Credo Technology Group Holding
The wiring between AI chips: cables with a signal chip in each end, optical parts, and the chips other makers put inside their own.
The scores
Growth3/4 ↑
Seven quarters above 100% and organic, but two customers are 71% of it and one product line drove almost all of it.
Business quality3/4 →
Qualification really does lock a socket, but only for one technology generation, and the next lane-rate step has not been won yet.
Financial position4/4 →
No debt at all, no maturities, cash generation positive, and a $769.6M acquisition paid from cash with $764.3M still on hand.
Execution record3/4 ↑
Eight straight beats and a raised guide, but four years public and the $1.25B of acquisitions are too recent to judge.
Per share capture2/4 ↓
Share pay is 18.4% of revenue, the count rose 9.7% in fifteen months, a deal was part-funded in stock, and no buyback exists.
Potential2/4 →
The prize is large and the base does not compose; and the upside is hyperscaler spending decisions with $4.2M contracted behind them.
Risk profile2/4 ↑
Two customers at 71% who can stop with limited notice and no penalty, plus one exclusive foundry - externally controlled, resolving in one step.
Valuation3/4 ↑
Down 13.6% since 25 September with no new print, the price still needs the plan delivered in full, with no room left to arrive late.
Evidence quality3/4 →
Every figure is filed and the rivals confirm each other, but the entire forward half rests on management claims graded rather than verified.
If it works 1.5× · if it fails you keep about 100%.
The Verdict
We rate Credo neutral. It sells the cables and chips that let thousands of AI processors talk to each other, and its sales have grown very fast. The business is mixed: real and growing, but operating profit fell as sales rose, and little of this year's plan is under contract. The price is fair: about right if management's plan is delivered in full, with no cushion if the second half of the year slips. As a bet it is defensive: the copper cable business alone covers today's price, so if the new optical line works the shares gain about a third, and if it sells nothing little is lost.
The long caseBigger AI clusters need far more optical links, and Credo wins if its chips go into modules other firms build, against much larger rivals already there.
The evidence under this - 8 claims
- Nothing is contracted behind it: $4.2M of orders are recorded as contracted against a year guided at about $2.47B, and the 10-K states there are no long-term purchase commitments.
- The same management guided fiscal 2026 at roughly half what it delivered, so the record argues the guide is conservative rather than stretched - which cuts against treating the raise as a warning sign.
- Four analysts on the Q1 call pressed on the second-half ramp and on the ability to scale supply for it; the answers were about engagement and supply-chain preparation, and named no order.
- Management has described this shape in advance twice - in June 2026 as mid-single-digit sequential growth in the first half with an inflection in the second, and again in September - so the back-loading is disclosed rather than discovered.
- The first half is running ahead of what was guided for it: Q1 came in above the high end and Q2 is guided at about 10% sequential against the mid-single digits described in June, which lowers what the second half has to do relative to the original plan.
- Both forward figures are struck on management's own guide and management's own adjusted profit measure, which excludes share pay running at 18.4% of revenue. On reported earnings the same guide implies something closer to 38 times.
- The guide has been raised once already and the company has beaten its own revenue range in each of the last two quarters, so the denominator may prove conservative.
The company
Inside an AI data centre, thousands of processors swap data constantly, and one failing link can stall the cluster. Credo sells those links: copper cables with a signal chip in each end, plug-in optical parts, the signal chips others build into theirs, and software that watches them all.
The buyer's alternatives are plain copper, which stops working reliably as speeds double, or optical cables, which work but use more power and cost more.
The evidence under this - 2 claims
How it makes money
It sells parts, mostly to a handful of the largest cloud companies and the manufacturers building for them. Two customers were 71% of sales in the latest quarter. Nothing is ordered far ahead: the company has no long-term purchase commitments, so each quarter's sales are won from the designs it is already inside.
The evidence under this - 2 claims
Where the money comes from, stream by stream
| Stream | Share of sales (FY2026) | How it earns it |
|---|---|---|
| Active electrical cables | not disclosed | ZeroFlap copper cables with Credo's signal-processing chip at both ends, linking servers to switches inside hyperscale AI clusters. |
| Optical: transceivers, silicon photonics and optical DSPs | not disclosed; guided above $600M of fiscal 2027 | ZeroFlap optical modules, DustPhotonics silicon photonics chips and the DSP chips module makers buy, sold into plug-in optics. |
| Retimers, line-card PHYs and SerDes chiplets | not disclosed | Chips that clean up signals on boards and inside switches and processors, sold to equipment and chip makers. |
| SerDes IP licensing | 2.9% | Licence fees for Credo's serialiser designs, which other chip companies build into their own silicon; 14.5% in fiscal 2024. |
| Product engineering services | 2.8% | Paid engineering to tailor products to a customer's design before volume shipments. |
Who runs it
Bill Brennan has run Credo since 2014; its two co-founders still run technology and operations, and the finance chief has been there since 2015. Insiders own about 9%, mostly a board member and the co-founders, not Brennan. Their bonus rides on sales and profit growth that leaves out stock pay, and last year's targets were set so low they paid the maximum. Brennan's big new award pays only if sales and the share price both climb a long way.
