Credo
The Connectivity Tax on AI

Snapshot: Executive Summary

Credo Technology has become the public-market embodiment of the "connectivity tax" thesis on AI: as hyperscalers buy more GPUs and custom accelerators, every additional rack requires denser, faster, more reliable copper and optical interconnects to stitch silicon into systems. Credo's FY26 revenue of $1.335B (+206% YoY)¹, more than triple the prior year's $436.8M, captures that mechanic with rare purity. The company sells nothing into PCs, smartphones, automotive, or industrial; the entire revenue base is AI-back-end connectivity silicon and systems. Two hyperscalers concentrated approximately 81% of FY26 revenue (49% / 32%)², and four hyperscalers were each ≥10% of Q4 FY26 revenue. That concentration is the binary risk axis on which the thesis turns.

The setup:

• Active Electrical Cables (AECs) are the originating engine. Smart copper interconnects with embedded Credo DSPs and SerDes; rack-to-rack and inter-rack at 50G/100G/200G per lane scaling to 1.6T aggregate. Branded ZeroFlap for reliability against optics' link-flap problem. Credo invented the category and remains the dominant supplier; Marvell launched "Golden Cable" in late 2025 as a competing ecosystem.
• Optical DSPs are the next leg. 800G and 1.6T Cardinal family plus 3nm Bluebird DSP at 224G per lane. Management commentary frames >$600M of FY27 revenue from the optical portfolio³ versus a low base in FY26, the single largest mix shift in the model.
• The DustPhotonics acquisition closes the silicon-photonics gap. Closed May 28, 2026, $750M cash + 0.92M shares upfront (up to ~3.21M shares contingent)⁴. Opens 800G/1.6T/3.2T near-package optics (NPO) and co-packaged optics (CPO) exposure, the long-term displacement risk to pluggable DSPs becomes a Credo product line rather than a threat.

At ~22x FY27E sales (consensus $2.154B), CRDO trades inside the AI-infrastructure premium tier but below pure connectivity peer ALAB. The compression risk from the Q1 FY27 non-GAAP gross margin guide of 67.0%–69.0% (versus 68.3% in Q4 FY26) is real, optical mix is structurally lower-margin than AECs in the near term, before scale and ASP normalization. Reiterate Buy, 12-month PT $280 via EV/Sales scenarios (Bull $360 / Bear $145). Probability-weighted blended fair value ~$270, within rounding of the headline target. Customer concentration is the headline risk; multiple compression on any execution miss is the second-order concern.

Tactical: CRDO is trading at $250.81, ~11.6% upside to our $280 12-month target, anchored on FY27 revenue exceeding $2.15B (mgmt commentary >80% YoY), >$600M optical contribution, non-GAAP gross margin holding 67–69%, and continued hyperscaler diversification (target ~20% non-hyperscaler mix). The Q4 FY26 print on June 1, 2026 produced a 12–15% after-hours drawdown despite a clean beat, illustrative of how lofty expectations have become and how thin the margin for error is. Pairs analytically with the MRVL custom-silicon view. Rating Buy.

Investment Thesis

The investment thesis reduces to three orthogonal questions: (1) does the optical DSP portfolio inflect past $600M in FY27 as management has effectively guided, (2) does the top-customer concentration (49% in FY26) reset materially as the 4th and 5th hyperscaler ramps, and (3) does the EV/Sales multiple hold 25x as growth decelerates from triple-digit YoY toward "merely" 60–80%? The three scenarios below isolate each lever.

Rating: Buy. Probability weights: 30% Bull / 45% Base / 25% Bear → blended fair value ~$270, within rounding of the headline PT $280, indicating the rating survives the scenario distribution rather than depending solely on the bull outcome. The 30% bull weight reflects optical-inflection visibility and DustPhotonics optionality; the 25% bear weight is meaningful because customer concentration is real and the post-Q4 FY26 12–15% drawdown showed how aggressively the market punishes anything short of acceleration. We pair this Buy with the MRVL Buy thesis for an integrated optical-DSP and AEC view of the AI back-end.

Business Overview

Credo Technology Group Holding Ltd was founded in 2008 by Bill Brennan (CEO since 2013), Cheng "Lawrence" Cheng, and Job Lam⁵. The company is headquartered in San Jose, California (registered in the Cayman Islands), employs over 500 people worldwide, and priced its IPO on January 27, 2022 at $10.00 per share (20 million shares)⁶. Fiscal year ends on the Saturday closest to April 30; FY2026 ended May 2, 2026.

