Earnings Digest: Credo (CRDO) | Q1 FY2027 — Optics Becomes the Second Engine, but the 85% Full-Year Pledge Loads $1.46B into the Second Half

2 days ago

11 min read

Credo posted $479M in revenue this quarter, up 114.7% YoY and its seventh straight quarter of triple-digit growth. But the real story of this call wasn't that number. It was three things:

First, optics (DSP, silicon photonics PICs, ZeroFlap Optics) officially graduated from side dish to second engine, with management reiterating that full-year optical revenue will exceed $600M;

Second, the company is reaching into NPO and memory fan-out (OmniConnect), both pointing to FY2028;

Third, the full-year pledge of “revenue growth above 85%” effectively loads about $1.46B of revenue into the second half — 1.45x the first half. This is not guidance that can be delivered slowly.

This article is for technology and industry trend analysis only and does not constitute investment advice.

1. Three Core Takeaways

Takeaway 1: AEC isn't being replaced by optics — it's being redefined. Bill Brennan deliberately switched frames this time. Stop asking whether copper will lose to optics; instead, treat AEC as the 1-to-7-meter segment of the pluggable transceiver market. Unit volume in that market is expected to grow from 60 million in 2026 to 175 million in 2030. Through this lens, AEC isn't an installed base being eroded — it's incremental volume growing alongside the entire pluggable market.

Takeaway 2: This quarter marks Credo's shift from selling chips to selling the entire optical link. Optical DSP hit a record, the silicon photonics PIC business from the DustPhotonics acquisition recognized revenue for the first time, and ZeroFlap Optics entered volume shipment — all three lines now have numbers. More importantly, Credo has started co-optimizing DSP and PIC, then layering firmware, telemetry, and PILOT software on top. We broke down this path in After Copper Runs Out for AI: Seven Paths for Scale-Up Optical Interconnect: vertically integrated players can capture both a COGS advantage and an ASP premium.

Takeaway 3: Two FY2028 options are now officially on the table. One is NPO/Open CPX (near-packaged optics for scale-up), with confirmed design wins set to ramp starting in FY2028. The other is OmniConnect — using SerDes plus the Weaver gearbox to solve memory fan-out for inference workloads, which management describes as “thousands of dollars of Credo content per GPU.” Neither is in this year's numbers, but both will determine how big a company Credo is in FY2029.

2. Revenue and Financials: Seven Quarters of Triple-Digit Growth, with Cash Flow and Inventory as the Variables

Q1 FY2027 (quarter ended August 1, 2026) key figures:

・Revenue of $479.0M, up 114.7% YoY and 9.6% QoQ, above the high end of guidance. This is the seventh consecutive quarter of triple-digit YoY growth.

・Gross margin: GAAP 64.5% (down 2.9 pts YoY); non-GAAP 68.0% (up 0.4 pts YoY), at the midpoint of guidance. The 3.5-point gap between GAAP and non-GAAP comes mainly from amortization of intangibles from the DustPhotonics acquisition — not a collapse in core margins.

・Operating income: GAAP $120.7M, a 25.2% operating margin (up 98.7% YoY); non-GAAP $230.6M, a 48.2% operating margin.

・Net income: GAAP $129.4M (up 104.1% YoY); non-GAAP $236.3M (up 140.5% YoY), a record non-GAAP net margin of 49.3%.

・Diluted EPS: GAAP $0.67 (up 97.1% YoY); non-GAAP $1.20.

・Operating expenses: non-GAAP $95.2M, up 16% QoQ, above the high end of guidance — management said explicitly this is stepped-up R&D, not spending out of control.

Three easily overlooked balance-sheet signals:

1. Operating cash flow of $90.2M, down $92M QoQ, mainly due to working-capital changes. Capex was $7.3M and free cash flow $82.9M.

2. Inventory of $313.1M, up $62.2M QoQ. Bill Brennan admitted in the Q&A that this is deliberate: “We're leaning forward from a supply-chain standpoint,” to support second-half volume.

3. Cash and equivalents of $764.3M, down $679M QoQ, almost entirely the purchase consideration for DustPhotonics.

3. AEC: Not Replaced by Optics, but Placed in a Bigger Frame

AEC is still Credo's largest business, and it's still growing. Three concrete developments this quarter:

・Deep relationships established with five hyperscalers, with continued expansion of engagement with NeoCloud customers.

・Penetration at existing customers rises with deployment scale — the more racks a customer deploys, the more AEC it uses.

・200G per lane and 1.6T ports are the next growth vectors, starting to contribute in the second half and ramping in volume in FY2028.

More memorable than the numbers is the frame Bill Brennan swapped out. For two years the market has viewed AEC through a “copper vs. optics” lens; this time he defined AEC outright as “the 1-to-7-meter segment of the pluggable transceiver market” — Credo's AEC reaches up to 6.5 meters.

In the 800G generation, what we saw was copper replacing optics — specifically laser-based optical modules — because that first link (GPU to the first-tier switch) demands reliability first, and power second.

This framing does two things. First, it moves AEC's ceiling from the “copper cable market” to the “pluggable market,” which grows from 60 million units in 2026 to 175 million in 2030. Second, it reserves a seat for ALC (Active LED Cable, up to 30 meters), discussed later — the third chair in the same frame.

In other words, Credo doesn't intend to pick a side in the copper-vs-optics debate. It wants to sell chairs on both sides.

4. Three Optical Lines: Record DSP, First PIC Revenue, ZeroFlap in Volume Production

Optical DSP: Revenue hit a record this quarter, spanning both 50G and 100G per lane. Management stressed that 800G ports still have “a long tail,” while customer engagement on 1.6T (200G per lane) spans both fully retimed and LRO solutions, with first 1.6T DSP revenue coming later this fiscal year.

Silicon photonics PIC: First revenue recognized after the DustPhotonics acquisition. Initial design wins are in 800G and 1.6T optical transceivers, expected to ramp through the year. The Q&A revealed an important detail — the design wins at these two “heavyweight players” do not include Credo's DSP, meaning a DSP-plus-PIC bundle remains upside, not a number already in the bag.

ZeroFlap Optics: Already shipping in volume, with more customers ramping in FY2027 across 800G and 1.6T, and across hyperscalers and NeoClouds.

Adding up the three lines, management reiterated full-year optical revenue above $600M, with ZeroFlap Optics, silicon photonics PICs, and optical DSP each exceeding $100M. The $600M target was first announced last quarter, and we broke down its components and assumptions in Earnings Digest: Credo (CRDO) | Q4 FY26 — The Copper King Turns Reliability into a $600M Optics Business. What this quarter adds: all three lines now have real revenue, not just roadmaps.

5. PILOT: What Credo Is Really Selling Isn't Modules — It's the Yellow Light

The most technically dense part of the call was Bill Brennan's answer to a question on telemetry data. It's worth recording in full, because it explains why ZeroFlap Optics commands a premium over ordinary optical modules.

How it works: Credo first had to build a custom DSP to light up telemetry on every single link between the XPU and the switch — six per connection (three in each direction, six total). What it senses includes real-time eye height, signal-to-noise ratio, and post-FEC histograms, measured all the way down to the SerDes level.

The value:

Today the industry state is: if it's up, it's green; if it's down, it's red. What we've added is the yellow ‘check engine’ light.

With a yellow light, “mitigation” becomes possible — you can gracefully drain the GPUs attached to a soon-to-fail transceiver out of the cluster one by one, or make centralized network-level decisions. Credo has customers doing both.

Two concrete cases already delivered: first, detecting very slight electrostatic discharge (ESD) damage — not the kind that kills a transceiver immediately, but the kind that becomes a latent defect later — so customers know at rack bring-up which unit was mishandled; second, detecting dust on fiber, since back-reflection from a single dust particle creates multipath interference.

Translated into financial terms: cluster bring-up time compressed from 6–8 weeks to 5–6 days, and 99%+ availability after deployment. Bill Brennan converted it straight into money — at the scale of today's publicly announced AI compute contracts, a one-month gap represents hundreds of millions to a billion dollars of revenue opportunity cost.

This is the hardest moat argument in the entire transcript: Credo isn't selling optical module specs, it's selling cluster availability.

6. Retimers and Two FY2028 Options

Retimers also hit a record this quarter. Growth was driven by Screaming Eagle (100G per lane) in scale-up deployments, with Blue Heron (200G per lane) starting to contribute; Toucan is ramping with PCIe Gen 6 adoption. Credo calls itself “protocol-agnostic” — the same SerDes IP supports Ethernet, UALink, and PCIe, so it can participate wherever scale-up protocols finally converge. For why the PCIe line went from optional to essential, see 2026 OCP APAC Summit | PCIe 7 Turns the Retimer from Optional to Essential.

Option 1: NPO and Open CPX. Credo joined OCP's CPX MSA this quarter, signaling clearly that it will compete in near-packaged optics for scale-up. Management's wording is worth noting — “confirmed design wins, expected to begin ramping in FY2028”. That's “confirmed,” not “expected,” a notch more certain than a typical pipeline. The approach starts with silicon photonics PICs and moves toward full optical engines over time. On the route split among CPX, XPO, NPO, and CPO, Bill Brennan is “agnostic”: build whatever customers want, because all these paths solve the same problem — scale-up needs 10x the density of scale-out. For the full picture of this debate, 2026 OCP APAC Summit | CPO / NPO / XPO Panel: It's Not a Route War, It's the 409.6T Bottleneck 18 Months Out has a complete write-up.

Option 2: OmniConnect and memory fan-out. This is the most underrated part of the call. Credo's thesis: the real bottleneck for inference workloads is memory fan-out — not enough XPU beachfront, not enough reach between GPU and memory. The solution has two parts:

・Licensing a highly optimized high-speed SerDes: tiny area, ultra-low power, with reach up to 10 inches.

・The Weaver gearbox: one side connects to the SerDes embedded on the XPU, the other converts to an LPDDR interface. The first does LPDDR5, the second LPDDR6 — the point being that when memory generations change, customers don't need to re-tape out; they just swap the gearbox.

The first customer is Positron, whose first product reaches 2 TB of memory capacity, pitched as fitting an entire frontier model in memory. Management also offered a claim that needs ongoing verification: a bandwidth roadmap “quite comparable” to HBM5, while avoiding the reliability issues of co-packaging XPU and HBM. Revenue likewise starts in FY2028.

ALC (Active LED Cable) is a third, smaller card: using microLED emitters to extend AEC's reliability and power advantages to 30 meters, demoed at OFC in October, with first revenue in FY2028.

7. Management Outlook: The 85% Math Problem

Official guidance, numbers only:

・Q2 FY2027 revenue of $525M–$535M (up 9.6% to 11.7% QoQ)

・Q2 non-GAAP gross margin of 67.0%–69.0%

・Q2 non-GAAP operating expenses of $100M–$105M

・Q2 diluted share count of approximately 200M

・FY2027 full-year revenue growth above 85% YoY; non-GAAP gross margin roughly flat with FY2026; non-GAAP operating expenses up about 55% YoY; non-GAAP net margin around 50%

・FY2027 optical revenue above $600M, with each of the three product lines above $100M

Now turn it into a math problem. FY2026 revenue was $1.3351B, so 85%+ growth means FY2027 needs to reach $2.47B or more. First half = Q1's $479M + Q2 guidance midpoint of about $530M = $1.009B. In other words:

The second half must deliver at least $1.46B — 1.45x the first half. That's an average of about $730M per quarter, 38% above the Q2 guidance midpoint.

