Credo Technology Group Holding Ltd
NASDAQ:CRDO
Earnings Call Analysis
Q3-2024 Analysis
Credo Technology Group Holding Ltd
In the third quarter, the company experienced an encouraging 20% sequential revenue increase to $53.1 million, although there was a slight 2% dip year-over-year. The standout performer was the product business, which surged by 41% sequentially and 24% year-over-year, contributing the lion's share of the revenue. However, the IP (Intellectual Property) revenue faced a sharp decline both sequentially and year-over-year, accounting for a mere 2% of total revenue. Despite this, management remains optimistic about IP revenue aligning with their long-term expectations of making up 10-15% of total revenue.
The company reported robust non-GAAP gross margins of 62.2%, notably higher than their guidance. This impressive margin was driven by their product business which achieved a striking year-over-year jump thanks to a significant increase in product NRE (Non-Recurring Engineering) revenue. Meanwhile, the non-GAAP operating income turned positive, swinging from a loss in the prior quarter to $2.4 million, thanks to better top-line leverage. However, operating expenses also crept up by 19% year-over-year, largely due to R&D investments, with the goal of staying at the forefront of innovation.
Thanks to a successful follow-on offering, the cash and equivalents ballooned by nearly $170 million from the previous quarter, reflecting a robust financial position. This positions the company to invest in growth opportunities while maintaining a solid cash buffer. The accounts receivable also saw a marked increase, indicating more revenue is due to be collected. However, the company noted an outflow in free cash flow, which could be attributed to capital expenditures driven by R&D and production investments.
Looking forward, the company projects Q4 revenue to range between $59 million and $62 million, which would mark another sequential uptick of 14% at the midpoint. Gross margin expectations are also set to remain strong, targeted within 64-66%. However, as the company doubles down on R&D to keep its edge, operating expenses are anticipated to climb even higher.
The company puts considerable emphasis on its AEC (Application-Specific Integrated Circuit) business and optical DSP (Digital Signal Processing) as long-term revenue drivers. In particular, the AEC segment is receiving firm attention due to its alignment with the industry's shift towards Ethernet networks for operational efficiency. The company's dedication to delivering the most energy-efficient solutions is expected to maintain its competitiveness. Chiplet designs, particularly in AI-related domains, are also gaining traction, with multiple customers engaging with the company for these solutions.
Nonetheless, the revenue from AEC is reliant on the scale of deployment by hyperscalers—a factor not wholly within the company's control. And despite the higher-than-average product NRE in Q3, future NRE income is expected to revert to historical averages. Hyperscalers' timelines and large program ramps introduce an element of unpredictability to the revenue forecast. Still, the executive team is confident about the AEC's path to becoming a preferred choice thanks to its value proposition in Ethernet network architectures.
You don't have any saved screeners yet
Good day, and thank you for standing by. Welcome to the Credo Fiscal 2024 Q3 Earnings Conference Call. [Operator Instructions]I would now like to hand the conference over to your speaker today, Dan O'Neill.
Today, I'm joined by Bill Brennan, Credo's Chief Executive Officer; and Dan Fleming, our Chief Financial Officer. As a reminder, during the call, we will make certain forward-looking statements. These forward-looking statements are subject to risks and uncertainties that are discussed in detail in documents filed with the SEC, which can be found in the Investor Relations section of the company's website. It's not possible for the company's management to predict all risks nor can the company assess the impact of all factors on its business or the extent to which any factor or combination of factors may cause actual results to differ materially from those contained in any forward-looking statements. Given these risks, uncertainties and assumptions, the forward-looking events discussed during this call may not occur, and actual results could differ materially and adversely from those anticipated or implied. The company undertakes no obligation to publicly update forward-looking statements for any reason after the date of this call to conform these statements to actual results or any changes in the company's expectations except as required by law. Also during this call, we will refer to certain non-GAAP financial measures, which we consider to be important measures of the company's performance. These non-GAAP financial measures are provided in addition to and not as a substitute for or superior to financial performance prepared in accordance with U.S. GAAP. A discussion of why we use non-GAAP financial measures and reconciliations between our GAAP and non-GAAP financial measures is available in the earnings release we issued today, which can be assessed using the Investor Relations portion of our website. I will now turn the call over to our CEO.