The evidence under this - 4 claims
Insiders filed 716 open-market sales and no purchases in the year to 15 September 2026 (C29), so the stake is being run down, mostly under plans adopted in advance.
- The goals were set before the year's acceleration and the fiscal 2027 cycle moved goal-setting to after the year's results so goals can be better calibrated, per the proxy; the plan also requires a minimum level of performance before any bonus pays.
- No metric pays on per-share results, cash conversion or dilution, so share pay of 18.4% of revenue (C06) costs the plan nothing.
Why it wins, and who could beat it
Why customers come back
A link is approved as a whole system: Credo's cable is tested at full speed against the buyer's own switch and network card. Changing supplier means running that approval again while very expensive processors sit idle, and management says failing links already cost customers 10% to 20% of their processors' time. Its rival Astera Labs names the same near-zero tolerance for defects as how this market competes.
The lock lasts one technology generation. Credo's own annual report says prices in its markets fall over time.
The evidence under this - 4 claims
- It is the seller relaying what buyers told him. No hyperscaler publishes cluster utilisation, and no independent body measures the cost of link failure, so the figure cannot be checked from outside.
- It is the cost of the problem across a whole cluster, not the share Credo's products address - a link is one of several failure modes, and the company sells into only part of the link budget.
The threat
Marvell's annual report lists every chip Credo makes and names Credo as a direct rival. Astera Labs sells cable parts for the very product Credo created. And NVIDIA now builds the optical engine into its switches, which removes the plug-in optical part altogether at that end of the link.
The evidence under this - 4 claims
- NVIDIA's own announcement says its photonics will grow alongside pluggable optical transceiver technologies rather than replacing them.
- It removes the transceiver at the switch face only. The first hop from accelerator to switch, which is where Credo's cables sell, is not addressed by it.
- No independent measurement of deployed volume was found, so the announcement establishes availability and not adoption.
The rivals: MRVL, ALAB, NVDA, AVGO.
Who else does this, rival by rival
| Ticker | Rival | Where it competes | How far along | Priced against it |
|---|---|---|---|---|
| MRVL | Marvell TechnologyPriced against it: Yes | Optical | On sale | Yes |
| ALAB | Astera Labs Taurus Ethernet smart cable modules and Aries paddle cardsPriced against it: Yes | Non-optical connectivity | On sale | Yes |
| NVDA | NVIDIA Quantum-X Photonics and Spectrum-X Photonics switchesPriced against it: Different product | Optical | On sale | Different product |
| AVGO | BroadcomPriced against it: Shown, not counted | - | On sale | Shown, not counted |
How big it could get
How big this could get
Management says the plug-in optical part market grows from 60 million units in 2026 to 175 million by 2030, citing forecasters it does not name. The outside check is Fabrinet, the largest independent builder of these parts: its data-centre sales grew 40.8% to $2.23B in the year to June 2026.
Nobody publishes a price per unit, so the market can be counted in parts but not in dollars.
The evidence under this - 3 claims
- The forecaster is not named and the series is not public: the firms that publish it - LightCounting and Cignal AI - sell it by subscription, and LightCounting's own site refuses an automated fetch, so the figure could not be checked against its source.
- It counts units, not value, and the company's own risk factors say selling prices in its markets fall over time, so unit growth overstates the growth in the money available.
Who has signed
Almost nobody. Orders already agreed were $4.2M at 1 August 2026, against $479.0M of sales in the quarter. Two large optical design wins are promised for fiscal 2028, but management would not name the buyers.
The evidence under this - 2 claims
- The customers are not named - management said it could not say which major players - so neither win can be checked against any counterparty's own disclosure.
- A design win is not an order: the company's contracted orders total $4.2M and it discloses no long-term purchase commitments.
The results
The numbers
Sales grew 165% over the twelve months to Aug 2026, slower than the 206% in the year to May 2026.
Operating margin is 32%, up from −19% in the year to Apr 2024.
Running costs take 35¢ of each sales dollar, down from 81¢ in the year to Apr 2022.
Each dollar put into the business earns 29% a year after tax, up from 5.8% in the year to May 2025.
It turned 28% of sales into free cash, $439M over the last twelve months.
Yearly sales growth has ranged from +4.8% to +206% over the last 5 years.
The share count rose 120% in 4.3 years, and pay in shares costs 15% of sales.
All the numbers
Filed figures from the company's 10-K and 10-Q, read 2026-10-04. Fiscal years end in May.
| Apr 2022 | Apr 2023 | Apr 2024 | May 2025 | May 2026 | 12 mo to Aug 2026 | |
|---|---|---|---|---|---|---|
| Size and growth | ||||||
| Sales | $106M | $184M | $193M | $437M | $1.34B | $1.59B |
| Sales growth | +81% | +73% | +4.8% | +126% | +206% | +165% |
| Profit | ||||||
| Gross margin | 60% | 58% | 62% | 65% | 68% | 67% |
| Operating margin | −21% | −12% | −19% | 8.5% | 33% | 32% |
| Net margin | −21% | −9.0% | −15% | 12% | 35% | 34% |
| What growth costs, per sales dollar | ||||||
| Research and development | 45% | 42% | 50% | 34% | 21% | 21% |
| Selling, general and admin | 33% | 26% | 31% | 23% | 14% | 14% |
| Pay in shares | 8.6% | 13% | 20% | 18% | 14% | 15% |
| Capital spending | 17% | 12% | 8.1% | 8.3% | 4.3% | 3.9% |
| Cash | ||||||
| Cash from operations | −$30.8M | −$24.6M | $32.7M | $65.1M | $464M | $500M |
| Free cash flow | −$48.4M | −$46.3M | $17.1M | $29.0M | $407M | $439M |
| Balance sheet | ||||||
| Cash and short-term investments | $259M | $218M | $410M | $431M | $1.44B | $764M |
| Returns and shares | ||||||
| Return on capital | - | - | - | 5.8% | 32% | 29% |
| Shares (diluted) | 88.4M | 147M | 155M | 181M | 188M | 194M |
What growth costs
Growing has made Credo far more profitable: research and overhead rose much more slowly than sales, so each new sale cost less to win. Stock pay is the exception and keeps rising with sales. The latest quarter ran the other way, with costs growing faster than sales.
The evidence under this - 3 claims
The fall per sales dollar is revenue outrunning spending, not spending cut: R&D dollars rose 5.8 times from fiscal 2022 to fiscal 2026, and Q1 fiscal 2027 R&D was up 118% year on year against revenue up 115%, so the next flat year would reverse the ratio.
The last report (1 Sep)
Quarter ended August 2026, filed figures.
The first quarter of fiscal 2027 was another record: sales of $479.0M, up 114.7% on a year earlier. Yet reported operating profit fell $35.1M in a quarter when sales rose $42.0M.
The company's own reconciliation explains the fall: pay in shares rose $38.3M, and the May purchase of DustPhotonics added $11.2M of amortisation. That explains the fall without excusing it. Pay in shares was $88.0M, 18.4% of sales and 97.5% of the cash the business generated.
What it did to the thesisIt confirmed demand is still racing ahead. It also exposed the cost the Neutral view worries about: pay in shares and a recent purchase swallowed the gain, so reported profit fell. Certainty about the second half of the year, which rests on orders not yet placed, did not return.
The quarter in detail
Sales growth from the previous quarter, filed figures.
Earnings per share $1.20 against $1.17 expected; $0.67 as filed, the gap being what the company adds back.
The evidence under this - 5 claims
- The same arithmetic shows non-GAAP operating profit rose $13.9M, so on the company's own measure the quarter improved. Both readings are correct and the choice of which to use is the valuation frame, not a fact about the quarter.
- Part of the fall is amortisation of intangibles bought three months earlier, on a stated schedule and costing no cash. But it gets larger before it gets smaller - see C49.
What management did with the money
It bought three companies in nine months, DustPhotonics for $1.251B of it, and still holds $764.3M of cash with no debt. It has bought back no shares, and the count rose 9.7% in fifteen months.
Insiders in 2026Q2: 7 purchases and 202 sales, from their Form 4 filings.
The evidence under this - 3 claims
The price, and where we disagree
In short. Fair if this year's plan lands in full, dear if it slips. The price pays for the whole plan delivered, and its second half rests on orders no customer has placed yet.
The shares closed at $150.09 on 15 September, 51.4% below their high, after falling 20.0% the day after results. Every one of the eight companies moving with it did better over the year.
The test comes on 2 December with second-quarter results. The company has guided $525M to $535M, and the full-year plan needs the second half to deliver about $1.46B, 44.8% above the first.
The evidence under this - 4 claims
- 2026-09-28editedre-struck at the 2026-09-25 close of $210.97 on the daily review's maintenance route -- drift +40.3%, past the 12% trigger.
- 2026-10-22editedre-struck at the 2026-10-21 close of $182.28 (from $210.97) on the daily review's maintenance route -- drift -13.6%, past the 12% trigger. Trailing operating profit unchanged since no new quarter has printed; net cash assumed unchanged since the last 10-Q.
- The prior quarter's range of $465M to $475M was beaten at $479.0M, and the range before that was beaten too, so the guide is set to be exceeded rather than met.
- It is a range issued eleven weeks before the quarter ends, and management states it assumes the current tariff regime, which it describes as fluid.
- Management has described this shape in advance twice - in June 2026 as mid-single-digit sequential growth in the first half with an inflection in the second, and again in September - so the back-loading is disclosed rather than discovered.
- The first half is running ahead of what was guided for it: Q1 came in above the high end and Q2 is guided at about 10% sequential against the mid-single digits described in June, which lowers what the second half has to do relative to the original plan.
How we read the multiple
That 54.5 was struck at the mid-September price. After a rally and a pullback on no new results, the company costs about 66 times last year's operating profit.
On management's plan for this year that is about 28 times, but only on adjusted profit that leaves out pay in shares; about 46 times on reported profit. The plan needs the second half of this year to:
- Bring in about $1.46B of sales.
- That is 44.8% more than the first half, which grew 19.5%. Arriving late, the price does not hold.
Against its own history. Dearer than its own normal on the measures that can be compared, though it genuinely earns more now than it used to.
The evidence under this - 4 claims
- 2026-09-28editedre-struck at the 2026-09-25 close of $210.97 (from $150.09) on the daily review's maintenance route -- drift +40.3%, past the 12% trigger. Trailing operating profit unchanged since no new quarter has printed; net cash position assumed unchanged since the last 10-Q.
- 2026-10-22editedre-struck at the 2026-10-21 close of $182.28 (from $210.97) on the daily review's maintenance route -- drift -13.6%, past the 12% trigger. Trailing operating profit unchanged since no new quarter has printed; net cash assumed unchanged since the last 10-Q.
- Both forward figures are struck on management's own guide and management's own adjusted profit measure, which excludes share pay running at 18.4% of revenue. On reported earnings the same guide implies something closer to 38 times.
- The guide has been raised once already and the company has beaten its own revenue range in each of the last two quarters, so the denominator may prove conservative.
- Management has described this shape in advance twice - in June 2026 as mid-single-digit sequential growth in the first half with an inflection in the second, and again in September - so the back-loading is disclosed rather than discovered.
- The first half is running ahead of what was guided for it: Q1 came in above the high end and Q2 is guided at about 10% sequential against the mid-single digits described in June, which lowers what the second half has to do relative to the original plan.
Priced against its rivals
The competitors its own annual report names, each confirmed against the rival's filing. Measured on company value for each $1 of operating profit, cheapest first:
| Ticker | Company | company value for each $1 of operating profit |
|---|---|---|
| CRDO | Credo Technology Group Holding | 54.5× |
| MRVL | Marvell Technology | 122.2× |
| ALAB | Astera Labs | 160.7× |
| AVGO | Broadcom not in the median | 39.0× |
The middle of the other 2 is 141.4. Broadcom Inc. competes with it in one product line inside a much bigger company, so its price says little about this business and it is not counted in the middle. Credo Technology Group Holding Ltd is cheapest of 3.
Rivals' figures come from a data provider, so read the spread of the group rather than any single pair.
The cheapest of the rivals its annual report names while growing faster than any of them. That position is its own, though the group is small.
Private or foreign rivals that cannot be priced: various cable and optical transceiver suppliers.
The bet: what it pays if it works
This bet is on one business only: Optical, the light-based links and chips Credo began selling this year. Everything else, mostly its copper cables, is counted separately at what its listed rival Marvell is valued at, and on management's plan for this year that alone pays for today's price.
So today's price asks nothing of Optical. If Optical hits management's target the company is worth only a little more than today, an expensive bet. If it sells nothing, little is lost, as long as the copper plan holds.
| Optical revenue in two to three years | Points of Fabrinet's data-centre revenue needed | At the stated multiple |
|---|---|---|
| $600.00M (the big win) | about 27 at 100% kept | 1.5× |
| $1.02B | about 46 at 100% kept | 1.8× |
| $1.25B (market forecast) | about 56 at 100% kept | 1.9× |
LightCounting's forecasts put Ethernet optical transceivers near $50B in 2028; at 2.5%, under double the share the fiscal 2027 guide implies, Credo's optical line earns about $1.25B in fiscal 2029. On that forecast, $1.25B pays 1.9× from today's value: an Expensive bet on the market case. Optical would be worth about $21.62B, and the price pays nothing for it.
To pay 2× (a Fair bet) it would need about $1.42B of Optical revenue (about 64 points of Fabrinet's data-centre revenue); to pay 5× (a Cheap bet), about $6.85B of Optical revenue (about 307 points of Fabrinet's data-centre revenue), at the same multiple.
Each row is Optical alone, plus the rest of the company (Non-optical connectivity) counted at about $38.15B.
The steps it has to take
- Done DustPhotonics bought, bringing silicon photonics in house
- Done Two silicon photonics design wins secured for fiscal 2028 ramps
- In progress The more-than-$600M optical guide held or raised at the Q2 fiscal 2027 print due 2026-12-02
- Not started Fiscal 2027 optical revenue of at least $600M reported due 2027-06
- Not started Both silicon photonics design wins ramping in fiscal 2028 due 2028-05
The detail behind the label
The evidence under this - 4 claims
- Nothing is contracted behind it: $4.2M of orders are recorded as contracted against a year guided at about $2.47B, and the 10-K states there are no long-term purchase commitments.
- The same management guided fiscal 2026 at roughly half what it delivered, so the record argues the guide is conservative rather than stretched - which cuts against treating the raise as a warning sign.
- Four analysts on the Q1 call pressed on the second-half ramp and on the ability to scale supply for it; the answers were about engagement and supply-chain preparation, and named no order.
Where we disagree with the market
Nineteen analysts, 94.7% positive and unchanged since June, see the plan as safe.
The business did not weaken; certainty did. The market repriced it, and certainty has not come back.
Analysts have kept raising their targets while the shares halved. The filings show what they look past: $4.2M of contracted orders, and operating profit falling as sales rose.
The case against us is strong. In March the market made the same objection, a margin forecast to fall, and the margin held; the shares went from $87.81 to $302.52 by June.
What analysts say, and the evidence
The evidence under this - 5 claims
What could go wrong, and how we would know
The risks
- Two customers carry it. Two customers were 71% of the latest quarter's sales, and neither is committed to buy anything beyond current orders.
- A second half nothing backs. The year needs about $1.46B in six months, and the optical target it depends on was not raised with the headline forecast.
- Rivals with the same parts. Marvell and Astera Labs sell into the same sockets, and prices in this market have always fallen.
- Pay in shares. Guided to rise to $97.0M next quarter, it already takes nearly all the cash the business generates.
The evidence under this - 9 claims
- Management has described this shape in advance twice - in June 2026 as mid-single-digit sequential growth in the first half with an inflection in the second, and again in September - so the back-loading is disclosed rather than discovered.
- The first half is running ahead of what was guided for it: Q1 came in above the high end and Q2 is guided at about 10% sequential against the mid-single digits described in June, which lowers what the second half has to do relative to the original plan.
Guided share pay has been close to delivered in recent quarters, so the risk here is that it is understated rather than overstated: Q1 non-GAAP operating expenses came in above the high end of the guided range.
What we still do not know
- Who the two big customers are. The filing names whoever places the order, which may be a manufacturer buying for someone else.
- How much of the latest quarter's sales were optical. The company reports one business and gives no product split.
- What one optical part sells for. No free source publishes it.
The evidence under this - 4 claims
The filed concentration note reports two customers above 10%, not four, because it counts the contracting party rather than the end customer. The wider count is management's cut and is not in any filing.
- Nothing is contracted behind it: $4.2M of orders are recorded as contracted against a year guided at about $2.47B, and the 10-K states there are no long-term purchase commitments.
- The same management guided fiscal 2026 at roughly half what it delivered, so the record argues the guide is conservative rather than stretched - which cuts against treating the raise as a warning sign.
- Four analysts on the Q1 call pressed on the second-half ramp and on the ability to scale supply for it; the answers were about engagement and supply-chain preparation, and named no order.
- The forecaster is not named and the series is not public: the firms that publish it - LightCounting and Cignal AI - sell it by subscription, and LightCounting's own site refuses an automated fetch, so the figure could not be checked against its source.
- It counts units, not value, and the company's own risk factors say selling prices in its markets fall over time, so unit growth overstates the growth in the money available.
What would change our view
- Neutral→AttractiveThe guide lands with margin and optical intact
It needs all of:
- ○Sales for fiscal 2027 grow at least the promised 85%, with optical sales reaching the promised $600M.checked at the next results, due by 2 Dec
- ○Reported operating profit rises with sales instead of falling again, and the adjusted gross margin stays at 62% or better.checked at the next results, due by 2 Dec
- Neutral→UnattractiveA customer or the growth rate falls away
It needs any of:
- ○Either of the two biggest customers falls below a tenth of sales within two quarters.checked at the next results, due by 2 Dec
- ○Fiscal 2027 growth lands below 60%, or optical sales finish below $450M.checked at the next results, due by 2 Dec
What is coming
- October 2026 Open Compute Project summit: light-emitting-diode cable demonstration: Whether the committed demonstration happens on time, the first public test of the Hyperlume technology bought in September 2025.
Every dated event ahead
- October 2026 Open Compute Project summit: light-emitting-diode cable demonstration: Whether the committed demonstration happens on time, the first public test of the Hyperlume technology bought in September 2025.
- 12 October 2026 Annual general meeting: say-on-pay vote and three directors: How shareholders judge a pay plan that paid its maximum against goals set far below the outcome; a weak vote would press the board on goal-setting.
- 2 December 2026 Q2 fiscal 2027 results: Revenue against the $525M to $535M guide, the above-85% year held, optical above $600M repeated or raised, and whether reported operating profit rises again.
- June 2027 Q4 fiscal 2027 results and the fiscal 2027 10-K: Whether the guided second half arrived, optical delivered above $600M, and trailing revenue crossed the chief executive's first $2.5B pay hurdle.
Since we wrote this
- 22 OctPrice movePrice re-struck, -13.6%; view heldNo print or development since the last re-strike. The plan is priced at about 66x trailing operating profit rather than 77x on the same quarter.
- 5 OctThesis reviewThe bet now names what it ridesThe owner asked that every bet name the macro trend behind it, after BWXT's bet was missing its chief executive's case for nuclear over gas.
- 4 OctThesis reviewOwner's checklist: revenue streams, cost of growth, management, catalystsThe owner's checklist found the method silent on who runs the company, what growth costs, where the money comes from and what is dated ahead; this pass fills those four blocks from the proxy and filings.
- 28 SepPrice movePrice re-struck, +40.3%; view heldNo print, filing or confirmed development since the thesis. The plan is now priced at about 77x trailing operating profit rather than 54.5x on the same quarter.
- 15 SepThesis writtenNeutral, medium convictionTwo weeks after results sent the shares down 20.0% in a day, although sales had more than doubled and the year's forecast went up.
- 15 SepThesis reviewArgument rebuilt around the rating, not the business. View heldFalling after a beat turned out to be common among its rivals. The better points: the whole year's fall was the rating, and March's margin worry proved wrong.
- 15 SepThesis reviewThe buyer's cost of a failed link added. View heldManagement says link failures cost customers 10% to 20% of their processors' time. That explains the margin without making it more durable.
Also checked 4 times with nothing new to act on.
Sources 30
Sources - 30
Every claim above cites one of these, or states the reasoning that derived it. The quote is frozen at the date it was read: a link can rot, and an agent fetch returns a reading of a page rather than the page. Backs counts the whole claim ledger, so it can name a claim this page does not surface.
- S01primarycompany filing2026-06-15the product line and Item 1 business description, the named competitor set, customer concentration by year, the tax-rate reconciliation and the Pillar Two statement, headcount, the TSMC sole-source disclosure, the average-selling-price risk factor, and the audited fiscal 2026 statementsbacksC01C02C08C09C10C15C18C24C26C27C35C36C37C38C39C45C48C64C65C70
- S02primarycompany filing2026-09-02the DustPhotonics purchase price allocation, remaining performance obligations of $4.2M, customer concentration at 43% and 28%, revenue by destination, the cash flow statement and the share-count roll-forwardbacksC03C06C08C09C10C15C23C24C25C28C41C48C49C64C65C70
- S03primarycompany filing2026-09-01the GAAP-to-non-GAAP reconciliation line by line, the quarterly income statement and balance sheet, and the Q2 fiscal 2027 outlook including the $97.0M share-pay linebacksC03C04C06C07C25C32C33C40C47C48
- S04primarytranscript2026-09-01Q1 fiscal 2027 earnings call transcriptthe raised full-year guide of above 85%, the $600M optical target, the end-customer split of 33/28/13/10, the 60-to-175 million unit market forecast, Positron as first OmniConnect customer, and ten analysts' questions on the second-half ramp and on scaling supply
we continue to expect an inflection in the second half driven by more than $600 million in optical revenue, with zero-flap optics, silicon-photonics PICs, and optical DSPs each contributing more than $100 million. Resulting in more than 85% year-over-year total revenue growth for the full year
backsC07C11C18C36C51C52C53C54C62 - S05primarycompany filing2026-06-01the fiscal 2026 full-year figures and the Q1 fiscal 2027 outlook the quarter was measured againstread, but no claim rests on it
- S06primarytranscript2026-06-01Q4 fiscal 2026 earnings call transcriptthe original fiscal 2027 guide of above 80% growth and about 50% expense growth, so the September raise could be measured rather than asserted
we expect mid single digit sequential growth in the first half with an inflection beginning in the second half
backsC11C54 - S07primaryvendor data2026-09-15Research pack frozen with this thesis, 15 September 2026price and returns, the cohort multiples, the analyst-rating series, theme co-movement and member returns, FINRA short interest and the insider base ratebacksC13C14C15C16C17C45C46C47C48C55
- S08primarycompany filing2026-02-20a rival's own account of the same market: its product families, its naming of Credo as a principal competitor, its statement that qualification is a principal competitive factor, and its cable-module productsbacksC15C19C21C35
- S09primarycompany filing2026-03-11a rival's own list of interconnect products covering Credo's whole silicon range, its naming of Credo as a direct competitor, its data-centre revenue share, and its statement that customers develop semiconductors internallybacksC15C20C35C42
- S10primarycompany filing2025-12-18the segment split that decides Broadcom's perimeter in the cohort: $36.9B of semiconductors against $27.0B of infrastructure software, with no connectivity line disclosedread, but no claim rests on it
- S11primarycompany filing2026-08-18a filed, outside measurement of how fast the optical module supply chain is actually growing - $2,225.1M of data-centre revenue, up 40.8%backsC31
- S12primarycompany release2025-03-18the attacker's own timetable for moving the optical engine onto the switch, and its own statement that this happens alongside pluggable transceivers rather than replacing them
NVIDIA Quantum-X Photonics InfiniBand switches are expected to be available later this year, with NVIDIA Spectrum-X Photonics Ethernet switches coming in 2026
backsC22 - S13primarycompany filing2026-09-15Form 4 filings read directly from EDGAR by scripts/insider_fetch.pythe Rule 10b5-1 flag no vendor feed carries: 716 open-market sales, no purchases, 83.7% of shares sold under a plan adopted earlierbacksC29
- S14primarycompany release2025-03-18the attacker's own statement that co-packaged optics grows beside plug-in transceivers rather than replacing them - the single most important qualifier on the threat
NVIDIA photonics will drive massive growth for a new wave of state-of-the-art AI factories, alongside pluggable optical transceiver technologies
read, but no claim rests on it - S15primarytranscript2026-09-01Q1 fiscal 2027 earnings call - the market-size passagethe only unit base available for the prize, and the fact that management attributes it to unnamed market forecasters
Just that piece alone is expected to grow from 60 million units in 2026 to 175 million units by 2030
backsC30 - S16primarytranscript2026-09-01Q1 fiscal 2027 earnings call - the end-customer splitmanagement's end-customer cut of concentration, which is wider than the filed contracting-party table and appears in no filing
our largest customer was a third of our revenue at 33%; second largest, 28%; and then followed by 13% and 10%
backsC44 - S17primarytranscript2026-09-01Q1 fiscal 2027 earnings call - the silicon photonics design winswhat stands behind the optical ramp beyond the guide: two unnamed wins, and management's statement that neither includes Credo's own signal chip
we have got design wins with 2 major players for next generation ramps that will occur in fiscal 28
backsC43 - S18primarycompany filing2026-03-02the print and the guidance behind the March 2026 repricing: revenue of $407.0M at a 68.6% adjusted gross margin, and the following quarter guided to 64.0%-66.0%backsC46
- S19primaryvendor data2026-09-15Astera Labs and Marvell earnings dates and closing prices, FMPthe pressure test's peer channel: how the two comparable rivals' own shares reacted to their own beats, which is the base rate the 20% fall has to be read againstbacksC50
- S20primaryindustry data2026-03-31an independent analyst's dollar size of the 2025 Ethernet optical transceiver market, the base the market case grows; read through ir_fetch
We now estimate sales of optical transceivers and related products at $23.8 billion in 2025, up 55% from 2024. First quarter of the year was seasonally slow, but the last three quarters were incredibly strong, as illustrated in the Figure below. Sales of Ethernet optical transceivers should be close to $18 billion in 2025 - up 70% from last year.