This is the real magnitude of what management calls the “second-half inflection.” It isn't a gradual ramp; it's a step function. It rests on three pillars: AEC moving to 1.6T, ZeroFlap Optics ramping at multiple customers simultaneously, and first recognition of 1.6T DSP revenue. If any one pillar slips by a quarter, the full-year 85% turns into some very ugly arithmetic.

Note the difference in wording: the $600M optics target is “remain on track,” the NPO design wins are “confirmed,” and OmniConnect and ALC are “expect.” Three levels of certainty that shouldn't be discounted at the same rate.

8. Customer and Supply Chain Clues

Customer mix: Four customers accounted for 10% or more of revenue this quarter — 33%, 28%, 13%, and 10%. The top three are the same as last quarter, just reordered; the fourth changed, though that customer has been a 10% customer before. Management maintained that there will be “three to four 10%+ customers in the coming quarters and for the full year,” and stressed that the customer base is diversifying across hyperscalers, NeoClouds, and others. The CFO added that these hyperscale customers are buying more than AEC — the product line is broadening.

Supply chain: Bill Brennan said Credo started securing supply 18 to 24 months ago and “feels good” about supply capacity for the second half and the next two fiscal years. Read this alongside the $62.2M QoQ inventory increase and the $92M QoQ drop in operating cash flow — two sides of the same coin.

Channel signal: In the optical components business, Credo's direct customers are module makers, but hyperscalers directly specify which components go into the modules. In other words, hyperscalers are both the de facto decision-makers on component sales and direct customers for ZeroFlap modules. Management expects components and modules to be complementary over the long run — the broad standards-based market goes through components, while the extremely reliability-sensitive segment goes through ZeroFlap modules.

9. Risks and Counterarguments

1. Second-half step-function risk. As noted above, 85% full-year growth requires the second half to deliver 1.45x the first half. Guidance shaped like this leaves very little margin for error.

2. Extremely high customer concentration. The top two customers account for 61% of revenue combined. Management's “diversification” is a direction for now, not a state.

3. GAAP gross margin hasn't bottomed yet. Q2 GAAP gross margin guidance is 62.9%–64.9%, with the midpoint below this quarter's 64.5%. Non-GAAP is holding at 68%, but the GAAP/non-GAAP gap widened to 3.5 points after the acquisition and won't converge in the near term.

4. Operating expenses keep overshooting. Non-GAAP opex came in above the high end of guidance this quarter, up 16% QoQ, and is set to grow 55% for the full year. Revenue is still growing faster, so leverage remains, but this ratio needs constant watching.

5. The two big PIC design wins don't include Credo's DSP. Management itself called this “upside” — in other words, the vertical-integration premium hasn't yet materialized in these two programs.

6. Competition is getting denser in the same box. In the same cycle, Marvell also cited scale-up optics and NPO as the single largest source of its FY2028 upside. Credo's differentiation pitch is “owning the entire technology stack” and “speed of innovation,” but neither is a structural moat — both must be re-proven every generation.

10. Conclusion

Credo delivered an “engine transition in progress” report card this quarter. The AEC engine is still accelerating, but the narrative's weight has clearly shifted toward optics — and not to a single product, but to the entire vertical stack of DSP, PIC, modules, and software. What really holds this together isn't any single spec; it's PILOT's yellow light. Once availability becomes quantifiable and priceable, Credo shifts from “selling parts” to “selling clusters that stay up.”

But the report card also pins three checks to the same timeline: $1.46B in the second half, NPO design wins in FY2028, and OmniConnect in FY2028. The first is execution risk; the latter two are options.

Three metrics to track next quarter:

1. Whether first 1.6T DSP revenue shows up in Q2 or Q3 — the first checkpoint on whether the second-half step-up holds.

2. ZeroFlap Optics customer count and the 800G/1.6T mix — of the three optical lines each crossing $100M, ZeroFlap is the hardest.

3. Whether non-GAAP gross margin holds in the 68% range — as module revenue mix rises, this number is the referee on whether vertical integration truly delivers a COGS advantage.