Credo: Reliability Leader Aggressively Moves into Optics

Credo is a force to be reckoned with in the networking space. The company brought to market active electric cables (AECs) that integrates a DSP and retimer, which were more reliable and power efficient than the alternative, laser-based optical modules for short-reach connections.

As Nvidia’s racks scaled-up to 72 GPUs, the impact of Credo’s AECs were instrumental to the systems seeing “up to 1,000x greater reliability than commodity laser-based optical modules” while consuming much less power. In past our past coverage on Credo, we’ve included pictures of their signature purple cables blanketing Nvidia’s in-rack connectivity, which resulted in hypergrowth fundamentals for Credo especially when Blackwell and Blackwell Ultra shipped.

However, as we approach Rubin and certainly Rubin Ultra, it’s expected the copper-to-optics boundary will move in optics favor. As a growth investor, I have to be cautious anytime a company’s market share recedes to the adjacent market, even if that company’s revenue is growing YoY. In other words, Credo’s AECs alone are unlikely to sustain 200%+ growth, especially given that optics are becoming the preferred approach as reach and speed requirements increase, combined with tough comps from the Blackwell architectures.

And yet, despite the AEC story offering less gunpowder, the stock has been particularly resilient and is approaching an important technical level. The reason is that, as the reliability-and-power-efficiency supplier, Credo may be innovating within its space again – but this time, with optics.

The ramp that Credo foresees is fast at $600 million this fiscal year across three product lines. When pressed in the earnings call, management agreed that the ZeroFlap optics platform is expected to be the fastest-growing product within optics as it solves the same problem that Credo’s AECs are known for, which is network reliability. By continuously monitoring link health, the ZeroFlap optics platform autonomously detects and mitigates link instability before it impacts the cluster.

The ZeroFlap optics platform also comes with an upsell opportunity for the company’s PIC technology, which can reduce the number of lasers that a link requires, and 100G and 200G DSPs. Combined, the platform combines hardware and software to keep the network stable.

Awaiting the Optics Inflection in 2H FY27

Credo’s growth is decelerating as optics replace AECs with growth peaking at 201% YoY last quarter, and growth is now at 157% this quarter for revenue of $437 million. As we move along, Credo’s YoY growth is expected to sustain in the 50% to 80% range.

However, Credo’s guide makes it abundantly clear that optics is their growth story moving forward. The company is expected to exit the year with about $150 million or more in optics revenue per quarter, which represents the entirety of the QoQ growth as we enter FQ3 and FQ4. In other words, without optics, Credo would have been moving into flat to negative QoQ growth by the January and April quarter.

This is important because it helps to frame the challenge that lies ahead for Credo, which is that already in this fiscal year, they had to innovate quickly to keep growth rates sustained. As the company exits the year, it’s expected to report 30% QoQ growth by combining a strong base off AECs with its new optics growth market.

The growth is expected to sharply inflect in the back half of the year as Credo prepares to ramp a trio of optical products – optical DSPs, SiPho photonic ICs (acquired from Dust), and ZeroFlap optics. Credo expects each of the three to contribute more than $100 million in revenue in FY27, with optics in total contributing more than $600 million for the year with the ramp accelerating in 2H. As it stands, current estimates point to growth reaccelerating to nearly 28% QoQ in FQ3 (the January 2027 quarter) and maintaining roughly 24% QoQ in FQ4 as optics layers in to growth.

While analysts rightfully picked up on the fact that the $100 million each and $600 million total optical revenue guidance implied revenue is skewed towards one of the three products (it was later revealed to be ZeroFlap), the bigger takeaway is that optics are contributing half of Credo’s YoY growth on a dollar basis:

“If you kind of break that down or dive into that a little bit deeper, what you'll see is based on that guidance, if you look at absolute dollars year-over-year expectation for fiscal '27, about half of that growth in absolute dollars is coming from our optical portfolio and about half of that is coming from our existing copper portfolio, predominantly AECs, but also retimers.”

Credo also provided some context ASPs for its optical suite, with the ZF optics ramp in part due to it carrying a triple-digit ASP versus a double-digit ASP for PICs and DSPs:

“To give a little more color on your first question earlier, the ASPs on the discrete components, optical DSPs and cyclo PICs, those ASPs are typically 2-digit ASPs. On the ZF Optics, we're going to see 3-digit ASPs. And so I didn't mean to be a bit elusive or not answer your question. But I think the math clearly says that as we ramp ZF Optics, that's going to be clearly our largest revenue contributor for our optical portfolio that the potential there is great.”

CEO Bill Brennan also commented that a majority of the growth in FY27 will likely be attributed to scale-out demand, with scale-up only beginning to ramp with FY28 being more substantial, though mix will ultimately depend on customer deployment.

