Stock Analysis · Bulls vs Bears

Credo Technology Group CRDO

🐂 Bulls vs Bears 🐻Report #19 Oct 2026
★ Willem owns this stock · 2% of the portfolioOlder version — see the latest →
51
high risk
🐂 9vs9 🐻
Even
42%
▲ Chance of the biggest tailwind the forecasts miss
30%
▼ Chance of the biggest headwind the forecasts miss
The verdict

Credo is a very profitable, fast-growing, but heavily concentrated supplier whose core copper product is today's default for short AI links, and which is buying and building its way into optics before copper's window narrows. Owning it is a bet that its system-level know-how keeps it the preferred supplier through the 1.6T generation, and that optics and scale-up grow fast enough to matter before competition or…

The Bulls vs Bears podcast is being made
Two AI hosts debate this report, NotebookLM-style. It appears here automatically within a few hours.

Written by Claude (Anthropic’s AI) from public sources · Willem owns CRDO · AI can make mistakes — numbers in brackets point to the sources · Not financial advice

🐂 Bull arguments 9

  1. 1Credo is the leading supplier of the default short-reach link at 800GStrong
  2. 2Optics is a credible second engine, and Credo owns more of the stack than most rivalsMedium
  3. 3A record of conservative guidance and fast executionMedium
  4. 4Outstanding economics and a clean balance sheetStrong
  5. 5Scale-up networking opens a new, larger socketMedium
  6. 6Option value from products nobody models yetWeak
  7. 7Reliability and telemetry create practical switching costsMedium
  8. 8AI data-centre spending is still acceleratingMedium
  9. 9Management incentives point at a much larger companyWeak

🐻 Bear arguments 9

  1. 1Extreme customer concentration with no contractual protectionHigh
  2. 2Copper's window may close: optics could take over short linksMedium
  3. 3The AEC moat is being commoditisedMedium
  4. 4Optics is a harder, lower-moat business, and Credo is new to itMedium
  5. 5Single points of failure in supplyMedium
  6. 6High expectations and violent share-price swingsMedium
  7. 7A capex digestion phase would hit Credo hardMedium
  8. 8Geopolitics and compliance exposureLow
  9. 9Insiders only sell, and dilution through stock pay is heavyLow

Bull arguments show how well they survive the devil’s advocate (strong / medium / weak); bear arguments show the damage they could do after the counter-argument (high / medium / low).

Bull case — all positive points (most important first)

🐂 Bull #1Credo is the leading supplier of the default short-reach link at 800G

  • The point: Hyperscalers switched from optical cables to copper AECs for in-rack and neighbouring-rack links. Credo was first and is the leader.
  • Evidence:
    • FY26 revenue rose 206% to $1,335M, and AECs drove more than 99% of the increase [1].
    • Credo is in volume with 5 of 6 hyperscalers [5].
    • Nvidia's Gilad Shainer: "Copper is the best connectivity, if you can use it." Pluggable optics would have added about 20 kW per NVL72 rack [14].
    • Credo frames AECs as the "1-meter to 7-meter segment" of the link market [5].
  • Why it matters: every new AI cluster needs thousands of these links. A preferred-vendor position across most hyperscalers turns capex growth directly into Credo revenue.
  • 😈 Devil's advocate:
    • "Leader at 800G" does not guarantee leadership at 1.6T, where copper reach shrinks (Credo's 1.6T AECs reach 6.5 m [4]).
    • The only third-party AEC market size found, 650 Group's $644M in 2025 [13], is smaller than Credo's own revenue, so nobody outside the company can measure its share properly.
  • Verdict: Strong The results prove the position today. Its durability is the open question, covered in Bear #2–#3.

🐂 Bull #2Optics is a credible second engine, and Credo owns more of the stack than most rivals

  • The point: Credo is using AEC profits to become a full optical supplier: DSPs, silicon-photonics chips and complete modules. That reduces the copper-substitution risk and enlarges the market.
  • Evidence:
    • FY27 optical guidance is above $600M, with ZF optics, PICs and DSPs each above $100M. It was raised from above $500M in April and held in September [4,5,7].
    • Record DSP revenue in Q1 FY27 and first PIC revenue [4].
    • Two major PIC design wins ramp in FY28 [4].
    • The Oracle co-development of ZF optics, and an OCP reliability workstream co-chaired with Oracle [1].
  • Why it matters: the optical module market is several times larger than AECs; LightCounting puts AI Ethernet optics plus CPO at about $8B in 2025 and about $20B by 2030 [16]. If optics grows to half of revenue, as Brennan said "we can make a case" for, Credo's fate is less tied to one product [5].
  • 😈 Devil's advocate:
    • Optics is crowded: Broadcom and Marvell in DSPs; Innolight, Eoptolink and Coherent in modules.
    • Credo is a newcomer there and paid about $1.25B (including earn-out value) for Dust [2].
    • The FY27 optical guide was not raised in September, and analysts cited that in the 20% one-day drop [24,26].
    • The PIC design wins exclude Credo's DSP, so the "full stack" pitch is not yet showing up in the wins [4].
  • Verdict: Medium The revenue is real and growing, but the moat in optics is unproven and the margin profile is likely lower than AECs (interpretation).

