Quick ReadPlain-language overview · read this first
Credo is a U.S. chip company that sells “connectivity cables” to AI data centers. The report’s rating is “Watch”: the business is excellent, but the stock is too expensive now, so watch it rather than chase it.
Its edge is a chip-equipped copper cable category it created. In AI data centers, thousands of chips need to communicate at high speed. Ordinary copper cannot transmit beyond 2 meters, while optical fiber is expensive and power-hungry. Credo installs one in-house chip at each end of a copper cable, extending the cable’s reach to 3 to 7 meters while making it more power-efficient and reliable. It holds about 75% of this market segment, and Amazon, Microsoft, and Google are customers.
The financials look very strong: revenue tripled this year to $1.335 billion, about $35 of net profit remains for every $100 of goods sold, and the company has about $1.4 billion of cash on the balance sheet with almost no debt.
Why only “Watch,” then? Because the price has already pulled forward several years of good performance and leaves no safety cushion. The bigger problem is customer concentration: the largest customer once accounted for 67% of annual revenue. In 2024, one large customer cut orders, leaving growth almost flat for the year. Add in executives selling only and not buying for more than a year, cashing out about $496 million, plus manufacturing concentrated in Greater China, and geopolitical friction becomes a problem. The report says it would be better to wait until the stock falls below $155 before seriously considering it.
The above only explains this report and is not investment advice. Stock markets involve risk; invest with caution.
LeadThe pioneer of the AI data center copper interconnect category (Active Electrical Cable, AEC) and the leader with roughly 75% share. FY2026 revenue tripled to $1.335 billion, with a 68% gross margin, 35% GAAP net margin, $1.4 billion in net cash, and almost no debt—a connectivity picks-and-shovels play that is top tier on both quality and growth. Rating Watch: a superb business bought at a price that has already prepaid years of flawless growth, with tail risks from customer concentration and geopolitics holding it down.
Prices in the article are as of publication; see the valuation band above for the live price.
Research Summary: The Top-Quality, Fastest-Growing Picks-and-Shovels Leader in AI Copper Connectivity, but the Price Has Already Prepaid Perfection
Start with the conclusion up front. Credo Technology (NASDAQ: CRDO) is a picks-and-shovels supplier of "connectivity" inside the AI data center, and its most valuable identity is a product category it created itself—the Active Electrical Cable (AEC).
The problem it solves is concrete: inside an AI cluster, hundreds or thousands of GPUs/accelerators need to communicate with one another at high speed. For short-reach interconnects within the rack, there have traditionally been two paths—cheap, power-efficient passive copper cable (DAC) that can only carry 1–2 meters, and optical modules (AOC/fiber) that reach far but are expensive, power-hungry, and have many failure points. Credo's AEC takes a third path: it places one in-house DSP/SerDes chip at each end of a copper cable, with the transmit side shaping the waveform and the receive side correcting channel impairments, extending the reach of copper from 1–2 meters to 3–7 meters while preserving copper's low power, low cost, and low latency. Versus optical modules, AEC uses about half the power, costs less, and achieves a mean time between failures (MTBF) of about 100 million hours—roughly 100x that of optical transceivers (Chipstrat AEC breakdown; Credo official blog).
Its market narrative today is "the copper-connectivity picks-and-shovels play for AI infrastructure." As hyperscalers like Amazon, Microsoft, Google, and xAI build out clusters of custom ASICs (in-house AI chips such as AWS Trainium, Google TPU, and Microsoft Maia), the racks need vast quantities of this kind of short-reach high-speed connectivity—and Credo holds roughly 75% share of the AEC segment, a real foothold in this wave of AI capex that others cannot easily replicate in the near term (CNBC, as-of 2025-10-17; Goldman's view via Kavout).
What has driven the stock over the past year? A violent inflection from "narrative" to "delivery." Credo's fiscal year ends in April. In FY2024 (ended April 2024), revenue was still stuck at $193 million, up just +5% year over year, and the company was still losing money. FY2025 jumped to $437 million (+127%, its first GAAP profit), and FY2026 tripled outright to $1.335 billion (+205.7%), with GAAP net income of $472 million, a 35.4% net margin, and a 68% gross margin (Credo Q4/FY2026 results, as-of 2026-06-01). The stock has gone from an IPO price of $10 to roughly $217 today (about 21x), up about +202% over the past year, with a market cap of about $40 billion. The market believes this is not an ordinary cyclical rebound but a structural ramp in AI copper-connectivity demand.
The most important bull-bear debate right now condenses into a single sentence: this is a flawless business, but at this price and this level of concentration, should you buy it? The bulls say growth is the fastest in the entire space, the gross margin is 68%, it is GAAP profitable, it has $1.4 billion in net cash, holds roughly 75% AEC share, has a patent moat (having already forced Amphenol into a licensing settlement), its valuation multiple is being caught up by earnings, and analysts are unanimously bullish. The bears say the absolute valuation is still extreme (P/E 87, EV/EBITDA 117), customer concentration is extreme (a single customer was 67% of revenue in FY2025), insiders have sold one-directionally across the board for 14 months to the tune of about $496 million, and both revenue and R&D have heavy exposure to Greater China—the geopolitical tail risk is real.
Qualitative portrait tag: top-quality, fastest-growing AI-connectivity picks-and-shovels play × extreme absolute valuation, held down by extreme customer concentration and geopolitical tail risk. Compared with Astera Labs (ALAB), which I covered earlier, CRDO is a mirror image—another AI-connectivity leader at an extreme valuation, but CRDO grows faster, is already GAAP profitable, holds net cash, and actually trades at a lower price-to-sales ratio (31x vs. Astera's 61x). So CRDO's problem is not "the most expensive among peers" (it is not); it is "the absolute valuation has already prepaid years of flawless growth, and the tail risks (concentration + geopolitics + insider selling) are hard enough." The investment lean and rating are held to section ten, derived naturally from the facts in the first nine sections.
