Credo Technology Group Holding Ltd

Thesis

Credo is the purest public play on active electrical cables becoming the default rack-scale interconnect for AI clusters: FY2026 (ended 2026-05-02) revenue tripled to $1,335.1M, management guided FY2027 to grow another 80%+ with a second $600M+ optical leg, and the model already prints 68% gross margins and ~50% non-GAAP net margins with $1.44B net cash. The thesis holds if revenue compounds roughly 28%/yr through FY2031 (~$4.6B) as AEC and optical-DSP attach rates rise with each 100G-to-200G-per-lane transition across at least four hyperscaler customers. At $236.50 the market already capitalizes materially more than that — this publishes as a watchlist thesis with an entry gate near $85-90, not a buy at today's price.

Why Now

The inflection is no longer a forecast; it is printed in the filings. Revenue for fiscal 2026 (ended May 2, 2026) was $1,335.1M versus $436.8M in FY2025 — up 206% (Q4 FY2026 press release, furnished on Form 8-K, issued 2026-06-01). The exit rate is even steeper: Q4 FY2026 revenue of $437.0M grew 157% year over year — the fourth quarter alone matched the entire prior fiscal year. Operating leverage arrived with it: GAAP operating margin swung from 8.5% to 33.3% year over year, GAAP net income rose from $52.2M to $472.3M, and non-GAAP net income quintupled to $661.5M — a 49.5% non-GAAP net margin (same release, GAAP-to-non-GAAP reconciliation).

Why now, mechanically: at 100G-per-lane signaling, passive copper (DAC) reach collapses below ~2 meters while pluggable optics burn roughly 5x the power and introduce link flaps that stall GPU training jobs. AECs — copper cables with retimer silicon at each end — landed in the gap, and Credo's ZeroFlap positioning turned a cable into a reliability product hyperscalers now spec by default for intra-rack and adjacent-rack connectivity. Adoption is visibly broadening, not concentrating: four hyperscale customers each contributed 10%+ of revenue in Q4 FY2026 (top three at roughly 34%/27%/16% per Zacks' summary of the quarter), versus the FY2025 profile where a single customer (Amazon) dominated. The FY2026 Form 10-K (period ended 2026-05-02, filed June 2026) shows the top ten customers at ~90% of revenue with two customers each ≥10% for the full year.

The forward evidence is equally concrete. Q1 FY2027 guidance is $465-475M (+~110% YoY at the midpoint; same 8-K press release), and on the June 1, 2026 earnings call management guided fiscal 2027 revenue growth above 80%, with the optical portfolio — optical DSPs, silicon-photonics PICs, and ZeroFlap optical transceivers, each guided to $100M+ — contributing more than $600M, while non-GAAP opex grows only ~50%. The balance sheet corroborates the demand signal: inventories rose from $90.0M to $250.8M year over year (build-ahead for committed demand), and cash plus short-term investments ended at $1,443.3M with zero borrowings (press release balance sheet, 2026-05-02).

Business Quality

Credo's moat is SerDes IP expressed as a system. The founders' serializer/deserializer pedigree lets Credo build retimer and DSP silicon with industry-leading power per bit, then sell it in the highest-margin form factor available: a finished cable qualified into a hyperscaler's rack design, wrapped with the PILOT diagnostics software layer. Qualification is the real switching cost — once an AEC SKU is designed into a cluster generation, ripping it out mid-buildout risks GPU utilization, which is the one metric hyperscalers will not trade for pennies on a cable. The unit economics show it: 68.2% GAAP gross margin in Q4 FY2026 — extraordinary for anything cable-shaped — and non-GAAP operating margin of 49.6%, with FY2026 non-GAAP net margin at 49.5%. The fabless model (TSMC) keeps capital intensity low: property and equipment is just $101.6M against $1.34B of revenue.

Competition is real and coming from above and below. Marvell sells AEC DSPs into competing cable vendors, Point2 Technology and Spectra7 attack the retimer socket on price, and in optical DSPs Credo is the challenger against entrenched Marvell and Broadcom. The structural risk is architectural rather than competitive: NVIDIA's rack-scale designs (NVL72-class) use an integrated copper backplane for scale-up, so Credo's AEC franchise is disproportionately levered to hyperscalers' custom-ASIC clusters (AWS Trainium, Microsoft, xAI). A world where NVIDIA's integrated racks take a larger share of installed accelerators dilutes Credo's attach rate. Concentration compounds this: ~90% of FY2026 revenue came from ten customers, and the top three were ~77% of the latest quarter. Any one of them dual-sourcing aggressively resets the growth curve.