I'll begin by providing an overview of our fiscal Q3 results and our outlook for the future. Our CFO, Dan Fleming, will then provide a detailed review of our Q3 financial results and your expectations for fiscal Q4. For Q3, Credo reported revenue of $53.1 million and non-GAAP gross margin of 62.2%. These results and our future growth expectations continue to be driven by the accelerating opportunity for high-speed and energy-efficient connectivity solutions throughout the data infrastructure market.Our connectivity solutions include active electrical cables or AECs, optical DSPs, Laser Drivers and TIAs, Line Card PHYs, SerDes chiplets and SerDes IP licensing, each leveraging our SerDes technology. Core to Credo's success is our SerDes technology, which is the fundamental connectivity building block of all of our solutions.Credo continues to invest deeply in SerDes architectures that enable application-specific solutions optimized for speed, reach performance, energy efficiency and cost. Our 100-gig per lane SerDes portfolio spans process geometries from 12-nanometer to 3-nanometer with optimized reach performance from the longest reach links to the most power-sensitive die-to-die links for chiplets. Credo will continue our innovation for 200-gig per lane series as the opportunity to differentiate increases, specifically related to the optimization of trade-offs between process geometry, DSP architecture, performance and energy efficiency. Now regarding our optical solutions, Credo continues to gain traction in the optical DSP market. During the third quarter, we continued production shipments of optical DSPs to multiple global hyperscale end customers for a variety of applications across numerous port speeds. We closely target the combination of optical module partners and hyperscale end customers, and we're making progress on optical DSPs for 400-gig and 800-gig optical transceiver and AOC opportunities. Enabled by our SerDes design approach, Credo wins by delivering a compelling combination of performance, energy efficiency and value as well as focusing on innovative ways to solve complex customer needs. Last quarter, we talked extensively about the industry call for action for better power efficiency for 800 gig and 1.6T optical solutions. As we discussed, we believe eliminating the DSP with the linear pluggable optics or LPO architecture is simply too far elite, especially for 1.60T. Credo quickly responded with our innovative linear receive optics or LRO DSP architecture that directly addresses the power challenge while delivering better signal integrity, maintaining industry standards, IEEE compliance and interoperability and extending to 1.6T solutions. Since our discussion last quarter, we've made meaningful progress. Our first 800-gig LRO DSP partner has built and tested 800-gig optical modules, and the results are exactly as expected, successfully delivering on the promise of much reduced power and great signal integrity while overcoming the shortfalls of the aspirational LPO architecture. Next month at OFC in San Diego, we'll be demonstrating both 800-gig LRO DSP and full DSP solutions, highlighting our progress. We'll be demonstrating 800-gig solutions with five different optical module partners. Our 1.6T road map includes both LRO DSP and full DSP solutions with 200 gig per lane speeds with our top priorities being energy efficiency and signal integrity, which are both critical to achieve robust 1.6T optical solutions. We believe the growth in our optical business will be primarily driven by U.S. hyperscaler end customers with further contributions from global hyperscalers. I'll now discuss our Line Card PHYs business, which delivered another solid quarter in Q3. In this segment, our customers include networking OEMs and hyperscalers and our products include Retimers, Gearboxes and MACsec PHY for data encryption. Our customers are the leaders in this space. We have close working relationships and our design wins typically have long life cycles once they ramp to production, contributing nicely to our overall results. In the upcoming months, we'll take out our customer sponsored 5-nanometer 1.6T MACsec and our power optimized 1.6T Retimers and Gearboxes. Credo is among the industry leaders for 50-gig and 100-gig per lane line card PHY applications. We have many customers that have deployed our 50 gig per lane solutions in production, and we count more than 10 customers designing in our 100-gig per lane Screaming Eagle Line Card PHYs. Now turning to our IP and chiplet business. Our SerDes IP and SerDes chiplet businesses continue to be a strategic component of our overall business. These solutions enable our partners to address the ASIC market for next-generation solutions. Due to revenue recognition rules, our SerDes IP revenue can vary meaningfully from quarter-to-quarter, which is what we saw in Q3, and we'll likely see in upcoming quarters.Our funnel for licensing opportunities remained strong, bolstered by the increased opportunity on ASICs with high-speed SerDes for data center applications. We have a comprehensive portfolio of SerDes IP for our AEC customers, including 100-gig per lane SerDes