Credo does not break out revenue by product line in its filings; the contour below is reconstructed from management commentary, OFC/ECOC product disclosures, and earnings-call discussion. AECs and optical DSPs together are the dominant contributors; SerDes IP licensing remains a smaller, high-margin tail; chiplets and silicon photonics (post-DustPhotonics) are the emerging optionality.

The five product pillars

Tabs: drill down by pillar

AECs: The defining product. Inside a hyperscaler rack, every cable carrying traffic between top-of-rack switches and accelerators is a candidate for either passive copper, an AEC, or an optical transceiver. Passive copper runs out of reach above ~3 meters at 200G/lane; optics work at any reach but cost more, draw more power, and have a meaningful link-flap rate (random momentary signal loss requiring a retransmit). AECs sit in the middle: copper economics, optics-class signal integrity via embedded DSP, and, Credo's ZeroFlap framing, quantifiably lower flap rates than optical alternatives. Credo invented the smart-copper category and remains the dominant supplier; Marvell's late-2025 "Golden Cable" ecosystem launch is the first credible competitive response.

Optical DSPs: The Cardinal 1.6T family (3nm, 224G/lane) and Bluebird 3nm DSP target the next generation of pluggable optical transceivers. At OFC 2026 Credo demonstrated 400G/800G ZeroFlap optics, a 1.6T AEC, the 3nm Bluebird DSP, and PCIe-over-AEC at Gen6⁷. NVIDIA's Vera Rubin GPU NVL144 and Kyber Ultra NVL576 reference platforms include Credo 1.6T AECs, the most visible AI-platform inclusion to date.

IP & Chiplets: SerDes IP licensing predates the systems business and remains a small, high-margin contributor. Chiplet-class products (OmniConnect gearboxes, HyperLume Active LED cables) and the post-DustPhotonics SiPho PIC roadmap give Credo a platform story that extends beyond pluggable optics into NPO/CPO, mitigating long-term displacement risk if hyperscalers move integration in-package.

Partners: Named hyperscaler customers in AECs include Microsoft, Amazon (AWS), and xAI, with a 4th and 5th hyperscaler now ramping per management commentary⁸. NVIDIA platform inclusion at the 1.6T AEC level adds reference-design momentum. Foundry relationships center on TSMC for advanced-node DSPs (3nm Bluebird).

AEC vs Optical: the structural comparison

AI Connectivity Silicon: Industry & Cycle

The most useful mental model for Credo is the AI rack as a system. A modern AI training rack, NVIDIA NVL72 / NVL144 / Vera Rubin and successors, contains 36–144+ GPUs, multiple NVSwitch / NVLink fabrics, top-of-rack and middle-of-rack switches, scale-up and scale-out networking, and several hundred high-speed interconnects per rack. Every link is a candidate for one of three solutions: passive copper (cheap, limited reach), active electrical cables (smart copper with DSP, Credo's home turf), or pluggable optics (long reach, higher power and cost). At 200G/lane the practical passive copper reach collapses below 3 meters, which is why AECs have become the standard for rack-to-rack and inter-rack within a row. Above that reach, optics, and the DSPs that power them, are the only option, which is why Credo's optical portfolio matters for the next leg of the thesis.

AI revenue / serviceable TAM trajectory ($B)

Customer mix: Q4 FY26 and FY26 full year

The competitive map

The competitive landscape for Credo splits into three layers, with different incumbents in each:

• ALAB (Astera Labs): Closest pure-play peer; leads PCIe retimer market while Credo leads AECs. The two compete most directly in the PCIe Gen6 retimer space and in scale-up connectivity for AI clusters. ALAB's NTM EV/Sales (~16x) anchors the high end of the pure-play comp set.
• MRVL (Marvell): Most direct optical DSP and SerDes competitor; the late-2025 "Golden Cable" AEC ecosystem launch is the first direct challenge to Credo's AEC franchise. MRVL has scale (~$8.2B FY26E revenue), a deeper SerDes IP heritage, and is the head-to-head comp for any prose on optical DSP share.
• AVGO (Broadcom): Dominates optical transceiver/switch silicon at scale; the overhang on DSP pricing. AVGO's customer book at Google + Meta + ByteDance is a multi-program backdrop that pressures Credo's pricing power in optical even where the products do not directly overlap.
• COHR, Lumentum, InnoLight, Eoptolink: Optical transceiver-system competitors; meaningful as customers (they buy Credo DSPs) and as systems-level alternatives where vertical integration is the strategy.
• CRWV (CoreWeave) and the neocloud cohort: Customers, not competitors. The "non-hyperscaler" diversification mgmt targets at ~20% of revenue runs through these GPU-as-a-service operators.

