalers + neoclouds expanding the served base (guidance).
- Rising bandwidth-per-rack (800G→1.6T→3.2T) increasing connectivity content per system (analysis).
Reported financials — SEC EDGAR
Audited GAAP figures pulled from SEC filings · latest filing 2026-06-15. The audited primary-source spine — not financial advice.
Revenue — annual (GAAP)
Margins & balance sheet — FY’26
Bull & bear
Credo is the toll-collector on AI-data-center connectivity — it sells the link layer under every fabric regardless of which accelerator wins, and is parlaying AEC dominance into a full electrical-plus-optical stack just as bandwidth-per-rack explodes. FY26 tripled revenue at ~68% GM and ~48% non-GAAP op margin with >40% FCF and net cash; FY27 guided >80% growth with optics inflecting.
- AEC is a real, owned category with a power/cost edge inside the rack; copper isn't going away for sub-7m scale-up this cycle (analysis/fact).
- Architecture-agnostic — wins whether NVIDIA, AMD, or custom hyperscaler ASICs take share (analysis).
- Optical leg (ZeroFlap, SiPho via DustPhotonics, optical DSP) adds >$600M FY27 TAM and answers the CPO-substitution bear (fact/guidance).
- Best-in-class unit economics: ~68% GM, ~48% non-GAAP op margin, >40% FCF, ~$1.4B net cash funds M&A without dilution stress (fact).
- Long runway: PCIe Gen6 retimers, Active Line Cards, OmniConnect memory are incremental multi-$B vectors (estimate/guidance).
You are paying ~107x earnings for a single-product, single-end-market name where one customer was 67% of revenue and the core copper franchise faces a 3-5yr CPO substitution clock against better-resourced rivals. Any AI-capex digestion, a hyperscaler design loss, or optics slipping turns a hyper-growth story into a de-rating.
- Customer concentration is existential: ~67% from one customer (FY25 fact) — one re-design or in-sourcing event re-rates the stock.
- Co-packaged optics could commoditize/displace in-rack copper before Credo's optics scale, eroding the AEC moat (analysis — the flagged 3-5yr clock).
- Valuation (~107x P/E, ~$50B cap on $1.3B revenue) demands flawless execution; QoQ already decelerated to +7.4% in Q4 FY26 (fact).
- Optical DSP is Marvell/Broadcom's turf — Credo is the challenger there, not the incumbent (analysis).
- Pure AI-infrastructure exposure with no counter-cyclical ballast; a capex air-pocket hits hard (analysis).
What it is worth
Forward growth-adjusted multiple (P/E and EV/S) cross-checked vs FY27 guidance; reverse-DCF sanity on the ~$50B cap.
~$184
a hyperscaler design loss / capex air-pocket or CPO scare triggers a growth-multiple de-rate (analyst low; estimate).
~$270
in line with the ~$270 average 12-mo target; FY27 delivers ~80% growth as guided (fact).
~$350
optics hits >$600M FY27, concentration eases, multiple holds on >80% growth (analyst high; fact/estimate).
At ~107x trailing / ~$50B cap on $1.3B revenue, the stock already discounts years of >50% compounding — execution-perfect pricing; the swing factors are optics ramp + concentration de-risking vs the CPO clock.
SWOT
Strengths
- Owns the AEC category — HiWire copper cables with integrated DSP/retimer are lower-power and cheaper than optics for sub-7m in-rack links, the dominant scale-up reach (fact).
- Architecture-agnostic 'connectivity tax' — sells into any GPU/accelerator fabric (NVIDIA, AMD, custom ASICs) regardless of who wins compute (analysis).
- Structurally high ~68% gross margin and ~48% non-GAAP operating margin with operating leverage as revenue scaled 3x (fact).
- Fortress balance sheet: ~$1.4B cash, no debt, capital-light fabless model, >40% FCF margin (fact).
- Vertically integrating optics via DustPhotonics SiPho acquisition — owns SerDes→DSP→silicon-photonics stack (fact).
Weaknesses
- Severe customer concentration — one customer ~67% of FY25 revenue, top-10 ~90% (fact) — a single hyperscaler decision swings the P&L.
- Whippy QoQ: growth decelerated from +51.9% (Q3) to +7.4% (Q4) FY26 (fact), signaling lumpy, order-driven demand.
- Optics is a newer, more competitive arena (Marvell/Broadcom-dominated) than AEC, where Credo must still prove share (analysis).
- Valuation leaves no room: ~107x trailing P/E prices in years of compounding (fact/analysis).
- Heavy reliance on AI-capex cycle; minimal diversification outside data-center interconnect (analysis).
Opportunities
- Optical inflection — ZeroFlap optics, SiPho PICs, and optical DSPs each guided >$100M (>$600M total) in FY27 (guidance).
- PCIe Gen6 retimer design wins converting to revenue in FY27 — new TAM beyond Ethernet (fact/guidance).
- New multi-$B vectors flagged by management — Active Line Cards and OmniConnect memory/scale-up (estimate/guidance).
- Customer diversification across hyperscalers, neoclouds, and enterprise lowers the 67% concentration over time (guidance).
- 1.6T / 3.2T and co-packaged/near-packaged optics from DustPhotonics expand reach as bandwidth scales (analysis).
Threats
- Co-packaged optics (CPO) commoditizing/displacing copper on a 3-5yr clock — the core substitution risk to the AEC franchise (analysis).
- Marvell and Broadcom out-resource Credo in optical DSP and can bundle/price aggressively (analysis).
- Hyperscaler in-sourcing of connectivity or a shift to a competitor's fabric (analysis).
- AI-capex digestion / cyclical air-pocket would hit a single-end-market name hard (analysis).
- Customer warrants (e.g. Amazon) create share dilution / pricing concessions tied to volume (fact/analysis).
Moats, dependencies & bottlenecks
Moats
integrated DSP/retimer copper cable strong now, exposed to CPO/optics substitution Power + cost advantage for sub-7m in-rack links; the franchise that tripled FY26 (fact).
reusable across AEC, DSP, chiplet, PCIe Core IP block licensed and embedded across the portfolio; hard-won analog/mixed-signal know-how (fact).
independent of accelerator winner Sells connectivity regardless of compute vendor; structural to the value-chain role (analysis).
sticky once qualified, but concentration cuts both ways Multi-quarter qual cycles raise switching cost; offset by 67% single-customer dependence (fact).
newly assembled, unproven at scale vs Marvell/Broadcom SerDes→DSP→silicon-photonics integration; defensive answer to the CPO bear (fact).
Dependencies
~67% of FY25 revenue from one customer Top-10 ~90%; 3-4 customers expected >10% (fact). Single design loss / in-sourcing is the dominant risk.
TSMC-class) + OSAT/cable assembly No owned fab; capacity allocation and advanced-node access gate supply (analysis/fact).
Effectively the entire end market; a capex digestion hits a single-vector name hard (analysis).
AEC franchise durability hinges on optics not commoditizing in-rack copper too fast (analysis — flagged 3-5yr).
AI-interconnect demand and end-customer geography exposed to US export rules (analysis).
Advantages
- Owns the AEC category with a defensible power/cost edge for in-rack scale-up (fact).
- Architecture-agnostic — monetizes connectivity regardless of which accelerator wins (analysis).
- ~68% gross margin and ~48% non-GAAP operating margin with strong operating leverage (fact).
- ~$1.4B net cash, no debt, >40% FCF margin — funds M&A (DustPhotonics) without balance-sheet strain (fact).
- Now spans the full electrical+optical interconnect stack post-DustPhotonics (fact).
Weaknesses
- One customer ~67% of FY25 revenue; top-10 ~90% — extreme concentration (fact).
- Single end market (AI data-center interconnect) with no counter-cyclical ballast (analysis).
- Challenger, not incumbent, in optical DSP vs Marvell/Broadcom (analysis).
- QoQ growth already decelerating (+51.9% Q3 → +7.4% Q4 FY26) — lumpy order pattern (fact).
- Valuation (~107x P/E) leaves no margin for execution error (fact/analysis).
- Core copper franchise carries a 3-5yr CPO-substitution overhang (analysis).
Bottlenecks
- Customer concentration — one customer ~67% of revenue caps how much of the P&L Credo actually controls (fact).
- Foundry/advanced-node capacity allocation as a fabless vendor in a tight AI-supply market (analysis).
- Optical-product scaling — must ramp ZeroFlap/SiPho/optical-DSP to hit the >$600M FY27 optics target on time (guidance).
- Qualification-cycle length — multi-quarter design-in/qual gates new-customer diversification (analysis).
- Engineering talent in high-speed SerDes/silicon-photonics is scarce and contested (analysis).
Top signals & trends
Top signals
Diversification is the single biggest de-risking lever; watch quarterly (fact).
Each guided >$100M; on-track ramp validates the anti-CPO hedge (guidance).
Q4 already +7.4% QoQ; a stall before the H2 inflection would spook the multiple (fact/guidance).
New TAM beyond Ethernet AEC; proof of portfolio breadth (fact/guidance).
Faster co-packaged-optics displacement of in-rack copper shortens AEC's runway (analysis).
Execution on the $1.3B optics acquisition de-risks the substitution bear (fact).
Trends
More links + higher speeds per GPU rack = more connectivity content per system (analysis).
Threatens AEC long-term but is exactly why Credo bought SiPho optics — hedged both ways (analysis).
Credo is architecture-agnostic, so more bespoke fabrics still need its SerDes/DSP/AEC (analysis).
Opens retimer and OmniConnect memory TAM beyond Ethernet (fact/guidance).
Drives the boom but deepens Credo's customer-concentration risk (analysis).
Credo/DustPhotonics, Marvell, Broadcom all vertically integrating — raises the competitive bar (fact/analysis).
Ecosystem & competitor graph
Suppliers feed the company; customers pull from it. Line thickness shows the strength of each tie (supply-chain dependency, customer earnings contribution). Hover to isolate a tie.
Fabless Credo outsources wafer fabrication; advanced-node capacity is the key supply gate (analysis).
Assemble AECs and package ICs/chiplets; gate AEC throughput (analysis).
Israeli SiPho PIC developer; now in-house, supplies the silicon-photonics layer for optics (fact).
Design-tool and IP dependencies for 28G-224G SerDes development (analysis).
Hyperscaler customer; Credo issued Amazon a warrant tied to purchases — among the largest accounts (fact/analysis).
Collaborated on HiWire Switch AEC; hyperscaler data-center customer (fact).
Meta, Google-class, CoreWeave-type) 3-4 customers expected >10% of revenue; concentrated hyperscaler base (fact/analysis).
Integrate AECs and PHYs into AI racks and top-of-rack switches (analysis).
Incumbent leader in optical DSP/PAM4 and broad data-center connectivity; the primary optical threat as Credo expands into optics.
Scale leader in SerDes/PHY, optical DSP, and switch silicon; can bundle and out-resource on price.
Closest niche peer — PCIe/CXL retimers and smart-cable modules; overlaps in connectivity for AI racks.
Challenger in PAM4 DSP / interconnect; smaller, less concentrated in AI but competes on signal-integrity ICs.
Analog/RF and optical components; overlaps in high-speed connectivity and optical front-ends.
Passive/direct-attach copper and connector incumbents; the low-cost alternative AECs displace at longer reach.