Capital allocation is conservative to a fault: $1.44B of cash and short-term investments, no debt, no buyback, one small acquisition in FY2026 ($92.8M goodwill + $29.3M intangibles, likely the OmniConnect memory line). The offset is stock-based compensation of $182.6M in FY2026 (~13.7% of revenue) — real dilution, with diluted weighted shares up ~5.8% YoY (182.1M to 192.7M in Q4). Management is founder-led (CEO Bill Brennan since 2013 alongside founding technologists), and execution over FY2025-26 — tripling revenue while expanding gross margin 320bps — has been close to flawless. The open question is not quality; it is duration of hypergrowth, and that is precisely what today's multiple assumes.

Financial Base

Reference figures as of 2026-07-23: trailing-twelve-month revenue of $1,335M, 192.7M diluted shares, net debt of $-1,443M, and a share price of $236.50. Sources:

• Q4 & FY2026 press release (furnished on Form 8-K), fiscal year ended 2026-05-02, issued 2026-06-01 — FY2026 revenue $1,335.116M; Q4 revenue $437.003M; FY2025 revenue $436.775M; GAAP net income $472.279M; non-GAAP net income $661.543M
• Same release, condensed balance sheet as of 2026-05-02 — cash & equivalents $1,164.952M + short-term investments $278.334M = $1,443.286M; no borrowings on the balance sheet (net debt shown as negative = net cash; excludes ~$25-30M operating lease liabilities)
• Same release, statements of operations — Q4 FY2026 GAAP diluted weighted-average shares 192.681M (FY2026 full-year diluted 188.232M); GAAP diluted EPS $0.88 (Q4)
• FY2026 Form 10-K, period ended 2026-05-02, filed June 2026 — top ten customers ~90% of FY2026 revenue; two customers each ≥10% of full-year revenue (per StockTitan/Zacks 10-K summaries)
• Q4 FY2026 earnings call, 2026-06-01 (Seeking Alpha summary) — FY2027 guided to >80% revenue growth; optical portfolio guided to >$600M with three $100M+ product lines; Q1 FY2027 revenue guidance $465-475M per the 8-K press release
• Share price $236.50, Nasdaq close 2026-07-23 (web quote aggregators; intraday range in July 2026 spanned ~$214-290, all-time high $302.52 on 2026-06-22 per Macrotrends)

Scenario Valuation

Five-year scenario ranges under the published methodology — the tables below are computed from these exact parameters at publication and sealed with this text. What each scenario assumes:

• Loss (5%): AI capex bust coincides with architectural displacement: NVIDIA-style integrated copper backplanes plus a Marvell/Point2 price war strand the AEC franchise; revenue halves and the multiple collapses onto a business worth little more than its ~$7.50/share net cash plus a shrunken royalty on connectivity silicon.
• Bear (30%): FY2027 largely lands (H1 is effectively booked) but 2028-29 brings hyperscaler digestion and AEC ASP compression; revenue stalls near $2.4-2.9B (FY28-31 growth ~5%/yr) and the stock re-rates to a mature mixed-signal multiple — 6x EV/S is where Marvell and Microchip have traded as single-digit growers with strong gross margins.
• Base (45%): FY2027 grows ~75-80% per guidance, then decelerates (+30%/+18%/+12%/+8%) to ~$4.6B by FY2031 as AEC broadens across hyperscalers and optical becomes a real second leg but competition normalizes pricing; 32% terminal FCF margin (vs 35.4% FY2026 GAAP net margin) at 26x P/FCF — the ADI/TXN mature-analog range for a low-teens grower with fabless economics.
• Bull (20%): AI cluster buildouts sustain through 2030, AECs become standard in scale-up as well as scale-out fabrics, optical DSP/transceivers scale past $1.5B, and Credo reaches ~$8B revenue at ~50% non-GAAP operating margins; 10x EV/S at exit matches where premium connectivity franchises (pre-froth Broadcom, peak-quality Marvell) have traded while still growing 15-20%.