Hyperscaler capex tailwind

The macro frame: AI infrastructure spend is trending toward $500B globally by 2026 per marketsandmarkets aggregation⁹. Within that, the scale-up networking TAM is >$40B by 2030 (650 Group via marketsandmarkets) and the AI back-end data-center switch market alone is >$100B by 2030 per Dell'Oro⁹. Credo's claimed serviceable TAM is >$10B and has more than tripled in 18 months across the five product pillars. The growth dependency is therefore not on the macro tailwind, that is intact, but on Credo's share within an expanding pie.

Competitive Position vs MRVL / ALAB / AVGO

The competitive analysis matters because Credo's premium multiple depends on category leadership across two-and-a-half of three battlefronts: AECs (clearly leads), optical DSPs (peer-level competition with MRVL/AVGO), and PCIe retimers (trailing ALAB but with Gen6 design wins). The risk in the bear case is not that Credo loses on every front; it is that share erodes at the margin and the market re-prices the durability premium.

Head-to-head against the three direct comps

Why CRDO's multiple sits above MRVL but below ALAB

CRDO at ~22x FY27E sales is structurally above MRVL (~9x) because Credo is a pure-play AI connectivity bet without diversified semi exposure (no carrier, no consumer, no automotive), and FY26 growth of +206% is more than 4x faster than MRVL's ~42%. CRDO sits below ALAB (~16–18x but on a smaller base and softer growth) primarily because customer concentration is more visible at Credo (49% / 32% disclosed) and because PCIe retimers (ALAB's home) are seen as a more defensible category than AECs against silicon-photonics displacement. Our base case maintains the 25x level through FY27; bull case rerates to 28x on optical inflection and concentration easing; bear case compresses to 14.5x as a concentration shock is re-priced.

Two-way pair with MRVL

For an integrated AI-back-end view, owning both CRDO and MRVL captures the full DSP/SerDes/AEC stack: CRDO is the higher-beta, higher-multiple, more concentrated bet on the AEC franchise and optical inflection; MRVL is the lower-beta, more diversified, AI-mix-as-percent vehicle on custom-silicon ASICs. The trade-off is purity vs. diversification.

Financial Health & Trends

Credo's FY26 produced one of the cleanest profitability inflections in the AI infrastructure cohort: revenue more than tripled while non-GAAP gross margin held at 68.1%, non-GAAP operating margin expanded from the high-20s to 47.8%, and non-GAAP net income grew >5x to $661.5M. Cash & short-term investments closed FY26 at $1.4B with no material debt disclosed in retrieved sources. The model leverage from here depends on Q1 FY27 gross-margin trajectory (guided 67–69%) and on optical mix scaling without compressing the blended margin profile.

Segment / pillar revenue mix (estimated)

Q4 FY26 print highlights (reported June 1, 2026)

The Q4 FY26 print itself was a clean beat. The market reaction (12–15% after-hours drawdown) reflected the in-line Q1 guide rather than any miss, illustrative of how aggressive consensus expectations have become. Q4 FY26 GAAP diluted EPS $0.88; diluted share count 192.7M (GAAP) / 195.9M (non-GAAP).

Profitability vs MRVL / AVGO benchmark

Bookings & Customers: Concentration as a Binary Risk Axis

The single most important slide in the deck. Per the FY26 10-K commentary (via Zacks/TradingView summary), Customer A accounted for 49% of FY26 revenue and Customer B for 32%: combined ~81% from two relationships². The Q4 FY26 mix moderated meaningfully: four hyperscalers each ≥10% of Q4 revenue, with the top three at 34% / 27% / 16%. The trajectory is therefore improving against the historic peak (a single customer, Amazon per Benzinga's March 2025 reporting, was ~67% of FY25 revenue and ~86% in Q3 FY25). But "improving from extreme" is not the same as "diversified."

Why concentration is binary rather than continuous

At 49% top-customer concentration, a 25% reduction in that customer's purchases, for example, a re-allocation of one-third of AEC share to Marvell's Golden Cable ecosystem, removes roughly $200–$250M of incremental FY27 AEC revenue from the model. That is ~10% of consensus FY27 revenue; the cap-rate impact on the multiple is larger than 10% because the market would re-price the durability of the AEC franchise simultaneously. A second-order effect: losing fast-growing AEC revenue would raise the top-2-customer concentration ratio (because the lost dollars are growing faster than the rest), not lower it, making the bear case worse rather than self-correcting.