Major Nvidia Supplier for AECs

Credo is known for its ‘purple cables’, its AECs, that were pivotal in helping Nvidia scale to 72 GPUs per rack, with systems seeing up to 1,000x greater reliability versus commodity optical modules with lower power consumption.

As Blackwell and Blackwell Ultra ramped, Credo’s dominance over the AEC market was quite clear, with analysts implying in late October that the company held as much as 88% market share, benefiting from high-volume deployments. For example, xAI asked Credo specifically for 7 meter AECs as it re-architected its 100K-GPU Colossus cluster for liquid cooling, with the extended reach allowing them to “connect all of their GPUs, switches with AECs, which are fundamentally bulletproof from a reliability perspective.”

However, it’s not just for Nvidia where Credo finds success, as this reliability improvement and an up to 7 meter reach for both intra-rack and multi-rack connectivity has seen other hyperscalers adopt Credo’s AECs. For example, Amazon displayed the hallmark purple cables for intra-rack connectivity for its Trainium servers:

Source: Amazon

Despite constant shifts in narrative over copper’s remaining useful lifetime at 200G speeds and concerns over reach and signal integrity, Credo’s copper growth is remaining in the near term, and could still have a solid foundation with Nvidia’s Rubin, as the copper-to-optics boundary is not expected to move heavily in favor of optics until Rubin Ultra later next year.

More on ZeroFlap Optics, Ramp and Supply Preparations in Place

Given its experience in AECs and the success it has built on the accelerator-agnostic purple cables, it’s not a leap to think Credo will do well on an optical version of cables, its ZeroFlap (ZF) optics. Credo’s ZF optics support 400G, 800G and 1.6T speeds with reach of up to 500 meters, with Credo’s PILOT software used for telemetry and link health monitoring.

As noted above, ZF optics are expected to drive Credo’s revenue inflection in 2H as the line ramps, though management explained that there could be shifts in timing for the upgrade cycle from 800G to 1.6T depending on Nvidia’s Rubin and customer preference. This could have some impacts on the shape of the revenue ramp, as 1.6T carries a higher ASP versus the 800G products, similar to ASP dynamics at other transceiver vendors:

“On the transition from 800-gig to 1.6T. Of course, the timing is going to shift a lot about the deployment of Rubin and really each of the customers' individual strategy. Some will be very delayed. Some will be first to deploy. But I think the exact timing of the transition will move somewhat as the platforms evolve.”

Management added that 200G/lane revenue (1.6T) is expected to be relatively light in FY27, implying larger 1.6T contributions could arise in FY28.

Regardless of the timing of the 1.6T transition for Credo and the ASP uplift, management has been quietly preparing for the optics pivot behind the scenes, with CEO Bill Brennan explaining at BofA’s conference shortly after earnings that the company has been building demand for six months, and securing supply for 12:

“We've been in the demand generation mode for probably 6 months or so since OCP last year. But we've been in a mode of locking in supply for more than 12 months. And so leaning in with 3 partners that will assemble these transceivers, locking in supply of lasers, locking in supply of every component and the overall capacity.

I mentioned on the call that exiting this year, we'll be producing numbers that are measured in 100,000 unit increments monthly, and then we're going to be doubling and tripling that in the following year. So we're going to have the supply to match the demand that we're generating.”exiting this year, we'll be producing numbers that are measured in 100,000 unit increments monthly, and then we're going to be doubling and tripling that in the following year. So we're going to have the supply to match the demand that we're generating.”

For a bit of speculative, back-of-napkin math, assuming $0.75/gig pricing for the 800G optics and assuming those take majority share in FY27, ramping to 100K monthly capacity and sustaining that for one quarter implies quarterly optics revenue of up to $180 million. Doubling or tripling capacity to ~300K and shifting to a higher mix of 1.6T in FY28 (say 50/50 with 800G), for example, could help push optics-driven quarterly revenue towards $900 million in that upper scenario.

Credo’s Reliability Advantage

As mentioned above, Credo’s main advantage comes down to reliability, and it’s why you will hear management refer to reliability as their ‘North Star’ that guides their product roadmap. This dedication to reliability also underpins Credo’s pivot to optics, and why the company could have more of an edge than expected with its optics portfolio against the incumbents via rich telemetry.

As clusters scale from tens of thousands to hundreds of thousands of accelerators, reliability becomes one of the most important factors. Credo’s management explained that connections between top-of-rack (TOR) switches and NICs have no redundancy, so any instabilities within these first link connections from GPU to switch could bring the entire cluster down.

