Credo Technology Group Holding Ltd CRDO

Key data

The business

Credo sells the plumbing that moves data between chips at high speed. Its signature product is the Active Electrical Cable, a thick copper cable with a small signal-processing chip built into each plug. Inside a rack of AI servers, thousands of these connect GPUs to switches. Copper is cheaper and runs cooler than the light-based optical cables it competes with, and Credo's chip is what lets copper carry 400 and 800 gigabits over the few meters inside a rack. Around that sit its SerDes and DSP designs, sold as chips and as licensed blueprints, plus a growing optical line for the longer runs copper cannot reach.

The moat is a head start in a category Credo largely created, defended by design wins that lock into a customer's hardware for years. The weakness is the other side of the same coin: a handful of hyperscalers drive most of the revenue, and Credo is fabless, so it owns the design but rents the manufacturing from a foundry. What it holds is narrow and deep, not wide.

The numbers

The ramp is the whole story, and it is violent. Revenue tripled in one fiscal year and the last five quarters read like a company that found a new gear.

Revenue rose every quarter, up 115% year over year in the latest. The September 2 print, the first read on the company since late August, settled part of what was then open: the revenue ramp held cleanly, up 9.6% sequentially, but the margin story broke. Net income fell 24% from the prior quarter even as sales rose, because operating spending jumped 33% in three months. That tension is the memo.

Revenue compounded about 94% a year from FY2023 to FY2026, and the company crossed from four straight years of losses into real profit only in FY2025. That is why the multiple has no history to lean on: two years ago there were no earnings to price. Diluted shares grew 28% over those same three years, so the owner's slice thinned as the pie exploded.

Here is the crack. Gross margin held near 68% for a year, then dropped to 64.5%, and operating margin fell from a 37% peak to 25% in two quarters as research and selling costs outran revenue. What this memo believes the market is waving off: the FY2026 margins may have been the cyclical high, not the run rate. The print that settles it is next quarter's gross margin, filed around December. Recover toward 67% and the dip was the cost of a ramp; stay near 64% and 61x trailing was 61x on peak earnings.

Management

The record is a founder-led team executing a once-in-a-decade demand wave, with two flags. Insiders sold about $69.5M across 99 sales in twelve months and bought nothing, the chief operating officer the largest seller; this pull does not split the routine scheduled selling from the discretionary kind, and at one-way selling into a tripling stock that distinction is the thing worth knowing. Pay is not the problem, the chief executive's roughly $14.4M is about 3% of net income. The buyback is theater, $19M repurchased at an average $179, above today's price, dwarfed by stock-based pay running near 15% of revenue, so the real capital story is dilution, not return. The balance sheet is a fortress: essentially no debt and about $765M in cash and investments. A $113M acquisition last year left goodwill and intangibles at 45% of assets, and the deal is not identified in the filings reviewed.

How it fails or surprises you

Customer goes in-house or dual-sources. The thesis rests on two or three hyperscalers. If one designs its own retimer or splits orders to Marvell or Broadcom, revenue steps down fast and the multiple with it. The tell shows up first as a named 10%-plus customer shrinking in the concentration disclosure, or a guidance cut citing a lost socket.

The margin dip is the trend, not the blip. Profit fell 24% sequentially while revenue rose 10%, on both a lower gross line and a spending surge. This is the fact the bull case explains least well. If December's gross margin fails to climb back above 66%, the story quietly changes from compounder to a company that already earned its best margin.

Optical and new sockets scale (right tail). The market prices the copper cable. If Credo's optical DSP and SerDes chiplet lines ride the 800-gig and 1.6-terabit transition into a new 10%-plus customer, the attach compounds faster than the copper-only model assumes, and the current spending surge is revealed as the cost of buying that next wave rather than margin decay.

Closing thoughts

Two forces sit on top of each other. One is an uncertainty a single print resolves, next quarter's gross margin, which converts the setup into either "ramp cost" or "peak passed." The other is an exposure no print resolves, the customer concentration, where only survivability applies and the balance sheet, not a model, is the answer. Judgment, not arithmetic: the right tail is genuinely live because the connectivity content per AI rack keeps rising, but the left tail is a single customer decision at 61 times earnings, and that is the one to respect. The fatter near-term tail is the downside, because the margin question and the concentration question can both break in the same quarter.

The bet, stated the way it started: cloud companies keep building AI data centers and keep buying Credo's smart copper cables, and the few giant buyers keep ordering. What breaks it is one of those buyers going in-house or to a rival. The one pair that tells you first is a 10%-plus customer's revenue stepping down while gross margin stays stuck below 66%. If both hold, the ramp is real; if either turns, the price was the story.