Named hyperscaler customers

• Microsoft (Azure): Confirmed AEC ramp; historically the largest AEC customer in FY24, displaced by another hyperscaler in FY25.
• Amazon (AWS): Historic peak concentration (~86% in Q3 FY25 per Benzinga, March 2025); now believed to be one of the top two FY26 customers.
• xAI: Confirmed AEC ramp; smaller dollar base but fast-growing on Colossus and follow-on builds.
• 4th and 5th hyperscaler: Each "now ramping" per management commentary; the bull case requires these to scale to ≥10% in FY27.
• NVIDIA (platform inclusion): Not a direct buyer, but Credo 1.6T AECs included in Vera Rubin NVL144 and Kyber Ultra NVL576 reference platforms; downstream pull-through from NVIDIA's customer ecosystem.

Diversification target: the management frame

Management has framed a target of ~20% non-hyperscaler mix (neocloud GPU-as-a-service operators, enterprise AI clusters, sovereign AI deployments). Within Q4 FY26 the non-hyperscaler share was below that, but moving in the right direction. The diversification narrative is more credible than at this time a year ago, but the math still requires the 4th and 5th hyperscaler ramps to materialize over the next four quarters.

Capital allocation

Credo entered FY27 with $1.4B cash & short-term investments and no material disclosed long-term debt, among the cleanest balance sheets in the AI-semi cohort. Capital allocation priorities in FY27: (1) close and integrate DustPhotonics ($750M cash deployed May 28, 2026; ~$650M remaining); (2) fund R&D for Cardinal 1.6T DSP scaling and SiPho/CPO development; (3) retain a buffer for additional bolt-on M&A in silicon photonics or CPO ecosystem. No dividend; no announced buyback program of material size. SBC is meaningful (typical of a recent IPO / growth semi) and will continue to dilute share count modestly.

Valuation Overview & Comps

Why EV/Sales over P/E for CRDO

Credo is in a profitability inflection phase, non-GAAP operating margin expanded from ~30% in FY25 to 47.8% in FY26 and may compress modestly in FY27 as optical investment scales. Forward P/E (47.3x on FY27E ~$5.30 non-GAAP EPS) is informative but the EPS denominator is bouncing materially as the company guides and reports a moving margin target. EV/Sales smooths these dynamics, ties to a single forward revenue figure, and allows clean comparison with ALAB (the closest pure-play peer) and MRVL (the head-to-head DSP comp).

The CRDO vs ALAB multiple question

ALAB at ~16x and CRDO at ~22x sit on either side of the "pure-play AI connectivity" tier. CRDO trades higher because growth is faster (+206% vs ALAB's ~+90% est.) and the AEC franchise has more demonstrated category leadership than ALAB has in PCIe retimers (where Credo is also competing). ALAB's discount to CRDO reflects (a) softer growth differential, (b) similar customer concentration, (c) CPO displacement risk to PCIe-over-cable that the market has begun to assign weight to. Our base case maintains CRDO above ALAB through FY27; bull case widens the gap; bear case compresses CRDO toward ALAB or below as growth decelerates.

Analyst consensus check

EV/Sales Scenario Model

FY27E revenue × EV/Sales multiple = enterprise value, plus net cash, divided by diluted shares = forward fair value, discounted one year at 15% (semis cycle visibility) to today's PT. Scenarios reconcile to data.js targets: Bull $360 (rev $2.45B × 28x), Base $280 (rev $2.15B × 25x), Bear $145 (rev $1.85B × 14.5x; multiple compresses from 25x to 14.5x as concentration is re-priced).

Sensitivity grid: FY27E revenue × EV/Sales multiple (PT in $)

Quick PT calculator: Scenario Targets / Build Your Own

The reconcile-to-priceOf() calculator: any (FY27E revenue, EV/Sales multiple) pair maps to an implied PT using the same engine that drives the scenarios above.

Risk / Reward calculator

Ask the Thesis AI-assisted checking…

Describe an AI-connectivity scenario in natural language; the assistant returns a structured impact analysis against this dashboard's EV/Sales model, customer-concentration assumptions, and competitive dynamic vs MRVL / ALAB / AVGO. Powered by Claude via a Cloudflare Worker proxy.