🐂 Bull #3A record of conservative guidance and fast execution

  • The point: management consistently delivers more than it promises.
  • Evidence:
    • Four straight quarters above the top of the range [3,6,10,11].
    • FY26 growth was roughly double initial expectations [5].
    • FY27 growth guidance was raised from >80% to >85% after Q1 [4,5].
    • Long-tenured, technical leadership (CEO since 2014, CFO since 2015) [8].
  • Why it matters: in a company priced for growth, credibility lowers the chance of a guidance shock.
  • 😈 Devil's advocate:
    • The beats are shrinking. Q4 and Q1 beat the top of the range by only about $2–4M. Mizuho called Q1 "a bit skinnier" than past reports [27].
    • Opex keeps exceeding guide.
    • The FY27 plan relies on a steep second-half inflection [4].
  • Verdict: Medium The credibility is earned, but the size of the beats is no longer a cushion.

🐂 Bull #4Outstanding economics and a clean balance sheet

  • The point: Credo funds its expansion from its own profits.
  • Evidence:
    • Non-GAAP gross margin of about 68% [1,3].
    • FY27 net margin guided at about 50% [4].
    • FY26 operating cash flow of $464M [1].
    • At 2026-08-01: cash plus short-term investments of $764M ($466.9M plus $297.4M) and total liabilities of only $284M, with no borrowings [2].
  • Why it matters: Credo can fund 3nm tape-outs, acquisitions and inventory build-ups without dilution or debt, and can survive a demand air-pocket.
  • 😈 Devil's advocate:
    • Cash fell by about $700M in one quarter on the Dust deal [2].
    • Stock-based compensation was $82M in Q1 alone [3]. Interpretation: that is a real cost for shareholders, hidden by "non-GAAP".
    • Inventory and receivables are rising faster than sales [2].
  • Verdict: Strong

🐂 Bull #5Scale-up networking opens a new, larger socket

  • The point: "Scale-up" is the network that ties many AI chips into one logical computer, inside or across racks. It needs far more links per chip than "scale-out", the network between servers. Today Nvidia's proprietary NVLink dominates. Ethernet-based standards are rising for everyone else.
  • Evidence:
    • Retimer revenue hit a record in Q1, "led by scale-up deployments" [4].
    • Scale-up revenue begins in FY27 and is "more substantial" in FY28 [5].
    • Blue Heron supports Ethernet, UALink and ESUN [5].
    • Dell'Oro expects Ethernet to lead scale-up by the end of the decade and scale-up to exceed half of AI back-end switching by 2030 [15].
    • The ESUN 1.0 specification (led by Meta and Microsoft) was released in March 2026 [19]. The UALink 2.0 specification was ratified in April 2026, with products expected in 2027 [20].
  • Why it matters: more links per accelerator means more Credo content per AI chip, beyond the scale-out cables that built FY26.
  • 😈 Devil's advocate:
    • Scale-up links are the shortest of all, the ones most likely to live on circuit boards or midplanes, as in Nvidia's Kyber design, or eventually in co-packaged optics.
    • Astera Labs is strong in scale-up switching and retimers [21].
  • Verdict: Medium A real, documented opportunity, but standards silicon is still a 2027 story.

🐂 Bull #6Option value from products nobody models yet

  • The point: Weaver/OmniConnect, ALCs and CPO/NPO all start in FY28.
  • Evidence:
    • Brennan sizes Weaver at "$2,000 and $3,000" of content per GPU [5]. Positron is the first named customer [4].
    • ALCs reach up to 30 m [1], the gap between copper (7 m) and optics.
    • An OCP demo of ALCs is planned for October 2026 [4].
  • Why it matters: if even one gets a hyperscaler win, it is a new multi-hundred-million-dollar line. Interpretation: at $2–3k per GPU, a few hundred thousand GPUs would be several hundred million dollars a year.
  • 😈 Devil's advocate:
    • The first Weaver customer is a startup, not a hyperscaler.
    • MicroLED links are unproven at data-centre scale.
    • Management calls these "multi-billion" markets [10], which is a TAM claim, not a forecast.
  • Verdict: Weak to Medium. It is a cheap call option, not a reason to own the company on its own.