backsC56 - S21primaryindustry data2026-09-30LightCounting newsletter - We Must Be at the Peak of This Super-cycle (Ethernet Optics report, September 2026) LightCountingthe latest growth forecast for 2026, the shape of the 2027-2031 path in both scenarios, and the warning that flat or negative years interrupt every growth period; read through ir_fetch
We now expect 120% growth in the Ethernet optical transceiver market for 2026, as shown in the figure below. Our detailed forecast for 2027-2031 is based on the “Soft Landing” scenario. ... Even in this scenario, we assume that the market growth rate will decline from 120% in 2026 to 76% in 2027 and 20% by 2031. We must be at the peak of this super-cycle in terms of growth rate.
backsC57 - S22primaryindustry data2026-07-31LightCounting newsletter - The industry supply chain remains red hot (Cloud Data Center Optics report, July 2026) LightCountingthe analyst's dollar end-point for 2031, its capex assumption of 70% growth in 2026 slowing to 30% in 2027, the laser shortage stretching to mid-2027, and the observation that transceiver sales turn down six months ahead of capex; read through ir_fetch
We now expect that sales of Ethernet optical transceivers will ramp by 73% in 2026, propelling the market to $80 billion by 2031, even with fairly conservative assumptions for growth in 2027 and beyond.
backsC58 - S23primaryindustry data2026-04-21a second analyst's unit count for 2025 high-speed modules, its $19B datacom component revenue for 2025 and Innolight's lead at more than $5B; read through ir_fetch
Over 42 million high speed (400G+) datacom modules shipped in 2025, up more than 80% over 2024, split between 400GbE and 800GbE.
backsC59 - S24primaryindustry data2026-06-18Cignal AI press release - Datacom Optical Component Revenue Doubles to $7.7 Billion in 1Q26 Cignal AIthe first-quarter 2026 unit run rate, 1.6TbE above 10 million units in 2026 and nearly all of it silicon photonics DR, and CPO growing to tens of millions of ports by 2030; read through ir_fetch
High-speed (400G+) datacom module shipments set a record; up 145% year-over-year. Innolight again topped the leaderboard in both Datacom revenue and high-speed module shipments, by a wide margin. Almost 20 million modules shipped in the quarter as AI build-outs accelerated.
backsC59 - S25primaryindustry data2026-05-31the two largest module makers' 2025 revenue, the scale a share of this market is won against; read through ir_fetch
The company reported $5.3 billion in revenue for 2025 - up 61% and just above of our estimates of 58% growth in the total transceiver market last year. ... Eoptolink reached record revenues in 2024 at $1.2 billion and again in 2025 at $3.5 billion - up 175% and 189%, respectively.
backsC60 - S26primarycompany filing2026-08-14a listed transceiver rival's filed growth: Datacenter & Communications revenue up $1,519 million, or 40%, in fiscal 2026, and its 800G/1.6T transceiver, CPO and silicon photonics development programmebacksC60
- S27primarytranscript2026-08-27Marvell Technology Q2 fiscal 2027 earnings call transcript MRVLthe incumbent optical-DSP seller's own description of its position and its 1.6T ramp; read through transcript_fetch (FMP)
Marvell's market-leading franchises in optical DSPs, switching and broadband analog components continue to see significant demand. On the optical DSP side, 800 gig demand remains strong, while our 1.6T business is ramping rapidly, a trend we expect to accelerate further in fiscal 2028.
backsC60 - S28primaryindustry data2026-02-26the analyst's size for 1.6T chipsets in 2026 and its expectation that linear drive erodes the DSP's share of the module from 2027; read through ir_fetch
Sales of 1.6T chipsets will exceed $2 billion in 2026 and grow rapidly through 2029. ... We expect that growth in sales of PAM4 chipsets will moderate in 2027-2031, as volume deployments of linear-drive solutions (LPO and CPO) will negatively impact DSP sales.
backsC61 - S29primarycompany filing2026-08-25beneficial ownership at 31 July 2026, director and officer biographies, the fiscal 2026 bonus plan and payouts, the fiscal 2025 PSUs, the June 2025 special PSUs and the fiscal 2027 CEO PSUs, and the meeting datebacksC66C67C68C69
- S30primarycompany filing2025-07-02the last filed split of revenue by type: product sales $412.2M, IP licence $12.5M and product engineering services $12.1M in fiscal 2025backsC64