Catalysts & Roadmap

Upside catalysts

• 1.6T Cardinal DSP volume ramp: Drives the >$600M optical FY27 commentary. Confirmation comes from the Q1/Q2 FY27 product mix disclosure.
• 4th and 5th hyperscaler design wins to ≥10% revenue: Material concentration relief; reduces the binary risk and supports multiple expansion.
• AEC adoption beyond top-of-rack: Inter-rack and middle-of-rack AECs at 200G/lane open a 2–3x volume expansion within the same customer accounts.
• DustPhotonics SiPho PIC product launches: NPO/CPO product reveals at OFC 2027 (March 2027) would be the first concrete validation of the $750M acquisition rationale.
• NVIDIA platform inclusion economics: Vera Rubin NVL144 and Kyber Ultra NVL576 revenue per platform disclosure would help size 1.6T AEC TAM more precisely.

Downside watch items

• Top-customer share loss to MRVL Golden Cable: The single largest bear-case trigger. Watch top-customer disclosure trend in Q1 FY27 10-Q.
• Co-packaged optics displacement: If hyperscalers accelerate CPO adoption faster than expected, pluggable DSP demand compresses (DustPhotonics partially mitigates).
• AI capex air pocket: Hyperscaler capex pause/reset would compress Credo revenue more than diversified semi peers because of pure-play exposure.
• Gross margin slippage below 67%: Q1 FY27 GM at low end of 67–69% guide signals adverse optical mix; sustained sub-67% would impair the operating-leverage story.

Roadmap table

Risk Factors

• Customer concentration, HIGH severity. Two hyperscalers concentrated ~81% of FY26 revenue (49% / 32%). Loss of the top customer would compress FY27 revenue by 15–25% and would re-price the durability premium in the multiple. Mitigation: 4th and 5th hyperscalers now ramping; non-hyperscaler diversification target ~20%.
• Multiple compression on any miss. At ~22x FY27E sales, CRDO leaves limited margin for error. The 12–15% after-hours drawdown post Q4 FY26, despite a clean beat, illustrates the asymmetry. A miss on Q1 FY27 revenue or a Q2 print below acceleration would likely re-rate the multiple toward the 13–17x range.
• AEC displacement risk. Marvell's late-2025 Golden Cable AEC ecosystem launch is the first credible competitive challenge to the Credo AEC franchise. Hyperscalers prefer dual-sourcing at scale; Credo can lose share at the margin even with a superior product.
• AI capex air pocket. Pure-play AI exposure means a 2026/2027 hyperscaler capex pause or reset would compress Credo revenue more than diversified semi peers. No carrier / consumer / automotive backstop.
• Co-packaged optics (CPO) displacement. If hyperscalers accelerate CPO adoption faster than expected, traditional pluggable DSP demand compresses. DustPhotonics acquisition is the partial mitigation, but the integration is unproven.
• ALAB and AVGO competitive overhang. ALAB leads PCIe retimers; AVGO dominates optical transceiver silicon at scale. Both have customer relationships at Credo's targeted hyperscaler accounts.
• Gross margin slippage from optical mix. Q1 FY27 guide of 67–69% (vs 68.3% Q4 FY26) signals near-term mix-driven pressure. Sustained sub-67% would impair operating leverage.
• Acquisition integration risk. DustPhotonics is Credo's largest acquisition by far; up to ~3.21M contingent shares add future dilution. Integration execution and product-roadmap delivery are the swing variables on the optionality value.
• SBC dilution. Typical of recent-IPO growth semis. Diluted share count creeps higher over time absent a buyback program of material size.
• Geopolitical / export controls. Hyperscaler customers are dominantly US; export-control risk is more about derivative ecosystem (foundry access, China-bound silicon) than direct CRDO revenue.

Bull vs Bear Debate

Technical Analysis & Positioning

Trader's view

• 52-week range $35–$251; trading near the upper end after the 2026 AI-interconnect re-rating. ATH close $236.03 on May 29, 2026 (per macrotrends); intraday and recent close ~$251.
• Key support: $180 (Q4 FY26 post-print drawdown level). A close below opens the path to $145 bear-case PT zone.
• Key resistance: $251 (the prior peak) and then $280 (base PT). A weekly close above $280 unlocks the path to $360 bull target.
• Volatility: realized 60-day annualized vol ~70–90%, extreme even for AI-thematic semis. The Q4 FY26 print produced a 12–15% after-hours drawdown despite a clean beat; this is the volatility character to expect through the FY27 print cadence.
• R/R from $251 to $360 target with $180 stop = 1.5:1, acceptable but not strong. For a Buy thesis at these levels, the trade requires conviction that optical inflects and concentration moderates; alternatives are to wait for a pullback to $200–$220 for better entry, or to size the position smaller given the volatility character.
• Short interest is not material per recent disclosures; positioning is dominantly long across institutional and retail constituencies.