🐂 Bull #7Reliability and telemetry create practical switching costs

  • The point: once a hyperscaler has qualified Credo's links and built monitoring around PILOT, swapping vendors means requalifying an entire fleet.
  • Evidence:
    • Microsoft co-created the SWITCH AEC [1].
    • Oracle co-chairs the optics reliability workstream with Credo [1].
    • Requalification takes months: 9–12 months if Credo itself changes supplier [1].
  • Why it matters: this raises the cost for a customer to dual-source each new generation.
  • 😈 Devil's advocate: hyperscalers deliberately dual-source, and each new speed generation is a fresh qualification anyway, so the lock-in resets every two to three years.
  • Verdict: Medium

🐂 Bull #8AI data-centre spending is still accelerating

  • The point: Credo's demand is a function of hyperscaler capex.
  • Evidence:
    • The four largest US hyperscalers' 2026 capex plans rose to about $735–750B by July, from about $640B in January. Alphabet says 2027 rises "significantly" (tracker data, not primary) [17].
    • Nvidia guided its next quarter to about $108B in revenue (secondary) [38].
    • Broadcom's AI semiconductor revenue was $16.7B in its latest quarter, up 221% (secondary) [39].
  • Why it matters: 2027 build plans are already in motion, which supports the H2 FY27 inflection.
  • 😈 Devil's advocate: this is the most widely known fact in the market and is fully in consensus. Alphabet's free cash flow turned negative in Q2 2026 (secondary) [17], which shows the spending pace has a financial limit.
  • Verdict: Medium It is a support, not a differentiator.

🐂 Bull #9Management incentives point at a much larger company

  • The point: the CEO's 2026 award pays only if revenue reaches $2.5B to $7.5B and the share price reaches $244.70 to $489.40 by 2031 [8].
  • Evidence: the six tranches are 239,500 shares each [8].
  • Why it matters: it rewards growth plus share-price performance, not just one of the two.
  • 😈 Devil's advocate:
    • Founders and executives have sold heavily (Bear #8).
    • The award also encourages acquisitions that buy revenue.
  • Verdict: Weak

Bear case — all negative points (most important first)

🐻 Bear #1Extreme customer concentration with no contractual protection

  • The point: three to four buyers decide Credo's revenue, and they can walk away at the next generation.
  • Evidence:
    • By end customer, Q1 FY27 was 33%, 28%, 13% and 10%. The top three were unchanged from Q4 [4].
    • The top 10 were about 90% of FY26 [1].
    • Two contracting parties hold 85% of receivables [2].
    • Customers have no long-term purchase commitments and can cancel or reschedule at short notice [1].
    • Credo may grant "exclusivity or most favored nation pricing" (10-K risk factor) [1].
  • Why it matters: losing one top-two customer's next-generation design could remove 25–35% of revenue within a year. That is a cliff, not a slope.
  • 🛡️ Counter-argument:
    • Concentration has improved from 67% for one contracting party in FY25 [9] to 43% in Q1 FY27 [2].
    • There are now four 10% customers, with neoclouds coming [4,5].
    • The top three have stayed the same and keep growing.
  • Severity: High Diversification is real but slow, and the top end customer is still a third of revenue.

🐻 Bear #2Copper's window may close: optics could take over short links

  • The point: as speeds double, copper's useful reach shrinks. CPO, NPO and linear optics aim to remove the cables Credo sells.
  • Evidence:
    • Credo's 1.6T AECs reach 6.5 m, versus 7 m at 800G [4].
    • Nvidia's Huang on scale-up: "We're going to do both" [14].
    • Nvidia's 2028 Feynman platform could use copper or CPO NVLink, which is undecided [14].
    • Nvidia invested $2B each in Coherent and Lumentum and announced a $2B optical tie-up with Marvell in March 2026 [14].
    • LightCounting says CPO adoption could exceed its forecast in 2028–2031 [16].
    • The stock fell on 2026-06-23 on exactly this worry [30].
  • Why it matters: AECs are the largest and highest-margin business. A shift in one major platform cuts both revenue and margin.
  • 🛡️ Counter-argument:
    • Copper beat optics at 800G on power and reliability, and the "1,000x" reliability gap is a big hurdle for CPO in production [5].
    • Credo is hedged: it sells optics and ALCs, and has joined the Open CPX group for CPO/NPO [4].
    • Kyber, Nvidia's all-copper Rubin Ultra rack, reportedly slipped to 2028 (secondary) [23], and Rubin keeps copper in-rack [14].
  • Severity: Medium High. It is not a 2027 problem, but it is the central question for 2028–2030.

🐻 Bear #3The AEC moat is being commoditised

  • The point: Credo's edge in AECs is know-how. Rivals are packaging that know-how for the big cable makers.
  • Evidence:
    • Marvell's "Golden Cable" programme (2025-12-09) supplies chips, firmware, reference designs and calibration data. Foxconn Interconnect finished a first design in two months, and Luxshare is a partner [13].
    • Astera Labs' Taurus AEC modules reached pre-production (secondary) [21].
    • Marvell is targeting about $20B of FY28 revenue and is investing heavily in custom silicon and optical I/O [22].
    • The 10-K names Broadcom, Marvell and Astera as competitors [1].
  • Why it matters: a qualified second source lets hyperscalers squeeze prices. A roughly 68% gross margin on a cable is the kind of margin that invites competition.
  • 🛡️ Counter-argument:
    • No public evidence of a Marvell AEC hyperscaler win or of Credo price cuts was found.
    • Non-GAAP gross margin has held at about 68% [3].
    • Credo integrates cable qualification and telemetry, not just a chip.
  • Severity: Medium High. No damage is visible yet, but the threat is built specifically against Credo's core.

🐻 Bear #4Optics is a harder, lower-moat business, and Credo is new to it

  • The point: the growth plan leans on optics, where Credo competes with entrenched giants and buys lasers and parts it doesn't control.
  • Evidence:
    • The $1.25B DustPhotonics price includes $310M of contingent consideration [2].
    • Goodwill rose from $93M to $986M [2].
    • GAAP gross margin fell to 64.5%, with 62.9–64.9% guided for Q2 [3,24].
    • Lasers are short across the industry [5].
    • The FY27 optical target was held, not raised, in September, and analysts (Rosenblatt, JPMorgan) want proof of the ramp [26,34].
    • The FY26 10-K has no dedicated risk factor on optical module manufacturing or laser supply [1].
  • Why it matters: if optics grows slower or at lower margins, the diversification story weakens. Goodwill would also be at risk of write-down.
  • 🛡️ Counter-argument:
    • First PIC revenue arrived on schedule, and ZF optics are shipping [4].
    • Two major PIC design wins are in hand [4].
    • Credo says it secured capacity commitments for ZF optics [5].
  • Severity: Medium

🐻 Bear #5Single points of failure in supply

  • The point: one foundry, one AEC assembler, and a tight 3nm process.
  • Evidence:
    • TSMC is the exclusive wafer source, and all 200G-per-lane parts are on 3nm, where capacity is tight [1,5].
    • BizLink is the AEC assembler [1].
    • Switching takes 9–12 months [1].
    • Inventory rose to $313M from about $90M a year earlier [2,6], and purchase commitments are $338M [1].
  • Why it matters: a capacity squeeze at TSMC 3nm, where larger customers get priority, could cap the 1.6T ramp. A demand slowdown would leave Credo holding inventory.
  • 🛡️ Counter-argument:
    • Every fabless chip company has this profile.
    • Credo says it prepared the supply chain over 18–24 months [4].
    • The inventory build is for a guided H2 ramp.
  • Severity: Medium

🐻 Bear #6High expectations and violent share-price swings

  • The point: the market already expects a lot, and small disappointments are punished hard.
  • Evidence:
    • On 2026-09-02 the stock fell 20% to $165.22, despite a beat and raise [24,25].
    • It fell from a $308.67 high on 2026-06-22 to about $150 in mid-September [37,12].
    • It trades around $210 (2026-10-06), about 33x FY27 and 22x FY28 consensus EPS. Consensus FY27 revenue of about $2.5B sits just above the guidance floor, and FY28 assumes about 55% more growth (secondary) [28].
    • Beta is about 3.2 [25].
  • Why it matters: a guide that merely meets expectations can still cost a third of the value. This is a risk to holders, not to the business.
  • 🛡️ Counter-argument:
    • The multiple has compressed sharply.
    • 19 of 20 covering analysts rate it Buy (secondary) [28].
    • Estimates are rising.
  • Severity: Medium

🐻 Bear #7A capex digestion phase would hit Credo hard

  • The point: Credo's growth depends on a few companies continuing to raise AI spending.
  • Evidence:
    • Alphabet's free cash flow turned negative in Q2 2026, and Meta's fell sharply (secondary) [17].
    • LightCounting expects optics growth to moderate from 2027 [16].
    • Customers can reschedule orders at short notice [1].
  • Why it matters: with concentrated customers and purchase-order-only sales, a pause would show up within a quarter.
  • 🛡️ Counter-argument: all large spenders guided capex up for 2027, and Dell'Oro calls the market "supply-constrained rather than demand-constrained" [15,17].
  • Severity: Medium

🐻 Bear #8Geopolitics and compliance exposure

  • The point: design and production straddle the US, China, Taiwan and now Israel.
  • Evidence:
    • About 99,000 sq ft of leased R&D space in mainland China, plus Taiwan manufacturing [1].
    • The 10-K says Credo has "in the past" found "certain exports may have been made in violation of Trade Controls laws" and has remediated. Penalties or disclosure outcome: not found [1].
    • The US outbound-investment rule may be revised in 2027 [1].
    • Tariffs are "fluid" per guidance [4], and Section 232 semiconductor duties are in place with US data-centre exemptions [35].
    • Pillar Two minimum tax from FY28 [1].
    • DustPhotonics adds Israeli operations [7].
  • Why it matters: a Taiwan disruption, new export rules on China-based R&D, or a tariff on cable assemblies could raise costs or interrupt supply.
  • 🛡️ Counter-argument:
    • Shipments to mainland China are now almost nil, at $1.1M in Q1 [2].
    • The data-centre tariff exemptions currently apply [35].
  • Severity: Low Medium.