I. The Vertical History: From a Small SerDes Shop of Ex-Marvell Engineers to the AI Copper-Connectivity Leader
To understand Credo today, start with its technical DNA and one near-fatal customer dependence.
2008—founded by a group of ex-Marvell engineers. Credo was founded by a group of high-speed interconnect engineers who left Marvell, centered on two co-founders still in their roles today: current CTO Lawrence Cheng (Chi Fung Cheng) and current COO Job Lam (Yat Tung Lam). The company spent its first decade "behind the scenes" honing its in-house SerDes (serializer/deserializer) IP—the physical-layer foundation of all high-speed data interconnect (Craft.co executives page; CNBC). One common misconception needs correcting: current CEO Bill Brennan joined in 2013 and is not a founder; his job was to commercialize the SerDes technology. And Lip-Bu Tan, often mentioned, was an early Credo investor and former chairman (founder of Walden International), not a founder or the CTO—he stepped down as Credo's chairman after becoming Intel's CEO in March 2025, keeping only a board seat.
2015–2020—the transition from IP licensing to AEC hardware. In 2015, Walden International led the first VC round; in 2017 came an early collaboration with Tesla's Dojo supercomputer. During this period Credo's business model was still primarily SerDes IP licensing (IP licensing was still about 15% of FY2024 revenue). The real strategic pivot was into hardware—Credo is a founding member of the HiWire Consortium, and the first AEC specification appeared in 2020. It upgraded "selling IP" into "selling systems": an AEC is not a copper wire but a full-line business of "one SerDes/DSP chip at each end + cable manufacturing + system integration + software" (Chipstrat).
2022-01—IPO, priced at the bottom of the range. Credo listed on Nasdaq on January 27, 2022, pricing at the bottom of the range at $10 and raising about $200 million, the first major U.S. tech IPO of that year (Bloomberg, as-of 2022-01-27). The listed entity is Cayman-incorporated Credo Technology Group Holding Ltd, which has no operations of its own and runs through subsidiaries in the United States, mainland China, and Hong Kong—a structure that matters a great deal to the geopolitical risk discussed later.
2023—a near-fatal customer dependence. This is the one entry in the vertical history most worth remembering. In February 2023, Credo's then-largest customer (pointing to Microsoft) lowered its product demand forecast for reasons unrelated to Credo's performance, directly hitting FY2023 Q4 and all of FY2024 revenue—as a result, FY2024 revenue nearly stalled at $193 million, up just +5% year over year (Credo FY2024 10-K, as-of 2024-04; background SemiAnalysis, as-of 2023-02-15). One big customer cutting orders was enough to flatten the whole year—this is the hardest fragility in the company's business model, and it will keep coming back.
2024–2026—the AI inflection + customer diversification + vertical integration. AI pushed AEC from supporting actor into the spotlight; Amazon replaced Microsoft as the largest customer, and revenue tripled from FY2025 to FY2026. Credo expanded in two directions at once: first, customer diversification, moving from "one customer at 67%" toward "four customers each above 10%"; second, extending into optical, developing an in-house 1.6T optical DSP (Bluebird), and in May 2026 completing the roughly $750 million acquisition of silicon-photonics company DustPhotonics and the roughly $92 million acquisition of Hyperlume, rounding out a "SerDes → DSP → silicon photonics → systems" vertically integrated stack to hedge the long-term risk of "copper being replaced by optical" (Credo completes DustPhotonics acquisition, as-of 2026-05-27).
Looking down the vertical history, Credo is a three-stage leap: "small SerDes IP shop → AEC category creator → AI copper-connectivity leader now extending into optical." Its technical foundation (in-house SerDes/DSP) is a real barrier, but that 2023 customer order cut is a reminder: its fate is highly tied to the capex cadence of a handful of hyperscale customers.
II. Financial Review: A Textbook High-Quality Breakout, but Three Definitions Must Be Seen Through
At first glance, Credo's financial statements show a textbook high-quality breakout, but three definitions must be seen through.
Revenue: an exponential curve, tripling from $193 million to $1.335 billion over three years. Zooming out (stockanalysis financials page, fiscal year ending April):
FY2021 $59 million → FY2022 $106 million → FY2023 $184 million → FY2024 $193 million (just +5%, the stall year from the big-customer order cut) → FY2025 $437 million (+127%, first GAAP profit) → FY2026 $1.335 billion (+205.7%);
The acceleration is clearer quarter by quarter: the four FY2025 quarters were $60M → $72M → $135M → $170M, and the four FY2026 quarters were $223M → $268M → $407M → $437M (Q4 +157% year over year, GAAP gross margin 68.2%, GAAP net income $169M);
FY2027 Q1 revenue guidance is $465–475 million (midpoint +6% sequentially), with management qualitatively signaling continued strength in FY2027, the market broadly expecting about +80% full-year growth, of which the optical business is expected to contribute about $600 million (Credo Q4 FY2026 8-K, as-of 2026-06-01; Seeking Alpha optical >$600M).
Profit quality: a 68% gross margin and 35% GAAP net margin—top tier for a hardware company. Credo's gross margin rose from 64.8% in FY2024 to 68% in FY2026, which is rare in connectivity hardware, an industry known for low margins—the root cause is that it sells "a system that includes an in-house DSP/SerDes chip" rather than a bare cable, and the higher the chip content, the better the margin. The GAAP net margin of 35.4% reflects a business that is already genuinely making money.
The first definition to see through—the $190 million gap between GAAP and Non-GAAP is almost entirely stock-based compensation. The company emphasizes Non-GAAP: FY2026 Non-GAAP net income was $661 million (a 49.5% net margin) and Non-GAAP EPS was $1.16 (Q4), while GAAP net income was only $472 million and GAAP EPS was $2.51 (full year). The roughly $190 million difference between the two comes almost entirely from stock-based compensation (SBC)—FY2026 SBC was about $183 million, or 13.7% of revenue (Q4 FY2026 8-K reconciliation table). This is a common ailment of high-growth semiconductors: Non-GAAP "adjusts out" employees' equity costs to make the margin look prettier. This report uses GAAP throughout for valuation and reminds readers that the 49.5% net margin the company promotes is watered down.
The second to see through—it is a net issuer of stock, not a repurchaser. Sitting on the balance sheet is about $1.443 billion in cash plus short-term investments, with only about $20 million of debt—almost no debt (Q4 FY2026 8-K), a flawlessly clean balance sheet. But this net cash is going toward acquisitions (DustPhotonics $750 million + Hyperlume $92 million) rather than shareholder returns; meanwhile, because of SBC + acquisition share issuance + option exercises, the company is a net issuer of stock—net equity issuance over the past year was about +$1.1 billion, and the share count keeps rising (FY2023 +65.8%, FY2025 +16.8%) (GuruFocus buyback rate). When reading EPS, remember the denominator is getting bigger.
The third—R&D intensity of 20.9%, the moat cost of this business. FY2026 R&D spending was $279 million, or 20.9% of revenue. What this money buys is generational leadership in SerDes/DSP (28G → 224G/lane), the root of the AEC foothold. High R&D intensity is not itself a problem, but it means this business requires sustained heavy investment to defend its technical position.
The financial portrait in one sentence: a textbook high-quality breakout—tripling growth, a 68% gross margin, a 35% GAAP net margin, net cash, no debt, with hardly a second like it among hardware companies. But you must look at it on a GAAP basis (Non-GAAP carries a 13.7% SBC dilution), remember it is diluting its share count, and note that the growth depends heavily on the capex of a handful of customers.
III. Business Model and Moat: A Real Foothold, with a Moat More Systematic Than Peers'
Credo's moat, unpacked from hard to soft, has four layers, and overall it is more systematic than pure-chip peers'.
Layer one (the hardest)—the underlying SerDes/DSP IP, developed in-house since 2008. An AEC is not "selling copper wire" but a system business of "one high-speed SerDes/DSP chip at each end + system integration." The design capability of this chip (signal processing from 28G to 224G/lane) is a decade-plus accumulation for Credo, and copper-cable makers simply do not have this DSP/SerDes capability. This is the fundamental barrier that distinguishes AEC from an ordinary cable business and the reason it can earn a 68% gross margin.
Layer two—the patent moat, already being monetized. This moat is not on paper: in March 2025 Credo sued Amphenol, Molex, TE Connectivity, and Volex at the U.S. ITC for infringing its AEC patents; in August 2025 it reached a settlement with Amphenol—Amphenol must obtain a license from Credo, while litigation against the other three continues (Credo-Amphenol settlement, as-of 2025-08-14). This is hard evidence that "AEC will not be freeloaded by the copper-cable giants"—even Amphenol, the connector leader, has to pay Credo a license fee before it can do AEC.
Layer three—systematic integration, more complete than pure-chip peers'. This is the key difference between Credo and Astera Labs. Astera is a pure-chip company (PCIe/CXL retimers, Scorpio switch fabric) and does not touch the cable system; Credo controls the full stack itself, from SerDes IP → retimer IC → full-line AEC → optical DSP → software (the PILOT diagnostic telemetry platform). When a cloud provider wants a complete solution to "get the AI cluster to stable operation faster, raise GPU utilization, and lower power" rather than a single chip, Credo's systematic integration is a stickier selling point (Credo PILOT launch).
Layer four (the softest, and also the outer limit)—customer-certification stickiness, whose flip side is customer concentration. Once you win a design-win in a given generation of a hyperscaler's platform, switching costs are high and you can capture the entire life cycle of that platform. But this "moat" has a dangerous flip side: it ties your fate to a very small number of customers (see the concentration risk in sections six and eight).
The essence of the business model: it sells "connectivity systems that include chips" to hyperscale cloud providers, earning hardware money by their AI-cluster shipment volume (product sales are about 97% of revenue), with gross margin propped up to 68% by chip content. This is a picks-and-shovels business of "heavy front-loaded R&D, earning a premium through generational technical leadership + patents + system integration."
The honest moat conclusion: Credo's moat is real and more systematic than pure-chip peers'—in-house SerDes/DSP (copper-cable makers don't have it) + patents (Amphenol must now license) + full-stack integration (Astera doesn't do cables). But the depth of its moat is constrained by two things: first, the highest-value connectivity (within Nvidia's NVLink domain) is self-supplied by Nvidia, and Credo mainly captures non-Nvidia custom ASICs and Ethernet networking; second, the customer side of this moat is extremely concentrated—it is a combination of "strong barrier + high concentration."
IV. Industry Cycle and TAM: Copper Connectivity Is a Hard Requirement for the AI Rack, but the TAM Definitions Must Be Honest
Credo's tailwind rests on a fact that has been repeatedly confirmed: the compute expansion of AI clusters is being bottlenecked by connectivity density inside the rack.
Why it is a hard requirement. Each generation of AI accelerator (Nvidia Blackwell/Rubin, various custom ASICs) doubles in performance, and the number of back-end network ports and cables between GPUs doubles along with it. Short-reach interconnect within the rack (scale-up) currently relies almost entirely on copper—because over a few meters, copper beats optical across the board on power, cost, latency, and reliability. SemiAnalysis notes that the TAM for scale-up interconnect within the rack "has already significantly exceeded the scale-out network" (SemiAnalysis CPO report). This is precisely AEC's home turf.
TAM—the definitions must be honestly separated and not exaggerated:
AEC chip market: 650 Group estimates growth of about 64% per year, reaching about $1 billion by 2028 (note it is about $1 billion, not the $4 billion in some second-hand claims) (650 Group AEC report);
Active copper cable + passive copper cable + optical cable combined: LightCounting puts 2028 at about $2.8 billion (the three combined, which cannot be attributed solely to AEC);
The overall data center interconnect (DCI) market: about $3.7–4.2 billion, but reaching that by 2032 and on a definition far broader than AEC—it cannot be conflated with AEC;
Credo's self-reported adjacent opportunity: PCIe retimer TAM exceeding $1 billion by 2027, seen as a "second growth wave" that could double its addressable market (BeyondSPX analysis).
These numbers are on different definitions and are nested, and they must never be added together. Honestly: AEC is a niche that "grows extremely fast but is still not large in absolute size" (on the chip definition, only about $1 billion by 2028), and Credo already did $1.335 billion in FY2026 revenue—its growth cannot rely on the AEC segment alone but must come from share (about 75%) + lateral expansion into optical DSP / PCIe retimers / 1.6T upgrades. This is also the fundamental reason it is rushing to acquire silicon photonics.
The 800G → 1.6T upgrade cycle. The industry is shifting from 800G ports to 1.6T to support the new generation of GPUs, and each doubling of speed requires a SerDes/DSP upgrade (112G → 224G/lane), a continuing replacement dividend for both of Credo's lines, AEC and optical DSP. Credo is already sampling 1.6T AEC and DSP (CRDO 1.6T deep dive).
The biggest space-level risk—copper being replaced by optical. Credo's AEC thesis bets entirely on "short-reach copper within the rack remaining the medium of choice." But if the power and reliability of co-packaged optics (CPO) / silicon photonics cross the critical threshold faster, AEC's addressable market will be compressed from above. This is not fantasy—it is exactly why Credo itself spent more than $800 million acquiring DustPhotonics and building in-house silicon-photonics capability as a hedge (Vik's Newsletter, CPO and AEC). In other words, even Credo is buying insurance against "the end of copper."
Industry conclusion: Credo stands on a real, strong tailwind (AI-rack copper connectivity is a hard requirement, with about 75% share), but the single-segment AEC TAM is not large, growth must come from lateral expansion and the upgrade cycle, and a long-term sword of "copper being replaced by optical" hangs overhead. The space is good, and within it Credo "has locked up the fattest copper-connectivity segment but must keep running into adjacent areas."
V. Horizontal Comparison: The Fastest Grower in AI Connectivity, Not the Most Expensive, with Multiples Still Compressing
Placing Credo into the AI-connectivity peer coordinate system yields several counterintuitive but important conclusions.
Valuation comparison (as-of 2026-06-04 close, multiples from stockanalysis):
| Company | Price | Market Cap | P/E-TTM | P/S-TTM | Revenue Growth | Gross Margin | Positioning |
|---|---|---|---|---|---|---|---|
| Credo (CRDO) | $217.50 | $40.1B | 86.7 | 30.9 | +206% | 68.0% | AEC + SerDes/optical DSP, vertically integrated |
| Astera Labs (ALAB) | $358.05 | $61.4B | 242.1 | 61.3 | +93% | 76.0% | PCIe/CXL retimers, Scorpio fabric |
| Marvell (MRVL) | $316.43 | $276.8B | 109.3 | 31.8 | +42% | 51.5% | Custom ASIC + optical DSP + interconnect |
| Broadcom (AVGO) | $418.91 | $1.98T | 81.6 | 26.3 | AI segment +143% | 76.3% | SerDes/optical/switch giant + custom ASIC |
| Amphenol (APH) | $146.77 | $180.6B | 42.2 | 7.0 | +54% | 37.9% | High-speed copper cable/connectors (AEC competitor, now must license) |
| MACOM (MTSI) | $382.74 | $29.2B | 164.0 | 27.2 | ~23% | 55.7% | Analog/RF, optical communication chips |
| Coherent (COHR) | $421.90 | $82.5B | 176.1 | 12.5 | — | 37.0% | Optical modules/lasers |
| Semtech (SMTC) | $169.35 | $15.8B | Loss | 14.5 | — | 52.4% | Active copper cable DSP (LinkX) |
Company by company + three key conclusions:
Conclusion one: CRDO is not the most expensive in the group; Astera Labs is. On price-to-sales, ALAB's 61.3x is nearly double CRDO's (30.9); on price-to-earnings, ALAB at 242x, COHR at 176, and MTSI at 164 are all higher than CRDO's 87. CRDO's P/S is on par with Marvell's (31.8), placing it at the "upper edge of the second tier." Within the AI-connectivity group, CRDO's valuation is "expensive, but not the most outrageous."
Conclusion two: CRDO grows the fastest in the field, with a first-tier gross margin. Its +206% revenue growth is the highest in the group (ALAB +93%, AVGO AI segment +143%, MRVL +42%); its 68% gross margin trails only AVGO/ALAB's 76% and is far above MRVL (51.5%) and APH (38%). From a PEG (valuation/growth) standpoint, CRDO's very high multiple is supported by current-period growth, so the mismatch is actually lighter than Astera's.
Conclusion three: CRDO's multiple is being caught up by earnings rather than hitting new historical highs. This matters—CRDO's absolute stock price is at a new high, but its price-to-sales has compressed from a historical peak of 54.3x to 31x, and its price-to-earnings from a three-year average of 240x to 87x (GuruFocus P/S). In other words, the stock has risen, but profits have risen faster, and the valuation multiple is actually converging. This is the opposite of the bubble pattern where "valuation gets bid ever higher."
Versus the direct competitor Astera Labs (ALAB). The two are mirror images in AI connectivity: Astera is strong in PCIe/CXL retimers (the industry-first Aries 6) and the Scorpio switch fabric and is a pure-chip company; Credo is strong in AEC copper-cable systems (which Astera does not do) and is extending into optical. Credo grows faster, is already GAAP profitable, holds net cash, and has a lower P/S; Astera has a higher gross margin, a lighter pure-chip model, but a more extreme valuation. I earlier rated Astera "Watch" as well—within the same group, CRDO actually has better quality and a better value proposition.
The key takeaway of this section: CRDO's "expensiveness" must be read in context—within AI-connectivity peers, it grows the fastest, is not the most expensive, and its multiples are still compressing. Its risk is not "more expensive than peers" but "the absolute valuation is still extreme (P/E 87/EV-EBITDA 117) + customer concentration and geopolitical tail risk"—which is the main thread of sections six and eight.
VI. Current Fundamentals, Bull vs. Bear: Quality Colliding with Concentration and Geopolitics
Put the hardest arguments from both sides on one table.
Bulls (where the bull case is hard):
The fastest grower in the entire space + FY2027 still guided to about +80%. FY2026 +206%, Q4 +157%, FY2027 Q1 guidance another +6% sequentially, full-year market expectation about +80%, optical contributing about $600 million.
Top-quality: 68% gross margin, 35% GAAP net margin, $1.4 billion net cash, no debt. Nearly one of a kind among hardware companies.
About 75% AEC share + a patent moat. The category creator; Amphenol must now license; copper-cable makers lack DSP/SerDes capability.
The valuation multiple is being caught up by earnings, and it is not the most expensive in the group. P/S compressed from 54 to 31, P/E from 240 to 87; Astera (P/S 61) is the most expensive.
Customers are diversifying. Moving from "one customer at 67%" toward "four customers each above 10%," with Amazon, Microsoft, Google, and xAI all using it.
Analysts are unanimously bullish. About 17 firms cover it, 16 buy, 1 hold, 0 sell, with a consensus target of about $248–251, +14% above the current price.
Bears (where the bear case is hard):
The absolute valuation is still extreme. P/E 87, EV/EBITDA 117, P/S 31—no matter how the multiples compress, this is a price that prepays years of flawless growth, with a very thin margin of safety.
Customer concentration is extreme, and FY2024 already proved it can be fatal. In FY2025 a single customer was 67% of revenue and the top ten about 90%; although recent quarters diversified to "four customers at 34%/27%/16%/10%," the largest is still about 34% and the top two are a majority. In 2024 the whole year stalled because one big customer cut orders—this is not a hypothetical risk; it is a fact that happened.
Insiders sold one-directionally across the board for 14 months, totaling about $496 million, with zero buying. The CTO sold about $232 million, the COO about $100 million, the CEO about $85 million, and the CFO and directors all sold, running through the entire move from about $90 to about $190 (OpenInsider CRDO). Although mostly 10b5-1 programmatic, the across-the-board, continuous, one-directional pattern is a strong signal at a high.
Large Greater China geopolitical exposure. About 60% of long-lived assets are in Taiwan, 14% in mainland China, and 10% in Hong Kong; in FY2025 over 70% of revenue was geographically attributed to Greater China (mostly assembly in China supplying U.S. customers). U.S. export controls or Chinese countermeasures could directly hit the supply chain.
Copper being replaced by optical over the long term. The AEC thesis bets on short-reach copper, and if CPO/silicon photonics accelerate across the threshold they will compress the AEC market—even Credo is spending $800 million on silicon photonics as a hedge.
Net stock issuer + 13.7% SBC. Behind the Non-GAAP cosmetics, real shareholder returns are diluted by SBC and share issuance.
The net verdict after the collision: the bull points—top quality, fastest growth, highest share, multiple compression—all hold, and this business's excellence is beyond doubt, and its value proposition within AI-connectivity peers is actually better than Astera's. But the bear arguments point to one conclusion: the problem is not business quality but price and tail risk. When a stock has a P/E of 87, its largest customer can be more than a third of revenue, insiders are selling across the board at a high, and it carries Greater China geopolitical exposure, its margin for error against "any single tail event" is very thin. This is not a judgment about "whether the foothold will work out" (it already has); it is a judgment about "whether the risk-reward at this price and this concentration is worth it"—and the answer is: the quality deserves the highest attention, but the price gives no margin of safety.
VII. Valuation: For a High-Growth Stock, See Through the Multiple; the Current Price Sits in the Fair Band, with Upside Resting on Beats
The biggest trap in valuing a stock growing +206% is using a static P/E—the denominator is growing fast. Here I build three scenarios around "earnings across the growth path + peer multiples + the analyst anchor." Definitions: about 184 million shares, FY2026 GAAP EPS of $2.51, and a market expectation of about +80% FY2027 revenue (about $2.4 billion).
Bear $110–150: scenario = growth decelerates sharply (a big customer cutting orders / inventory digestion as in FY2024), or AI capex cadence slows, and the market compresses the valuation back to the normal of "high growth but single-customer dependence" (P/S about 8–11x). This band also roughly covers the lower half of insiders' 2025 selling range (about $90–190) and the 52-week-low area.
Base $180–250: scenario = FY2027 revenue delivers about +80%, gross margin holds at 68%, customer diversification continues, and the market grants a high-quality growth premium. The current price of $217.50 sits right in the middle of this band; the analyst consensus target of $248–251 sits at the upper edge of this band. This is the neutral pricing that "top quality + fastest growth" deserves.
Bull $290–380: scenario = AEC upgrades smoothly to 1.6T + the optical business (including DustPhotonics silicon photonics) ramps into a second growth engine + PCIe retimers open a second wave + customers diversify further, and the market discounts years of high growth in advance. Roth's high-end target of $300 sits at the lower edge of this band.
The current price of $217.50 sits in the middle of the fair band ($180–250), with the analyst consensus target of $248–251 at the upper edge, implying about +14% upside. This means: the current price has already priced the AI copper-connectivity foothold and high growth to "fair," and the market sees mild further upside (from beating expectations, not from a valuation re-rating). Going higher requires FY2027's +80% growth + the optical business actually ramping; going lower, if concentration risk or a geopolitical event is triggered (a big customer cuts orders, export controls), a return to the lower edge of the fair band or even the bear band is a realistic risk.
Ideal buy zone: at or below $155 (fair_buy_price set at 155). The reasoning: to get a margin of safety on a high-growth stock at this kind of extreme P/E of 87, the price needs to fall below the lower edge of the fair band (180), near the bear-to-fair transition zone—roughly corresponding to about 35x forward FY2027 earnings, and below the level of dense insider selling. $155 is a ceiling that satisfies both "recognizing the top quality and foothold, willing to hold long term" and "leaving a buffer for extreme customer concentration + geopolitical tail + insider selling." This is not bearish on the business; it is requiring an entry point that is not at a historical high and does not chase before the tail risks have been released.
Valuation uncertainties that must be honestly flagged: ① a high-growth stock's P/E is extremely sensitive to the growth assumption, and FY2027's +80% is a market expectation, not hard company guidance (the company gave only Q1 guidance); ② the three bands use GAAP, while the company's preferred Non-GAAP carries a 13.7% SBC dilution and the share count is being diluted; ③ the single-segment AEC TAM is not large, the bull band depends on optical/PCIe lateral expansion delivering, and the execution risk is not small.
VIII. Risks: Customer Concentration, Absolute Valuation, Geopolitics, Insider Selling, and Copper Being Replaced by Optical—Five Swords
Ranked by their damage to the investment judgment, from high to low:
Risk one · extreme customer concentration, already proven fatal (the hardest). In FY2025 a single customer (pointing to Amazon, directly tied via warrants) was 67% of revenue and the top ten about 90%; although recent quarters diversified to "four customers at 34%/27%/16%/10%," the largest is still about a third and the top two are a majority. Most critically—FY2024 stalled the whole year (revenue just +5%) because one big customer cut orders, and this is not hypothetical but a fact that happened. Any change in the ASIC capex cadence of a single hyperscaler could violently swing Credo's growth curve.
Risk two · extreme absolute valuation, very thin margin of safety. P/E 87, EV/EBITDA 117, P/S 31—no matter how "cheap" relative to peers or how much the multiple is compressing, this is a price that prepays years of flawless growth. Any "merely in-line" earnings could be treated as bad news at this valuation level.
Risk three · large Greater China geopolitical exposure. About 60% of long-lived assets are in Taiwan, 14% in mainland China, and 10% in Hong Kong; in FY2025 over 70% of revenue was geographically attributed to Greater China (assembly in China supplying U.S. customers). Against the backdrop of 2026 U.S.-China tech friction, export controls, supply-chain decoupling, and geopolitical conflict could all directly hit its manufacturing/assembly/part of its R&D—a structural tail underestimated by the market narrative.
Risk four · insiders selling one-directionally across the board at a high. Over the past 14 months, about 100 Form 4 filings were all sales, zero buys, totaling about $496 million, with the CEO/CFO/CTO/COO/multiple directors all without exception. Although mostly 10b5-1 programmatic plans, "across the board, continuous, one-directional, running through the entire doubling of the stock" is a high-level signal that should not be ignored—the people who know the company best are systematically cashing out.
Risk five · copper being replaced by optical over the long term + SBC dilution. AEC bets on short-reach copper, and accelerated CPO/silicon photonics would compress its market (Credo itself spent $800 million on silicon photonics as a hedge); meanwhile, net stock issuance + 13.7% SBC continuously dilutes shareholders, and the Non-GAAP pretty margin is watered down.
Stacking the five: Credo's risk is not "whether the business is good" (top tier) but "the combination of buying at the extreme valuation of P/E 87 + the largest customer being able to take a third of revenue + insiders selling across the board at a high + Greater China geopolitical exposure"—any single tail event will be amplified at this no-margin-of-safety position.
IX. Catalysts: Upside Rests on Delivery, Downside on the Tail, with Relatively Balanced Timing
Upside catalysts:
FY2027 about +80% growth + the optical business ramping. The 1.6T AEC upgrade, optical DSP (including DustPhotonics silicon photonics) ramping to about $600 million, and PCIe retimers opening a second wave—any beat is a strong upside catalyst.
Further customer diversification: a fifth and sixth hyperscale customer crossing the 10% threshold, with Neo Cloud / second-tier cloud combined reaching 20% of revenue, would significantly ease the concentration discount.
Holding AEC share + winning the patent suits: another win in the ITC litigation against Molex/TE/Volex would consolidate the moat.
Downside catalysts:
A big customer cutting orders / ASIC cadence slowing: any move by a customer that is more than a third of revenue could replay the FY2024-style stall—this is the heaviest downside catalyst.
Geopolitical events: escalating export controls on China, or a Taiwan-Strait / supply-chain shock, would directly hit its Greater China manufacturing and revenue.
A quarter's growth or gross margin missing expectations: at the P/E 87 level, even one "merely in-line" print could trigger a violent pullback.
Continued insider selling + a heating CPO-replacement narrative: non-stop high-level selling, or CPO/silicon photonics being proven to replace copper faster, would both compress the valuation.
The timing of the catalysts is relatively balanced (unlike AMKR, where the bad news comes first): on the upside there is a real delivery path of FY2027 growth + the optical ramp, and analysts grant +14% upside; but on the downside, the three swords of concentration, geopolitics, and valuation could be activated at any time by some tail event. The net timing is "upside visible but needing delivery, downside tail random."
X. The Horizontal-Vertical Intersection Summary: A Top-Tier Picks-and-Shovels Play, Bought at a Price With No Margin of Safety
Intersecting the vertical (history and quality) with the horizontal (peers and valuation), the conclusion is clear.
Vertically, this is a top-quality AI-connectivity picks-and-shovels play with a real foothold. From a small SerDes shop of ex-Marvell engineers to the AEC category creator and the AI copper-connectivity leader with about 75% share, Credo's foothold is real and systematic—in-house SerDes/DSP (copper-cable makers don't have it), a patent moat (Amphenol must now license), full-stack integration (Astera doesn't do cables), and an extension into optical to hedge the end of copper. FY2026 tripled to $1.335 billion, with a 68% gross margin, a 35% GAAP net margin, and $1.4 billion net cash—financial quality nearly one of a kind among hardware companies.
Horizontally, the valuation is not the most expensive, and the multiple is still being caught up by earnings. Within AI-connectivity peers, Credo grows the fastest in the field (+206%), has a first-tier gross margin, and its price-to-sales (31) is actually below Astera's (61) and on par with Marvell's (32), and its P/S has compressed from a historical peak of 54 to 31 and its P/E from 240 to 87—it is not "getting bid ever higher" but "earnings catching up to valuation." Analysts are unanimously bullish, with +14% upside. From this angle, CRDO's quality and value proposition within the group are actually better than Astera, which I previously also rated "Watch."
The judgment at the intersection of the two lines: a top-tier picks-and-shovels play, bought at a price with no margin of safety. This business's excellence is beyond doubt and its value proposition within peers leans favorable; but the current price (P/E 87, EV-EBITDA 117) has already prepaid years of flawless growth, while it carries three hard tail swords—the largest customer can take more than a third of revenue (FY2024 already stalled from an order cut), insiders sold across the board at a high for 14 months for about $496 million, and Greater China geopolitical exposure is over 70% of revenue. This is a top-tier name that deserves a place at the very top of the watchlist, to be seriously considered after the price falls back or the tail risks are released—not a name to chase now at a historical high and a P/E of 87.
Rating: Watch. This is not bearish on the business (the quality deserves the highest attention, and the value proposition is even better than the group's Astera); rather, the risk-reward at this price, this concentration, and this geopolitical tail is not attractive enough—the margin of safety is too thin. The ideal buy zone is at or below $155; whether FY2027 growth and the optical ramp deliver, the pace of customer diversification, the AEC 1.6T upgrade and share, and marginal changes in insider selling and the geopolitical situation are the re-rating signals to watch closely.
Pre-mortem: If, Three Years From Now, "Buying at $217.50" Looks Like a Mistake, the Most Likely Reasons Are
A big-customer order cut replays FY2024: a hyperscale customer that is more than a third of revenue adjusts its ASIC capex or shifts to a competitor / in-house, the growth curve halts abruptly, and the market mercilessly kills the valuation at a P/E of 87—a script that has played out once before.
The timing of the absolute valuation: buying at a P/E of 87 and EV-EBITDA of 117, even if the fundamentals keep growing fast, makes no money for a long time because "the price was prepaid"—a good company, a bad entry point.
A geopolitical black swan: escalating export controls on China or a Taiwan-Strait / supply-chain shock hits its Greater China exposure of over 70% of revenue, directly interrupting manufacturing and delivery.
Copper being replaced by optical accelerates: the power and reliability of CPO/silicon photonics cross the critical threshold faster, AEC's addressable market is compressed, and Credo's silicon-photonics hedge has not yet come into its own—the core moat eroded from above.
Insider selling was prescient: the across-the-board, one-directional selling at a high turns out, in hindsight, to be management's true judgment on the sustainability of growth / the valuation, and outside investors took the last leg.
Conversely, if this purchase is right three years from now, it will certainly be because: AI copper-connectivity demand keeps beating expectations, AEC share is held and upgraded smoothly to 1.6T, the optical business (including silicon photonics) ramps into a genuine second growth engine, customers diversify to the point that no single one exceeds 20%, and the market is willing to pay a long-term premium for this top-tier picks-and-shovels play. This script exists, and because earnings are catching up to valuation it is not a low-probability one—but it requires surviving the volatility of an extreme valuation and requires not a single tail risk detonating. Holding the same top-tier script at a lower price (at or below $155, ideally a low printed by some concentration panic or geopolitical shock) gives much better odds.