When you factor in the cost of building a 100K GPU cluster, or roughly 1,390 GB200 NVL72 racks costing approximately $4.2 billion (at a $3 million ASP), extended downtime from link instability affecting cluster stand-up or deployment directly impacts revenue and margins. This is especially important for neoclouds in ramp phases, who cannot afford delays and are increasingly turning to Credo for its reliability and advantage in helping stand up clusters quicker:

“So basically, bringing up a cluster quickly, time to stability, which is dollars. If you buy billions of dollars of gear and it takes you 8 weeks or 12 weeks to bring the cluster up to a stable point so you can start generating revenue, that is hugely expensive compared to bringing it up in a week and then keeping it at near 100% uptime.”

As to how this fits in to Credo’s burgeoning optics portfolio, the acquisition of Dust Photonics brought SiPho PIC tech into Credo’s stack, complementing its 100G and 200G DSPs, Robin and Cardinal. Extending its ownership of the optical stack down to the DSP and PIC level for its ZF optics gives Credo a much higher degree of visibility into signal performance and link behavior:

“[ZF optics] was designed for that link or eliminating link instabilities between GPUs and switches, not at a core technology level, but by going up the stack. So designing a custom DSP that was capable of lighting up rich telemetry on every link. … The tighter DSP to PIC integration enables richer telemetry, enhanced diagnostics and more intelligent system-level optimization. …

So lighting up rich telemetry on all of those links so that we can predict when a link instability is going to happen because we're monitoring signal integrity continuously across the entire cluster where ZF Optics are deployed.”

This is how Credo finds its differentiation factor versus the incumbent optics vendors, within telemetry, reliability and visibility into optical performance. Credo is not competing against commodity optics but rather providing an upgrade with a more feature-rich product, offering an ability to preemptively identify potential causes of link failure, address it before it occurs, and increase cluster reliability and time to deployment. These are all key features that will help hyperscalers and neoclouds maximize GPU utilization, minimize downtime and maximize revenues.

Dust Photonics for CPO/NPO

Credo recently closed on its acquisition of Dust Photonics, not only bringing its SiPho PIC tech to its optics stack as mentioned above, but also providing Credo with a direct path to CPO and NPO architectures, letting the company quickly branch into the upcoming higher-growth vectors in the optical landscape.

Dust provides an immediate expansion of Credo’s optics portfolio to 800G and 1.6T, with a roadmap to >3.2T, giving Credo a quick path to entry into the leading data rates on the market with potentially simplified development timelines for next-gen 3.2T products. This may help position Credo at the forefront of the optical transitions from 1.6T to 3.2T over the coming years, considering 800G is still ramping in volume and the 1.6T transition is currently underway.

Credo also expects the path to CPO and NPO via Dust to happen relatively quickly, building off of its SiPho suite. Management outlined initial revenue for both CPO and NPO designs to arrive in FY28 (starting July 2027) based on current customer engagements, which aligns with current timing discussions from Astera expecting NPO to begin ramping in 2027.

However, the most important part of the Dust acquisition is that Dust’s designs enable simpler optical architectures and reduced laser counts, both of which could facilitate stronger optical growth for Credo in a laser-constrained industry. Dust says its L3C tech uses simple lasers and reduced laser counts, which reduces bill of materials and increases yield for optical modules, while also reducing power consumption and lessening supply chain constraints. This could facilitate Credo’s entry and ramp into CPO and NPO-based architectures considering the supply chain challenges that Lumentum and other incumbents are facing when it comes to InP-based lasers.

Next Year’s Catalyst: Active LED Cables and Weaver Chiplets

Credo has a few more cards up its sleeve and two new catalysts on the horizon for 2027 (fiscal 2028): active LED cables and its Weaver gearbox chiplets. We first covered the two back in December with Q2’s report.

Active LED cables are pluggable optical solutions that use micro LEDs as the light source (acquired with Hyperlume), which effectively extend its AEC strengths to longer distances, serving both scale-out and scale-up needs. ALCs represent Credo’s preparation for a cable-based future beyond AECs at 400G speeds, as management notes ALCs deliver the same reliability, power and cost efficiency as AECs with connections reaching up to 30 meters, or >4X that of the 7 meters AECs offer.

Sampling of first ALC products is expected to occur in FY27, though the production ramp and revenue generation is currently slated for FY28. As we discussed in our Q2 write-up, management ultimately believes the ALC TAM to be more than double the size of its AEC TAM, with the opportunity likely rooted in the extension towards a 30 meter reach. Management this quarter described ALCs as having a very natural evolution from AECs for customers deploying the product, which could set the stage for a rapid revenue ramp come FY28 and FY29:

“And so for certain customers, that's going to be a really natural product to get from a design in and beginning of production. So the dynamic there could be very much similar to AECs and ZF Optics in the sense that we can see large revenue quickly.”

