Credo Tech: A Problem, Yes. But Mostly Interesting.

by

Credo Technology Group (CRDO +3.87%) recently announced its quarterly results, which, as these things usually are, involved numbers. Quite a lot of them, in fact. $407 million in revenue, to be precise – a 201% year-over-year increase. A gross margin of 68.5%, which, when you think about it, leaves rather little room for error. And $1.3 billion in cash. One begins to suspect they might be hiding a small planet under the accounts receivable. (It’s probably just more cash.) Non-GAAP adjusted net income reached $208.8 million. All of which, on the surface, suggests a company doing… well, rather well.

So, what’s the snag? The mildly unsettling fact that two customers account for roughly 80% of that revenue. This, naturally, causes a degree of consternation amongst portfolio managers. And with good reason. When your business model is predicated on the whims of a handful of hyperscalers – those vast, data-intensive entities that seem to power everything these days – a delayed AI build-out (or, worse, an inventory digestion quarter – a phrase that sounds vaguely medical) can send the stock plummeting faster than a poorly-aimed satellite.

Concentration: A Risk, But One That’s… Shifting

Indeed, such a plummet occurred in early 2023 when Credo’s largest buyer revised its demand forecasts. The result? Two quarters of flatlined growth. A rather disconcerting experience, one imagines, for everyone involved. But things are changing. The customer base is, thankfully, diversifying. In fiscal 2024, two hyperscalers accounted for more than 10% of revenue. By Q1 of fiscal 2026, that number had increased to three, with a fourth rapidly approaching the threshold. And, crucially, they aren’t being lured with discounts. They’re coming to Credo because, quite simply, no one else makes what they make with this level of reliability.

Credo’s core product is its Active Electrical Cable (AEC), which connects GPUs inside data centers. Now, imagine a cluster with 100,000 or more GPUs. A truly staggering number, isn’t it? And within that cluster, a single faulty link – what’s charmingly referred to as a ‘link flap’ – can crash an entire AI training run. The system has to reload from a checkpoint. Millions of dollars in compute time, vanished into the digital ether. (It’s a bit like trying to build a sandcastle during a hurricane, really.) Credo’s AECs are, astonishingly, up to 1,000 times more reliable than the optical modules they replace. That’s the difference between completing a training run and experiencing a profoundly expensive failure.

The practical implication is that hyperscalers aren’t evaluating Credo’s cables on price. They’re evaluating them on whether their multibillion-dollar AI clusters actually, you know, function. This fundamentally alters the competitive dynamic. It’s no longer about being slightly cheaper; it’s about preventing catastrophic, wallet-emptying outages.

Management is currently guiding for Q4 fiscal 2026 revenue of $425 million to $435 million. And for fiscal 2027, they project more than 50% year-over-year growth, potentially propelling the company towards nearly $2 billion in annual revenue. A considerable sum, when you think about it. (Enough to buy a small country, perhaps? Or at least a very large collection of rubber ducks.)

Loading widget...

The Moat: It’s the Technology, Naturally

And new product lines are ramping up simultaneously. Credo is expanding into ZeroFlap optics for longer-reach connections, advanced line cards for switching fabrics, and OmniConnect gearboxes. All of which expands their total addressable market by, well, billions of dollars. (It’s a bit like discovering a previously unknown dimension, really. Full of potential revenue, naturally.)

They’ve also announced three new multibillion-dollar TAM expansions in recent quarters. PCIe retimers are gaining traction across AI servers. Optical DSP demand is increasing at 50, 100, and now 200 gig per lane. These are revenue streams that barely existed 18 months ago. (Time, as they say, is a curious thing. Especially when viewed through the lens of quarterly earnings reports.)

Yes, customer concentration remains a genuine risk. But Credo isn’t selling a commodity. They’re selling infrastructure that keeps hundred-thousand-GPU clusters online. And that, as any seasoned investor will tell you, is a rather compelling proposition.

I’m buying because the moat isn’t built of water and stone; it’s built of the product itself. And the product pipeline isn’t narrowing; it’s widening. And frankly, that’s a rather comforting thought in a universe that’s demonstrably, and increasingly, absurd.

Read More

2026-03-13 20:12