Published range (present value): $52.83 – $237.23 · probability-weighted expected value $117.31 against $236.50 at drafting (0.50×) · base-case IRR −5.1%/yr. Gates: base IRR ≥ 15% and EV ≥ 1.3× — both passed at publication. Ranges are graded at the 1-year checkpoint and the 5-year horizon. This is a range, not a target.

Sensitivity — present value across growth × exit multiple

The shaded cell is the one today's price of $236.50 most closely implies. Find your own cell — we show our work rather than assert a number.

Kill Conditions

1. 1. Two consecutive quarters of sequential revenue decline greater than 10% in the 10-Q — the digestion/displacement scenario has begun and the bear/loss branch is live.
2. 2. FY2027 revenue (10-K, period ending ~May 2027) comes in below $2.0B — i.e., below +50% growth versus the >+80% guided on 2026-06-01 — falsifying the demand-durability leg of the thesis.
3. 3. Optical portfolio revenue below $400M for FY2027 versus the >$600M guided (disclosed on earnings calls) — the second growth leg failed, leaving a single-product AEC story.
4. 4. Non-GAAP gross margin below 60% for two consecutive quarters (press releases; Q4 FY2026 was 68.3%) — evidence AEC pricing is commoditizing under Marvell/Point2 competition.
5. 5. Any single customer exceeds 45% of quarterly revenue for two consecutive quarters, or two of the current four 10%+ hyperscalers drop below 10% simultaneously (10-Q concentration disclosures) — the diversification that de-risked FY2026 is unwinding.

Risk Register

• Valuation — extreme severity: ~33x EV/TTM sales and ~18x guided-FY2027 sales leaves zero room for deceleration; in our bear case the stock falls ~65% even though revenue still doubles from FY2026.
• Customer concentration — high: top ten customers ~90% of FY2026 revenue, top three ~77% of the latest quarter; a single dual-sourcing decision resets the curve.
• Architectural displacement — high: NVIDIA integrated copper backplanes (scale-up) and, later in the window, co-packaged optics could shrink the AEC socket; Credo's franchise is levered to custom-ASIC hyperscaler clusters.
• Competition — medium-high: Marvell sells AEC DSPs to rival cable makers, Point2/Spectra7 attack on price, and in optical DSPs Credo challenges entrenched Marvell/Broadcom.
• Hyperscaler capex cyclicality — medium-high: revenue is tied to cluster buildout schedules that have historically paused with little warning (2019, 2022 precedents).
• Supply chain/geopolitics — medium: fabless at TSMC with assembly concentrated in Asia; Taiwan disruption or China export rules would hit both supply and a portion of demand.
• Stock-based compensation — medium: $182.6M in FY2026 (~14% of revenue), driving ~3-6%/yr dilution that compounds against shareholders in every scenario.

Theme Position

Credo is effectively a pure play: substantially all revenue is AI/data-center connectivity — AECs plus optical DSPs, transceivers, retimers, and line cards sold into hyperscaler AI clusters — so theme revenue exposure is realistically 95%+ (the legacy non-AI remainder is de minimis after FY2026's triple). Within the ai-infrastructure theme, it is the cleanest public expression of the specific angle: AECs became the standard intra-cluster rack interconnect, and Credo defined the category.

Alternatives considered: Marvell (MRVL) offers AEC-DSP plus optical-DSP exposure at roughly a third of Credo's sales multiple, but the AI story is diluted by a large legacy business and lumpy custom-silicon wins — you buy the theme at maybe 40% purity. Astera Labs (ALAB) is the closest analog (PCIe retimers/scale-up connectivity, hyperscaler concentration) but carries the same problem as Credo — a valuation that already prices the bull case — without Credo's finished-cable systems moat. Amphenol/TE Connectivity win physical-layer content in every AI rack and trade at defensible multiples, but AI is a minority of revenue, so they are diversified industrials, not theme vehicles. Credo wins on purity and business quality; the entire debate is price.

Benchmark

Beating means CRDO total return exceeds QQQ total return over the full window from publication (late July 2026) to late July 2031. Success means beating QQQ's total window return from publication; the scoreboard grades this thesis against that bar, live, whether we like the answer or not.