Moving over to Weaver, which is the first product of Credo’s OmniConnect strategy to leverage its copper (and optics) portfolio for die-to-die solutions. Weaver is Credo’s solution to attack the inference memory wall with a fanout gearbox chiplet leveraging 112G very-short-reach SerDes. This addresses the ‘fanout’ problem, where DDR memory systems become constrained by limited I/O density as memory dies can only be accessed by a limited number of pins. Credo/s fanout gearbox aims to increase I/O density by 10X and offer 16TB/s of memory bandwidth and 6.4TB of memory density to mitigate this limitation.

Credo’s first customer for Weaver, AI chip startup Positron, is preparing to launch two inference accelerator engines in 2027, Asimov and Titan. Asimov features LPDDR memory capacity of up to 2.3TB per chip (without using HBM), which Credo states is “more than 10x any other inference engine that's been announced in the market. So game changing from the standpoint of performance.” Compare this to the GB200 Superchip, which carries 372GB of HBM memory per chip, or AMD’s MI355X, which carries 288GB.

Asimov also packs in realizable memory bandwidth of 2.76 TB/s per chip with a power draw of just 400W, scaling to 16,384 chips in a single cluster, which could make it an interesting memory-optimized, power efficient choice for inference.

Positron’s Titan combines four Asimov chips into one system, offering more than 8TB of LPDDR memory with 32TB/s of external chip-to-chip bandwidth (half of Vera Rubin NVL72’s 65TB/s) in an air-cooled form factor capable of scaling to 4,096 chips per cluster – at that 4,096 chip cluster size, Titan would offer nearly 33 PB of LPDDR memory, whereas a similar sized cluster of Rubin NVL72’s would offer nearly 1.2 PB of HBM4. With this memory, Positron says Titan is capable of supporting 10 million token context windows and could enable a 2030-class of frontier- model in 2027.

While it is still unclear how quickly Positron will ramp these new chips, Credo is seeing a rather large revenue opportunity from Weaver, benefiting from a high ASP: “Now if 2 terabytes are deployed, that requires many, many Weaver chips. And what I've said in the past is that the revenue contribution per GPU can be between $2,000 and $3,000.  And so you can see that, that product category will ramp very quickly as well as we've got customers that go into volume with their inference GPUs that utilize that solution.”

Financials Overview

Credo’s growth decelerated in FQ4 to 157% YoY to $437 million, a more than 44 point deceleration, while QoQ growth came in at 7.4%, moderating from FQ3’s 51.9% print. Credo guided for this YoY deceleration to continue in FQ1 to 111% YoY at $465-475 million, while QoQ growth would be sustained at 7.6%. QoQ growth is expected to inflect come 2H FY27 (the Jan 2027 quarter) with Credo returning to >24% QoQ in both FQ3 and FQ4.

Credo is benefitting from strong operating leverage – while GAAP gross margin expanded 1 point YoY to 68.3%, GAAP operating margin expanded nearly 16 points YoY to a strong 35.7%. This operating leverage is visible in Credo’s strong GAAP EPS growth, up 340% in FQ4 to $0.88.

For a quick look at FY27, Credo guided for YoY growth of >80%, implying revenue exceeding $2.41 billion. Credo’s bottom line is expected to remain strong, with GAAP EPS currently expected to rise 98% YoY to $4.98.

Key Metrics

Credo’s inventories as of FQ4 were $250.8 million, up 20.6% QoQ from $208 million in FQ3, with this growth predominantly driven by a $49 million increase in raw materials, up 320% QoQ, offsetting a slight decrease in finished goods. This adds a layer of confidence in Credo’s upcoming growth inflection and optics ramp in 2H. Credo does have a high degree of customer concentration, reporting three >10% customers in FY26, accounting for 84% of revenue; its two largest customers accounted for 65%, at 32% and 33% respectively (with the 33% customer down from 63% in FY25).

Conclusion

Credo’s AECs were instrumental in driving up to 1,000x greater reliability versus commodity optics as Nvidia’s racks scaled to 72 GPUs, while consuming much less power. The company is looking to take this reliability and power-efficiency advantage to quickly iterate for optics, with its ZF optics platform expected to be the cornerstone of a rapid ramp to >$600 million in optics revenue this year.

The driving factor behind management’s confidence in achieving this accelerated ramp in a crowded optics landscape is two-fold: leveraging its reliability strengths, with ZF optics continuously monitoring link health, autonomously detecting and mitigating link instabilities to maximize GPU utilization and minimize cluster downtime; and building a full-stack platform with upsell opportunities for its PICs and 100G and 200G DSPs.

Damien Robbins, Equity Analyst at I/O Fund contributed to this analysis.

Please note: The I/O Fund conducts research and draws conclusions for the company’s portfolio. We then share that information with our readers and offer real-time trade notifications. This is not a guarantee of a stock’s performance and it is not financial advice. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis. Beth Kindig and the I/O Fund do not own shares in CRDO at the time of writing and may own stocks pictured in the charts.

Recommended Reading: