Credo Technology Group Holding Ltd. Q3 FY2025 earnings call

Key Points

• Q3 revenue was a record, driven by the largest hyperscale customer scaling AI platforms.
• Tiered innovation includes SerDes technology, integrated circuit design, system-level approach, and software/firmware platform.
• AEC product line saw growth, with ZeroFlap AECs replacing optics and progress with additional hyperscalers.
• Optical business has opportunities across customer base, with 3-nanometer tape-out for leading-edge opportunities.
• Retimer business gained momentum, achieving PCIe compliance and receiving platform commitment from an AI server ODM.
• Team execution highlighted for navigating significant demand ramp.

Credo reported revenue of $135 million in the third quarter of fiscal 2025, up 87% sequentially and 154% year-over-year. The product business generated $132 million in Q3. The AEC product line was driven by the largest hyperscale customer, with ZeroFlap AECs replacing optics for certain applications and progress with additional hyperscalers. The optical business had opportunities across the global customer base, and the retimer business gained momentum including entry into PCIe retimers. The largest customer accounted for 86% of Q3 revenue, with plans to diversify customer base in coming quarters.

Forward-Looking

• Dan Fleming mentioned Q4 guidance with midpoint expectations and anticipation of customer concentration decreasing in fiscal 2026.
• Expect 3 to 4 customers contributing over 10% of revenue in coming quarters and fiscal year, as additional hyperscalers ramp.

Risks

• Customer concentration risk due to largest customer accounting for 86% of Q3 revenue, though diversification expected.
• Market shifts impact on connectivity requirements, such as focus from training to inference.
• Competitive risks in AEC space, though Credo's system-level approach provides advantage.

Q: For the first one, if you could give us a sense of how large was the largest customer. I think Dan mentioned the number I didn't catch it. But I guess I have 2 parts to the question. One is, where are you in the adoption of AEC at that customer? And if you exclude that customer, how are you looking at the growth of your business among other customers? Because depending on that customer concentration, rightly, we get a slightly different trend outside of that large customer.

A: Daniel Fleming said the largest customer was 86% of revenue. Bill Brennan added on customer diversification, mentioning 3 hyperscalers in volume production and 2 in qualification, expecting solid diversity long term.

Q: For the first one, if let's say, more of the market turns towards inference rather than training, right, whether more [Audio Gap], but [Audio Gap] of the compute build-out. What does that do to these air clusters? And what does that do to connectivity requirements and the AEC potential? Is that good, bad, neutral for AEC?

A: William Brennan said inference market growth would create a larger AEC opportunity due to more deployments.

Q: So I think a lot of the tone of the call here has been about like a broadening of the portfolio. You're talking about PCIe retimers. You're talking about PCIe cables. To push you maybe 1 step further, if you're looking in the PCIe realm, an area of a lot of value is in the switching ecosystem as well. Do you guys have plans to move into the switching ecosystem? Is that like a natural progression from kind of the road traveling down already? And maybe talk about the challenges of doing retailers versus moving to switching and the time line that it would take to transition from product in the market today to maybe some new products on the switching side.

A: William Brennan said moving into switching is a natural progression, but currently focused on retimers and AECs, with challenges related to technology transition but seeing opportunities in scale-up networks.

Q: Two questions, if I may. The first one, I guess, what's driving the uptick in gross margins in the April quarter? Is it IP licensing revenue related? Or is there some other thing that we should be thinking about?

A: Daniel Fleming said gross margin uptick was due to scale, with product gross margin excluding engineering services up 200 basis points sequentially, and warrant with Amazon contributing to margin.

Q: Just a follow-up on your prepared comments, you spoke about how you may have 3 or 4 customers that would be 10% plus in the coming quarters or in calendar '26. Obviously, the first 3 today are AEC related. Is the fourth one also AEC-related, or would that just perhaps broadening into other aspects in product portfolios of your business?

A: Daniel Fleming said the 3-4 customers referred to are AEC related, with AEC being the largest driver of revenue growth.

Q: If I've got my numbers right, revenue outside of your largest customer went from about $48 million in October to about $19 million in January. You talked about kind of revenue diversifying again over the next few quarters and into fiscal '26 with 3 to 4 customers that could be over 10%. Can you just give us some sense. What do you expect your largest customer to do? Does it stay 80-plus percent of revenue? Are you anticipating that, that pulls back and that you see ramps at some of the other customers in the April quarter. Just any sense of revenue rediversifying would be helpful.

A: Daniel Fleming said largest customer's revenue would be similar to Q3, with Q4 guidance implying they'd be 2/3rd of revenue, and William Brennan added on customer visibility and ramp diversifications in fiscal '26.

Q: Bill, in your prepared remarks, you talked about several layers or multitiered innovation. And the one thing that obviously is that you talk about is the system-level approach. That's obviously very obvious on the AEC part of your business. But as you venture into some of these new segments and especially on PCIe. Will you take more of a system business model approach there as well, or should we assume that's going to be primarily selling chips?

A: William Brennan said would take system business model approach with PCIe AECs, demonstrating at GTC and planning similar system-level approach as with Ethernet.

Q: Just on the AEC side, I know you mentioned 2 new customers to new hyperscalers that you'll be working with. On the visibility, do you see them happening to -- as you look out 12 to 18 months, you see that ramping pretty nicely to significant volumes. How do you see that ramp.

A: William Brennan said design engagements are identified, with fiscal '26 expected for ramping, confident with 2 new customers in qualification.

Q: Congrats on a nice set of results and guide. I'd like to trend a little bit and ask about front-end networking. The original -- obviously, the original AEC deployment was on a front-end solution. And I wonder whether or how you and your customers are thinking about that front-end opportunity that still potentially exists? Is it just that in end networking is not where the innovation is happening in the space, and so customers are focused on the back end? Or is it a matter of time or some hybrid of the both? And then to squeeze an extra question on top of that, is there really a difference in the product that you ship to front end or go-to-market for the front end versus back end?

A: William Brennan said front-end and back-end networking are similar, with AI clusters driving faster connection needs, and largest customer using AECs for both.

Q: I guess -- yes, as it concerns your largest customer here and maybe any of these large ramps going forward, I think a lot of these things are project-based. We have very large deployments, sometimes 200,000 GPUs in just a couple of months. So as we think about these kind of large, lumpy deployments, should we be expecting these customers, for example, in this case, a pretty strong hand off in the next few quarters to other customers? Like Bill, maybe if you could talk to how project-based these revenue ramps are, that would be great.

A: William Brennan said revenue ramps are project-based, with diversification in shipments in fiscal '26, no air pockets, but transitions between technology generations.

Q: Analog, copper cable solutions seem to have missed perhaps the B-300 cycle. Is that an opportunity for you guys? Do you think you're gaining share and design wins just as that technology did not ramp, does that increase your TAM and your ramp expectation?

A: William Brennan said TAM exists in NVIDIA ecosystem, but not built into forecasts, with focus on adding value where opportunities arise.

Q: Bill and Dan, maybe I'll just ask a simple question here that I think you get most quarters here, Bill, but I just want to get your latest thoughts on competition in the AEC space. Obviously, you've got a very high share, you're doing exceptionally well here. It certainly makes sense for some of your large customers to attempt to do dual sourcing. Have you seen any evidence of that, either attempts or see that coming here anytime soon just kind of your latest thoughts on that, please?

A: William Brennan said focus is on being best partner, delivering innovations first, and system-level approach gives competitive advantage, no significant changes on competitive front.