🐻 Bear #9Insiders only sell, and dilution through stock pay is heavy

  • The point: no insider has bought shares, and stock-based pay is large.
  • Evidence:
    • From April to October 2026: no open-market purchases [12].
    • CTO Cheng sold about 567k shares (about $116M) and COO Lam about 367k, mostly under 10b5-1 plans (pre-scheduled selling plans).
    • CEO Brennan sold about 55k shares on 5 June, apparently outside a plan, and more at about $150 in September [12].
    • Stock-based compensation was $82M in Q1 FY27 [3]. Non-GAAP opex is guided up about 55% for FY27 [4].
  • Why it matters: this is a mild signal, and a real cost borne by shareholders.
  • 🛡️ Counter-argument: the founders still hold millions of shares each [8]. The selling is pre-planned diversification after a tenfold rise, and continued at prices both high and low.
  • Severity: Low

Tailwinds NOT in the forecasts

  • Scale-up Ethernet becomes a volume Credo market (retimers and AECs inside AI "pods") — 42% chance within 36 months · Impact if it happens: High
    • What it is: custom AI chips (Trainium, MTIA, TPU-class) and merchant GPUs adopt ESUN or UALink scale-up. Credo's Blue Heron retimers and short AECs ship per accelerator, not per server.
    • Why it's not priced in: management only says "more substantial" in FY28 [5]. Standards silicon arrives in 2027 [20]. Consensus FY28 growth of about 55% [28] looks mostly like a continuation of AECs and optics.
    • What would trigger it / the sign to watch: management quantifying scale-up revenue; a named hyperscaler scale-up deployment; retimers disclosed as a 10%+ line.
    • Why this probability: the direction is backed by Dell'Oro and two open standards [15,19,20], but the copper-vs-board-vs-optics choice inside the pod is still open.
  • Optics clearly beats the >$600M FY27 target and approaches half of revenue in FY28 — 38% chance · Impact: Medium-High
    • What it is: ZF optics "more than double or triple" next year, as Brennan suggested [5], and the PIC design wins ramp early.
    • Why it's not priced in: the target was held in September and the market sold off on it [24]. Analysts explicitly want proof [26].
    • What would trigger it / the sign to watch: the optical target raised at Q2 (early December 2026); a named second ZF-optics hyperscaler.
    • Why this probability: management has beaten every revenue guide so far, but optics depends on laser supply and new manufacturing, which Credo has never scaled before.
  • A hyperscaler-class win for Weaver/OmniConnect or ALCs — 30% chance · Impact: Medium
    • What it is: one of the FY28 products lands a major customer beyond the startup Positron.
    • Why it's not priced in: none of these products has revenue yet, and they barely feature in the analyst write-ups found [28].
    • What would trigger it / the sign to watch: the October 2026 OCP ALC demo; design-win commentary on the Q2/Q3 FY27 calls.
    • Why this probability: three separate shots on goal raise the odds, but each is a new product category with no production history.
  • US restrictions on Chinese optical modules favour Credo's own modules — 15% chance · Impact: Medium
    • What it is: Reuters reported that the FCC was preparing a ban on new Chinese optical transceivers. Chinese suppliers hold about 60% of datacom optical components [18].
    • Why it's not priced in: it is an unpublished proposal, and Cignal AI doubts it will stick [18].
    • What would trigger it / the sign to watch: an FCC order or rule text.
    • Why this probability: reported but contested, and hyperscalers would lobby against supply disruption.

Headwinds NOT in the forecasts

  • A top-three customer dual-sources or redesigns its next generation away from Credo — 30% chance within 36 months · Impact if it happens: High
    • What it is: for 1.6T or scale-up, a top customer qualifies a Marvell Golden Cable or Astera AEC for a meaningful share of its links, or uses an optical or board-level design instead.
    • Why it's not priced in: consensus assumes about +87% in FY27 and +55% in FY28 [28]. The top three have been stable [4].
    • What would trigger it / the sign to watch: a top-customer share falling sharply quarter on quarter; a Marvell or Astera hyperscaler AEC announcement; Credo commentary on "share" at 1.6T.
    • Why this probability: hyperscalers routinely dual-source mature parts, and Marvell built a tool for exactly that [13]. Three years covers at least one generation change.
  • AEC price erosion pushes non-GAAP gross margin below about 64% — 27% chance · Impact: Medium
    • What it is: competition plus a lower-margin optics mix push margins down.
    • Why it's not priced in: guidance says FY27 gross margin is "broadly consistent" with FY26 [4].
    • What would trigger it / the sign to watch: the non-GAAP gross margin guide at the bottom of the range for two quarters running.
    • Why this probability: there is no erosion yet, but optics mix and second sources both push the same way.
  • A major platform chooses CPO or optics over copper for in-rack links before 2029 — 28% chance · Impact: High
    • What it is: Nvidia Feynman (2028) or a hyperscaler custom rack goes optical for scale-up.
    • Why it's not priced in: the FY28 consensus assumes continued AEC growth [28].
    • What would trigger it / the sign to watch: Nvidia's GTC 2027 architecture disclosures; hyperscaler NPO deployments; LightCounting CPO forecast revisions.
    • Why this probability: Nvidia is openly hedging and investing in optics [14], but the power and reliability penalties of optics remain large.
  • Hyperscaler capex flattens or declines in 2027–28 — 22% chance · Impact: High
    • What it is: a digestion year at two or more of the top four spenders.
    • Why it's not priced in: consensus and guidance assume continued growth [17,28].
    • What would trigger it / the sign to watch: 2027 capex guides in the January–February 2027 earnings season; negative free-cash-flow pressure at the hyperscalers.
    • Why this probability: all the big spenders are guiding up for 2027, but cash flow strain is visible [17].
  • An export-control or tariff action directly hits Credo's supply chain or China R&D — 15% chance · Impact: Medium
    • What it is: Section 232 Phase 2 without data-centre exemptions, outbound-rule tightening, or a Taiwan event.
    • Why it's not priced in: guidance assumes the current regime [4].
    • What would trigger it / the sign to watch: Commerce Department rule changes; the 2027 outbound-investment revision.
    • Why this probability: policy is volatile, but chips for US data centres have so far been carved out [35].

Investment Risk Score: 51/100

Risk area Weight Score (0–100) Why
Business & competition 25% 55 Clear leader at 800G AECs, but the moat is know-how. Marvell and Astera are building second sources, and optics could substitute in 2028+.
Financial health & dilution 20% 25 $764M cash and short-term investments, no borrowings, roughly 50% net margin. Dilution comes from heavy stock pay and the FY26 $750M share sale.
Execution (delivering the plan) 20% 50 Four straight quarters of beats, but FY27 needs a steep H2, a new optics manufacturing business, 3nm ramps and the Dust integration all at once.
Dependency & concentration (customers, suppliers, commodities, key people) 15% 75 Top four customers about 84% of revenue, purchase orders only, TSMC exclusive, one AEC assembler.
Regulation, policy & geopolitics 10% 45 Taiwan production, China R&D, past apparent export violations, Israel, "fluid" tariffs. Direct China sales are now negligible.
Sentiment & expectations (how much good news the market already expects) 10% 60 About 33x FY27 EPS, consensus near the guidance floor-plus, beta about 3. A 20% one-day drop on a beat shows how fragile it is.

Overall = 51/100 (high, at the boundary with moderate). The business is financially very strong, which keeps the score from going higher. The risk sits in concentration (four customers, one foundry, no contracts) and in whether the AEC franchise survives competition and optical substitution across the next two speed generations. First dive, so no change to report.

Conclusion

  • What kind of company this is:
    • Credo is a fabless connectivity specialist. Its best product, a smart copper cable, became the industry default at 800G and made it one of the most profitable small-cap-turned-large-cap chip companies.
    • Owning it is a bet on two things. First, that hyperscalers keep choosing Credo's system-level reliability over cheaper second sources. Second, that Credo diversifies into optics and scale-up before copper's role shrinks.
  • The 2–3 things that will decide the outcome:
    1. Whether Credo keeps its share at 1.6T and 200G-per-lane, the first generation where Marvell's ecosystem is fully ready.
    2. Whether optics scales profitably beyond the >$600M target.
    3. Whether hyperscaler AI capex stays on its rising path through 2028.
  • Signposts:
    • Bull case confirmed by:
      • Q2 FY27 (early December 2026) beats $525–535M and the >$600M optics target is raised.
      • A fifth 10% customer appears, or the top customer drops below 30% while revenue still grows.
      • Management quantifies scale-up revenue for FY28.
      • Non-GAAP gross margin holds at 67%+ as optics grows.
    • Bull case broken by:
      • Top customer share falls sharply with flat or down revenue.
      • A Marvell or Astera hyperscaler AEC production win is announced.
      • The non-GAAP gross margin guide falls below about 65%.
      • Nvidia or a hyperscaler commits to CPO for in-rack scale-up in a 2027–28 platform.
  • Balanced verdict:
    • The strongest bull version: Credo owns the most reliable short link in AI data centres. It is in volume at five of six hyperscalers, earns software-like margins, and is using that cash to become a full copper-plus-optics supplier, with scale-up and new products as upside nobody models.
    • The strongest bear version: a cable with roughly 68% margins, sold to four buyers without contracts, is exactly what the buyers and Marvell are working to commoditise or engineer away, and the stock still assumes years of high growth.
    • Interpretation: the evidence today leans toward the bull side on the business. Results, margins and customer breadth are all improving, and no share loss is visible. The risk is concentrated and binary rather than gradual, which is why the score sits just inside "high".

Open questions & data gaps

  • Which hyperscaler is the 33% end customer? Filings use shifting letters, and Amazon's current share is not found. This matters a lot: the identity determines which capex plan Credo is most exposed to.
  • AEC revenue in dollars and Credo's AEC market share: not disclosed. The only third-party AEC market size (650 Group, $644M in 2025 [13]) is smaller than Credo's revenue, so its scope is unclear. Medium importance.
  • Marvell and Astera AEC production wins: not found. High importance for Bear #3.
  • Export-control "apparent violations": no details, penalties or self-disclosure outcome found. Low to medium importance.
  • The full Q4 FY26 and Q3 FY26 call transcripts were only available as third-party copies or summaries. The Q1 FY27 transcript is a third-party copy with transcription errors, and key numbers were cross-checked against the press release and 10-Q.
  • ZF optics margins and the optical gross margin versus AECs: not disclosed.
  • Consensus figures and valuation multiples are from secondary sources (24/7 Wall St, TIKR) and should be treated as approximate.
  • DustPhotonics share consideration: the press release, 10-K and 10-Q give slightly different share counts. The 10-Q total of $1.25B is used here [2].
Sources
  1. Credo Technology Group Form 10-K, fiscal year ended 2026-05-02 — SEC EDGAR — 2026-06-15 — https://www.sec.gov/Archives/edgar/data/1807794/000162828026043303/crdo-20260502.htm
  2. Credo Form 10-Q, quarter ended 2026-08-01 — SEC EDGAR — 2026-09-02 — https://www.sec.gov/Archives/edgar/data/1807794/000162828026060111/crdo-20260801.htm
  3. Credo Q1 FY27 results press release (8-K Ex. 99.1) — SEC EDGAR — 2026-09-01 — https://www.sec.gov/Archives/edgar/data/0001807794/000162828026059795/credoq12027ex-991.htm
  4. Credo Q1 FY27 earnings call transcript (third-party copy) — TradingKey — 2026-09-01 — https://www.tradingkey.com/news/transcripts/262145054-tradingkey
  5. Credo Q4 FY26 earnings call transcript (third-party copy) — allmind.ai — 2026-06-01 — https://allmind.ai/earnings/event/2911589
  6. Credo Q4 and FY26 results press release (8-K Ex. 99.1) — SEC EDGAR — 2026-06-01 — https://www.sec.gov/Archives/edgar/data/0001807794/000162828026039474/credoq42026ex-991.htm
  7. "Credo Agrees to Acquire DustPhotonics" (8-K Ex. 99.1) — SEC EDGAR — 2026-04-13 — https://www.sec.gov/Archives/edgar/data/1807794/000162828026024892/april20268-kex991.htm
  8. Credo 2026 proxy statement (DEF 14A) — SEC EDGAR — 2026-08-25 — https://www.sec.gov/Archives/edgar/data/1807794/000162828026058936/crdo-20260825.htm
  9. Credo Form 10-K, fiscal year ended 2025-05-03 — SEC EDGAR — 2025-07-02 — https://www.sec.gov/Archives/edgar/data/1807794/000162828025033813/crdo-20250503.htm
  10. Credo Q3 FY26 results press release — SEC EDGAR — 2026-03-02 — https://www.sec.gov/Archives/edgar/data/1807794/000162828026013205/credoq32026ex-9911.htm
  11. Credo preliminary Q3 FY26 results (8-K) — SEC EDGAR — 2026-02-09 — https://www.sec.gov/Archives/edgar/data/1807794/000162828026006370/feb20268-kex991.htm
  12. Form 4 insider filings, CIK 1807794, April–October 2026 — SEC EDGAR — 2026-04-01 to 2026-10-07 — https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=1807794&type=4
  13. "Marvell Launches Strategic Initiative to Accelerate AEC Ecosystem and Hyperscaler Adoption" (includes 650 Group figures) — Marvell — 2025-12-09 — https://investor.marvell.com/news-events/press-releases/detail/1002/marvell-launches-strategic-initiative-to-accelerate-aec-ecosystem-and-hyperscaler-adoption
  14. Nvidia optical scale-up at GTC 2026 — The Register — 2026-04-05 — https://www.theregister.com/2026/04/05/nvidia_optical_scale_up/
  15. "AI Back-End Switch Sales to Approach $1 Trillion Over the Next Five Years" — Dell'Oro Group — 2026-07-28 — https://www.delloro.com/news/ai-back-end-switch-sales-to-approach-1-trillion-over-the-next-five-years/
  16. "AI scale-up raises the stakes for transceivers and CPO adoption" (LightCounting data) — Lightwave — 2026-03-10 — https://www.lightwaveonline.com/home/article/55362899/ai-scale-up-raises-the-stakes-for-transceivers-and-cpo-adoption
  17. "Follow the CAPEX: Q2 2026 Scoreboard" (tracker, secondary) — Platformonomics — 2026-07 — https://platformonomics.com/2026/07/follow-the-capex-q2-2026-scoreboard/
  18. "FCC ban on new Chinese optical modules" — Cignal AI — 2026-08-04 — https://cignal.ai/2026/08/fcc-ban-on-new-chinese-optical-modules/
  19. "The OCP ESUN 1.0 Specification Has Been Released" (search summary) — Open Compute Project — 2026-03-10 — https://www.opencompute.org/news/the-ocp-esun-10-specification-has-been-released/
  20. UALink Consortium publishes four specifications including 200G (secondary) — HPCwire — 2026-04-07 — https://www.hpcwire.com/off-the-wire/ualink-consortium-publishes-4-specifications-defining-in-network-compute-chiplets-manageability-and-200g-performance/
  21. Astera Labs Q2 2026 financial results (search summary) — Astera Labs — 2026-08 — https://www.asteralabs.com/news/astera-labs-reports-second-quarter-2026-financial-results/
  22. "Marvell raises 2028 revenue forecast on strong AI data center demand" (Reuters, search summary) — KSL/Reuters — 2026-10-06 — https://www.ksl.com/article/51632987/marvell-raises-2028-revenue-forecast-on-strong-ai-data-center-demand
  23. "Nvidia's Kyber rack for Rubin Ultra slips to 2028" (search snippet only) — Tom's Hardware — 2026 — https://www.tomshardware.com/pc-components/gpus/nvidias-kyber-rack-for-rubin-ultra-slips-to-2028
  24. "Credo Shares Fall 20% Despite 115% Revenue Surge Amid Valuation Concerns" — ts2.tech — 2026-09-02 — https://ts2.tech/en/credo-stock-sinks-20-as-115-revenue-growth-fails-valuation-test/
  25. "Why Credo Technology Stock Plunged 20% Today" — The Motley Fool — 2026-09-02 — https://www.fool.com/investing/2026/09/02/why-credo-technology-stock-plunged-20-today/
  26. "CRDO stock stares at worst week in 2 years as analysts wonder if Credo's optical growth can keep pace" — Stocktwits — 2026-09-03 — https://stocktwits.com/news-articles/markets/equity/crdo-stock-stares-at-worst-week-in-2-years-as-analysts-across-wall-street-wonder-if-credo-s-optical-growth-can-keep-pace/cZsMA2vRJHC
  27. "Credo Technology stock plunges 9% despite beating earnings estimates on margin concerns" — Parameter — 2026-09-02 — https://parameter.io/credo-technology-crdo-stock-plunges-9-despite-beating-earnings-estimates-on-margin-concerns/
  28. "One semiconductor stock with a potential 200% upside" (consensus estimates) — 24/7 Wall St — 2026-10-06 — https://247wallst.com/investing/2026/10/06/one-semiconductor-stock-with-a-potential-200-upside/
  29. "Credo Technology crashed for 3 months; this Wall Street pro says it's about to double" (Stifel $350) — 24/7 Wall St — 2026-09-21 — https://247wallst.com/investing/2026/09/21/credo-technology-crashed-for-3-months-this-wall-street-pro-says-its-about-to-double/
  30. "Credo slides as Wall Street weighs optical replacing copper" — Crypto Briefing — 2026-06-24 — https://cryptobriefing.com/credo-technology-copper-optical-replacement-risk/
  31. "Credo Technology Group stock dives despite Q4 earnings beat" — Benzinga — 2026-06-01 — https://cdn1.benzinga.com/markets/earnings/26/06/52924428/credo-technology-group-stock-dives-despite-q4-earnings-beat
  32. "Credo just tied its CEO's pay to a $2.5 billion revenue goal" — TIKR — 2026-06-30 (approx.) — https://www.tikr.com/blog/credo-just-tied-its-ceos-pay-to-a-2-5-billion-revenue-goal-heres-where-the-stock-could-go
  33. "Credo expands hyperscaler partnerships with Amazon, Microsoft, xAI" — Benzinga — 2025-03 — https://www.benzinga.com/25/03/44149586/credo-expands-hyperscaler-partnerships-with-amazon-microsoft-xai-amid-nvidia-powered-ai-boom
  34. CRDO analyst ratings (JPMorgan $335→$310, Evercore $325→$292 on 2026-09-02) — Benzinga — 2026-09-02 — https://www.benzinga.com/stock/crdo/ratings
  35. "Guidance on Section 232 semiconductor import duties for 2026" — C.H. Robinson — 2026-01-15 — https://www.chrobinson.com/it-it/resources/insights-and-advisories/client-advisories/2026q1/01-15-client-advisory-guidance-on-section-232-semiconductor-import-duties-for-2026/
  36. "Credo Launches 800G ZeroFlap Optical Transceivers Engineered for AI Networks" — Credo IR — 2026-03-17 — https://investors.credosemi.com/news-events/news/news-details/2026/Credo-Launches-800G-ZeroFlap-Optical-Transceivers-Engineered-for-AI-Networks/default.aspx
  37. "Credo Technology (CRDO) declines 32% after strong results" (52-week high $308.67 on 2026-06-22) — Tickeron — 2026-09 — https://tickeron.com/blogs/credo-technology-crdo-declines-32-after-strong-results-what-the-pullback-means-16520/
  38. Nvidia Q2 FY27 earnings coverage (search summary) — Nasdaq — 2026-08-27 — https://www.nasdaq.com/articles/nvidia-q2-earnings-beat-blackwell-ultra-data-center-strength
  39. Broadcom Q3 FY26 results (8-K Ex. 99.1; search summary) — SEC EDGAR — 2026-09-02 — https://www.sec.gov/Archives/edgar/data/0001730168/000173016826000076/avgo-08022026x8kxex99.htm
All CRDO Bulls vs Bears reports
Disclaimer. AI-generated research and podcast (Claude by Anthropic, voices by Google Gemini) for educational purposes — not financial advice and not a recommendation to buy or sell. Willem (MR.INVEST) owns this stock. Check the company’s own filings before making any decision.
← All Bulls vs Bears reports