In February, Credo Technology did something companies almost never do voluntarily: they pre-announced their quarter, six weeks before the scheduled report.

The number wasn't a small beat. Their previous guidance for fiscal Q3 had been $335 to $345 million in revenue. The pre-announce: $404 to $408 million. That's a 20% upside surprise on the already raised guide, mid-quarter — the kind of magnitude that says the model and the company have decoupled.

Q3 actual came in at $407 million. Up 51.9% sequentially. Up 201.5% year over year. Non-GAAP EPS of $1.07 against expectations that hadn't even been refreshed for the new revenue trajectory.

Then they guided Q4 to $425-435 million.

This is a piece about what the pre-announce told me, what the Street still hasn't fully modeled, and why the revisions on Credo Technology (NASDAQ: CRDO) are nowhere near finished.

My framework is simple: I don't chase price targets and I don't trade on conviction headlines. I look for companies where the rate of change in earnings and revenue estimates is accelerating. The signal isn't where estimates are. It's the direction and velocity of where they're going.

Three things tell you a revision cycle is still mid-flight rather than maturing:

The magnitude of upward surprises is getting larger, not smaller. Multiple analysts are revising upward in the same window. The company is disclosing new revenue streams that aren't yet in the model.

By all three measures, Credo is still early.

Credo is in the business of moving data inside AI data centers. Specifically, they make Active Electrical Cables — AECs — that extend the reach of copper interconnect inside server racks. Copper is cheap. Optics are expensive. Every foot a hyperscaler can keep on copper instead of converting to optics saves real money at scale.

Their core insight, from the company's founding in 2008: as link speeds climbed from 100G to 400G to 800G, traditional passive copper cables would stop working at meaningful distances. You'd need active conditioning at both ends — chips that retime the signal so it can travel further. That's the AEC. Credo doesn't sell the cable, exactly. They sell the SerDes chip inside the cable that makes copper viable at speeds where it shouldn't be.

For most of its history this was a real but modest business. Then AI clusters happened.

A modern AI training cluster has hundreds of thousands of GPU-to-GPU connections. The shift from 400G to 800G is mostly done; 1.6T is ramping now. Each speed migration roughly doubles the dollar content per port. Each AI cluster has dramatically more ports than any traditional cloud workload. Credo's TAM tripled in three years almost mechanically — not because their share grew, but because the world they sell into got much bigger.

Microsoft has been the dominant AEC customer. That's the central concentration risk, and I'll come back to it.

Look at the trajectory:

FY2024 revenue: $193 million. FY2025 revenue: $437 million, up 126%. FY2026 guidance through Q4 is tracking to roughly triple that again. Six times revenue in two years. There are very few public companies — at any market cap — that have done this organically, without acquisitions or one-time customer-concentration spikes.

The Street has been chasing the number all year:

The original FY26 consensus formed in spring 2025 was meaningfully below where the company actually is now. Post-Q1, analysts revised up after a beat-and-raise. Pre-Q3 guide came in at $335-345M. The actual Q3 pre-announce landed at $404-408M — 20% above their own raised guide. Q4 guidance of $425-435M is above analyst expectations even after the Q3 reset.

This is the textbook shape of a stock where the model is structurally lagging the company. Each successive guide is being revised up. Each successive print is exceeding the revised guide. Revision breadth — the percentage of analysts revising upward in the trailing 30 days — is at or near 100%.

Estimate dispersion tells you how confident the Street is in the trajectory. Tighter dispersion means analysts agree. Wider dispersion means they don't. Credo's FY27 dispersion is unusually wide, which is exactly what you want to see in a revision-mid-cycle stock — bulls and bears are still converging on the new normal.

Three product expansions and one acquisition have been announced in the past six months, and each represents a separate vector that is not fully reflected in current consensus models:

The DustPhotonics acquisition. In April, Credo announced a roughly $750 million agreement to acquire DustPhotonics, a silicon photonics specialist. This is the single most important development in the company's recent history and is not yet in any consensus model. The acquisition transforms Credo from a copper-and-DSP company into a vertically integrated optical interconnect platform. They now own the silicon photonics IP that goes inside their own optical transceivers, rather than buying it on the open market. The strategic logic is straightforward: as data center connectivity migrates from pure copper to optics-augmented copper to integrated optics, owning the photonics stack is what separates a parts supplier from a platform vendor. Integration costs hit Q4 margins. The revenue synergies don't show up until FY27.

ZeroFlap optical transceivers. Credo's 400G and 800G ZeroFlap transceivers debuted at OFC 2026 in Los Angeles in March, with samples shipping now and material revenue expected late FY26 into FY27. ZeroFlap targets a real and persistent problem in AI clusters — link flaps, where optical connections repeatedly disconnect and reconnect, undermining training stability. This is a product that solves a problem hyperscalers care about, not a spec sheet looking for a customer.

Robin DSP family. Credo's sixth-generation 800G and 400G optical DSP architecture also debuted at OFC, putting Credo in direct competition with Marvell and Broadcom on optical DSP share. This is the largest TAM expansion of the three — optical DSPs are a multi-billion-dollar market — and it's the one with the most uncertain ramp because Credo is the new entrant against incumbents.

OmniConnect gearbox solutions. Connect different-speed networks together — solving a real architectural problem in clusters that mix 400G and 800G ports. Smaller TAM individually, larger in aggregate as cluster topologies get more heterogeneous.

If even one of these four vectors lands at the scale management is targeting, the FY28 revenue model breaks materially higher. If multiple land, FY28 is unrecognizable from current consensus.

Here's where the business gets really interesting — and where the recent DustPhotonics acquisition adds a wrinkle.

Q3 non-GAAP gross margin came in at 68.6%. That's not a one-quarter peak — Credo has been operating in the high-60s for several quarters. For context: Marvell's blended gross margin is in the high 50s. Broadcom's networking semiconductor business runs in the high 60s. Credo is operating at margin parity with the most established players in the space — at a fraction of their market cap.

The reason is the SerDes IP. Credo's intellectual property is the chip that makes the cable work. Once designed and qualified at a hyperscaler, it's exceptionally sticky. Customers don't requalify SerDes IP unless they're forced to. That stickiness translates to durable pricing, which translates to durable margins.

The wrinkle: Q4 gross margin guidance came in at 64-66%, down from Q3's 68.6%. That's primarily DustPhotonics integration costs flowing through, plus some product mix shift as optical transceivers ramp. Long-term guidance from management remains above 60%. So the question becomes whether the Q4 step-down is a transitory integration hit or the start of structural margin compression as Credo moves further into commoditizing optical transceiver markets. I think it's the former, but it's the central thing the bears are anchored on, and it's worth weighing.

Operating leverage on high-60s gross margins is still the fastest path to free cash flow per share growth in the sector right now. Even at 64% — the low end of Q4 guidance — the operating leverage math still works.

I don't write hagiography. The bear case is real and worth weighing.

Customer concentration. Microsoft has been the largest AEC customer for several years. If Microsoft slows AEC purchases or qualifies a second source, near-term revenue takes a meaningful hit. The mitigant is that Credo has been actively diversifying — Amazon, Meta, and at least one additional hyperscaler are now buying — but concentration remains the single largest near-term risk.

Margin compression risk. Q4 gross margin guidance came in at 64-66%, down from Q3's 68.6%. Some of this is DustPhotonics integration costs, which is transient. Some of it reflects optical transceivers carrying lower gross margins than pure SerDes IP, which is structural. The Seeking Alpha bear case anchors on this — arguing that as Credo's mix shifts toward optical, blended margins compress materially. My read: integration costs dominate the Q4 step-down, and structural margins should stabilize in the mid-60s as optics scales. But this is a real debate, and how it resolves over the next two quarters is the most important fundamental question facing the stock.

The CPO question. The most common pushback I hear: doesn't co-packaged optics eventually replace everything copper-based? The short answer is no, not in the way that question usually implies. CPO replaces pluggable optical transceivers at switch front panels — it doesn't replace intra-rack copper, which is where Credo's AECs play. CPO and AECs solve different distance regimes: CPO is a 3-to-30-meter solution at the switch, AECs are an under-3-meter solution at the rack. NVIDIA's Quantum-X and Spectrum-X CPO programs explicitly preserve copper for intra-rack connectivity, and the Rubin platform roadmap continues that architecture. The genuine long-term risk to AECs is not CPO but a future generation of short-reach optics displacing copper inside the rack itself — a 2028-2029 concern at earliest. For the 12-24 month revision horizon, CPO is more likely a tailwind for Credo's ZeroFlap optics product than a headwind for their AEC franchise.

Multiple compression on disappointment. Stocks growing this fast trade on revenue acceleration, not earnings. If a single quarter shows decelerating sequential growth — even from a high base — the multiple can compress 30-40% in a session. Adjacent names this cycle have shown exactly that pattern.

Cyclicality risk. AI capex is enormous but cyclical at the margin. If hyperscalers digest capacity in any quarter, AEC orders can decelerate sharply. The 2026 capex picture suggests this isn't an immediate concern — Microsoft, Meta, Alphabet, and Amazon have collectively committed to $725 billion in 2026 capex, with Alphabet explicitly telegraphing a higher 2027 number — but it's the binary risk for any company this levered to the build cycle.

Here is where the picture gets uncomfortable for new buyers and comfortable for the revision thesis.

Credo trades at a forward EV/sales multiple in the high teens. On a forward earnings basis, somewhere in the 50-60x range depending on which forward year you anchor on. By any traditional yardstick, this is expensive.

By the revision yardstick, it's something else entirely. If FY27 earnings estimates get revised up 30% over the next two quarters — well within the trajectory of recent quarters — the multiple compresses materially without the stock moving at all. The bull case isn't that Credo is cheap. The bull case is that Credo's earnings power keeps moving up faster than the stock, so the multiple naturally compresses through estimate revisions rather than price decline.

This is the cleanest expression of revision-driven investing you'll find in the current market: a stock that's expensive on trailing numbers, mid-priced on next-year numbers, and outright cheap on the trajectory implied by the revision rate. Whether you can hold it depends on whether you can stay disciplined when the multiple optically looks rich.

I don't issue price targets. I issue revision conviction.

Credo Technology is mid-cycle on a revision sequence that has run for four consecutive quarters and shows no signs of decelerating. The pre-announce from February was the tell. ZeroFlap optics, the Robin DSP family, OmniConnect, and now DustPhotonics integration are the four vectors that can keep estimate revisions running through FY27 and into FY28. Customer concentration is the binary risk. Margin compression from the optics mix shift is the central debate. Margin profile, even at the low end of Q4 guidance, remains best-in-class for the sector.

The methodology says: stay long the revision until the revision turns. Right now the revision is accelerating, not decelerating. Until the next quarter shows otherwise, the trade is still working.

I've added Credo to the Classical Investor watchlist with a near-term focus on the Q4 FY26 print — expected next month — the FY27 guide, and the first read on DustPhotonics revenue contribution, all of which I expect to be major revision catalysts.