Key Data Table
| Dimension | Value | Definition / Source |
|---|---|---|
| Current price | $217.50 (prior close $214.60) | as-of 2026-06-05 |
| Market cap | ~$40.1B | 184.4M shares × current price |
| 52-week range | $66.75 – $245.95 | up about +202% over 1 year; about +2,075% from IPO ($10) to date |
| P/E-TTM / Forward P/E | 86.7× / 46.7× | high-growth stock, denominator rising fast |
| P/S-TTM / EV-EBITDA | 30.9× / 117× | P/S compressed from a historical peak of 54 to 31; not the most expensive in the group (ALAB P/S 61) |
| P/B | ~22.8× | net cash, asset-light |
| FY2026 revenue | $1,335.1M (+205.7%) | fiscal year ending 2026-05; tripled |
| FY2024 revenue | $193.0M (+5%) | the stall year from the big-customer order cut (a cautionary precedent) |
| Q4 FY2026 revenue | $437.0M (+157% YoY) | GAAP gross margin 68.2%, GAAP net income $169.1M |
| FY2027 Q1 guidance | revenue $465–475M, GAAP gross margin 66.9–68.9% | full-year market expectation about +80%, optical about $600M |
| Gross margin | 68.0% (FY2026 GAAP) | first tier among hardware companies |
| Net margin | GAAP 35.4% / Non-GAAP 49.5% | difference of about $190M ≈ SBC (13.7% of revenue) |
| GAAP / Non-GAAP EPS | full year GAAP $2.51 / Q4 Non-GAAP $1.16 | valuation uses GAAP throughout |
| Net cash | ~+$1.42B (cash + short-term investments $1.44B, debt about $20M) | almost no debt |
| R&D intensity | 20.9% ($279.4M / revenue) | moat cost |
| Capital allocation | net cash directed to M&A (DustPhotonics $750M + Hyperlume $92M), not buybacks | net stock issuer (SBC + issuance dilution) |
| AEC market share | ~75% | category creator; Amphenol must now license |
| Customer concentration | FY2025 single customer 67%, top ten 90%; Q4 FY2026 four customers 34/27/16/10 | diversifying, but still extreme |
| Greater China exposure | long-lived assets Taiwan ~60% / mainland 14% / HK 10%; FY2025 revenue >70% geographically attributed to Greater China | geopolitical tail |
| Insider selling | past 14 months about 100 filings all sales, zero buys, totaling ~$496M | CEO/CFO/CTO/COO/directors all (mostly 10b5-1) |
| Analysts | about 17 firms, 16 buy / 1 hold / 0 sell, consensus target ~$248–251 | +14% above current price |
| Founders / CEO | co-founders Lawrence Cheng (CTO) + Job Lam (COO) (ex-Marvell); CEO Bill Brennan (joined 2013) | Lip-Bu Tan is former chairman/investor, not a founder |
| Rating | Watch | top-tier picks-and-shovels play, price with no margin of safety |
| Ideal buy zone | ≤ $155 | below the lower edge of the fair band, leaving a tail-risk buffer |
Research Uncertainties and Definitions
Price/market-cap anchor: the current price of $217.50 is on a 2026-06-05 basis (EODHD real-time + WebSearch cross-check, with recent intraday $200–227 high volatility and a 52-week high of $245.95); market cap is about $40.1B at 184.4M shares × current price (some data sources use $42B by applying a higher instantaneous price).
Fiscal-year definition: Credo's fiscal year ends in late April / early May each year, so FY2026 = through 2026-05-02. "Tripled" and "+157%" are the FY2026 full-year and Q4 single-quarter figures respectively, and should not be mixed.
GAAP vs. Non-GAAP: the company emphasizes Non-GAAP (49.5% net margin), while this report uses GAAP throughout (35.4% net margin); the roughly $190M difference is almost entirely SBC (13.7% of revenue) + acquisition costs + intangible amortization. Non-GAAP EPS also reduces diluted shares for "future unrecognized SBC buyback shares."
Customer-concentration definitions must be dated: FY2025 full-year "single customer 67%" and Q4 FY2026 "four customers 34/27/16/10" are of completely different magnitudes; there is also "Amazon 61% / Microsoft 12% / xAI 11%," "top two 84%," and other data on different periods/definitions, so always specify the quarter when citing. The largest customer "= Amazon" is a high-confidence inference (direct warrant evidence); the 10-K discloses it anonymously only as "Customer A."
TAM definitions (a key correction): the AEC chip market is estimated by 650 Group at about $1 billion in 2028 (not the $4 billion in some second-hand claims); AOC+DAC+AEC combined is about $2.8 billion per LightCounting for 2028; the overall DCI market is $3.7–4.2 billion but corresponds to about 2032—the three are on different, nested definitions and must never be added. Credo's "~75% AEC share" and "PCIe retimer TAM >$1 billion / 2027" are on company/analyst definitions.
FY2027 growth: about +80% and optical of about $600 million are market expectations / management qualitative remarks, with the company's hard guidance going only to Q1 FY2027; the bull band depends on lateral expansion delivering, and the execution risk is large.
Management fact correction: the founders are Lawrence Cheng (CTO) + Job Lam (COO, both ex-Marvell), and CEO Bill Brennan joined in 2013 and is not a founder; Lip-Bu Tan was an early investor / former chairman (now Intel's CEO), not a founder or the CTO; the early investors were mainly Walden/Celesta/BlackRock/Chinese capital (no evidence of IDG as a major pre-IPO investor).
Insider selling: the about $496M all-sales figure is taken from OpenInsider aggregation (including 10b5-1 programmatic plans); different sources give smaller figures such as $50M (recent months) due to different sample periods, but the direction is consistent.
This report is independent third-party research; all load-bearing numbers are annotated with sources and dates, and key financials have been cross-checked against company filings (10-K/8-K/earnings calls) plus multiple third-party sources. This report is based on public information and does not constitute investment advice. Markets carry risk; invest with caution.
Full report
Sign in to read the full report
Sign up free to unlock the full text, the Baillie growth scorecard, and full-text search.
Log in / Sign up free