across the broadest range of process geometries from 12-nanometer to 3-nanometer and the broadest range of reach performance from long reach to die-to-die reach. Our SerDes IP offering enables our ASIC partners to optimize their solutions for process geometry, reach and power.Last quarter, our chiplet business was highlighted by a win with significant NRE at one of our leading customers for next-generation 5-nanometer chiplet solution, which speaks loudly regarding our differentiated SerDes portfolio. When complete, Credo will be able to market and sell this chiplet to the broad market. Overall, we remain optimistic about the prospects of the chiplet category given our results to date and due to our belief that chiplets will be a key enabler in the most advanced system solutions. In summary, we're pleased with our results for fiscal Q3, and we're optimistic about the increasing market demand for high-speed connectivity. Credo's competitive advantage is driven by our focus and execution on our core SerDes technology, and that leads Credo to being one of the few companies capable of delivering the necessary breadth of connectivity solutions at the highest speeds while optimizing for energy efficiency and system costs. For these reasons, we expect continued long-term growth across a diversified customer base and a diversified set of connectivity applications. I'll now turn the call over to our CFO, Dan Fleming. Dan will provide additional financial details, and then we'll be happy to take questions. Credo's SerDes technology expertise, combined with our system level, customer-focused design approach has led to our success with a diverse and growing set of industry-leading customers. In 2023, the technology industry experienced an inflection point driven by generative AI applications. The acceleration in the deployment of AI clusters has put high-speed connectivity on center stage given the fundamental need for higher bandwidth. For Credo, this need for higher bandwidth translates to the demand for higher speed, higher density and more energy-efficient connectivity solutions. This plays directly to Credo's strength and underpins our growth expectations. I'll now provide more detail of our overall business. First, I'll discuss our AEC business, where Credo continued to build momentum during the third quarter. We believe our AEC leadership derives from our comprehensive systems-level approach to the AEC market. We've built this from the ground up over many years, and as a result, we now have the ability to quickly innovate to support the diverse AEC needs of our customers. Given the increasing single-lane speeds and the shortcomings of both passive copper cables and active optical cables and transceivers, we foresee continuing adoption of AECs for In-rack cconnectivity. We expect U.S. hyperscalers to remain the majority portion of AEC demand in the foreseeable future. Many of these customers have distinct architectural requirements that demand innovation and tight collaboration between the engineering teams, at Credo and the customer. We're engaged at different stages with the five U.S. hyperscalers as well as other global hyperscalers and Tier 2 data center operators.We've delivered a range of products with nonstandard optimized hardware and firmware features to meet our customers' needs for port speeds from 100 gig–800 gig depending on the customer application. Credo continues to work closely with our first two hyperscale customers, delivering AEC solutions for both front-end and back-end Ethernet networks as each publicly highlighted during their conferences last quarter. We're gaining better visibility with these customers into their near-term product ramps, and we're also engaged in a range of AEC solutions to address longer-term road maps. While advanced programs with this nature take time to achieve material deployment rates, we continue to expect an inflection point in the second half of our fiscal '25. Credo is also working with additional U.S. and global hyperscalers to develop 400 gig and 800 gig AEC solutions that we expect will yield significant revenue for Credo in the future as next-generation network architectures transition to AEC solutions. Additionally, we see broader acceptance of AEC solutions among service providers and Tier 2 data centers. As a group, these customers represent a meaningful and growing revenue opportunity. Last quarter, we discussed the introduction of our P3 pluggable patch panel solution, developed in collaboration with a lead service provider. Over the past quarter, we've been encouraged by customer feedback that the combination of the P3 and the AECs can help overcome multiple networking challenges related to power and thermal distribution, lane speed disparities and the trade-offs of operational efficiency, latency and power. We expect to generate meaningful future revenue as the P3 enables AECs to be easily utilized in a more broad set of opportunities, thereby expanding our addressable market. In summary, we were very pleased with our AEC progress in Q3. Due to our expanding customer base and focus on innovation, we expect further progress in Q4, fiscal '25 and beyond.
I'll first review our Q3 results and then discus our outlook for Q4 of fiscal '24. In Q3, we reported revenue of $53.1 million, up 20% sequentially and down 2% year-over-year. Our IP business generated $1.3 million of revenue in Q3, down 83% sequentially and down 90% year-over-year. IP remains a strategic part of our business, but as a reminder, our IP results may vary from quarter-to-quarter, driven largely by specific deliverables to pre-existing or new contracts. While the mix of IP and product revenue will vary in any given quarter over time, our revenue mix in Q3 was 2% IP, below our long-term expectation for IP, which is 10%–15% of revenue. We expect IP as a percentage of revenue to be within our long-term expectations for fiscal '24 and near the high end of the range. Our product business generated $51.8 million of revenue in Q3, up 41% sequentially and up 24% year-over-year. Our top three end customers were each greater than 10% of our revenue in Q3. Our team delivered Q3 non-GAAP gross margin of 62.2%, above the high end of our guidance range and up 235 basis points sequentially. Our IP non-GAAP gross margin generally hovers near 100% and was 92.7% in Q3. Our product non-GAAP gross margin was 61.5% in the quarter, up 877 basis points sequentially due to a large increase in product NRE revenue and up 1,420 basis points year-over-year. Total non-GAAP operating expenses in the third quarter were $30.6 million, above the high end of our guidance range, up 13% sequentially and up 19% year-over-year. Our OpEx increase was a result of a 24% year-over-year increase in R&D as we continue to invest in the resources to deliver innovative solutions. Our SG&A was up 12% year-over-year. Our non-GAAP operating income was $2.4 million in Q3 compared to a non-GAAP operating loss of $0.7 million last quarter due to increased top line leverage. Our non-GAAP operating margin was 4.6% in the quarter compared to a non-GAAP operating margin of negative 1.7% last quarter, a sequential increase of 622 basis points. We reported non-GAAP net income of $6.3 million in Q3 compared to non-GAAP net income of $1.2 million last quarter. Cash flow used in operations in the third quarter was $1 million. CapEx was $5.1 million in the quarter, driven by R&D equipment and production mask spending. Free cash flow was negative $6.1 million, an increase of $3.1 million year-over-year. We ended the quarter with cash and equivalents of $409.1 million, an increase of $168.6 million from the second quarter. This increase in cash came from the net proceeds of our successful follow-on offering of shares completed in December of 2023. We remain well capitalized to continue investing in our growth opportunities while maintaining a substantial cash buffer. Our accounts receivable balance increased 36.8% sequentially to $44.8 million, while days sales outstanding increased to 77 days, up from 68 days in Q2. Our Q3 ending inventory was $31.5 million, down $4.3 million sequentially. Now turning to our guidance. We currently expect revenue in Q4 of fiscal '24 to be between $59 million and $62 million, up 14% sequentially at the midpoint. We expect Q4 non-GAAP gross margin to be within a range of 64%–66%. We expect Q4 non-GAAP operating expenses to be between $33 million and $35 million. We expect Q4 diluted weighted average share count to be approximately 180 million shares. We are pleased to see fiscal year '24 playing out as expected. The rapid shift to AI workloads has driven new and broad-based customer engagement, which has continued to enable us to diversify our revenue through fiscal year '24 and beyond. As a result, we look forward to driving operating leverage in the coming quarters. With that, I will open it up for questions.
[Operator Instructions] Our first question comes from Toshi Hari with Goldman Sachs.
Bill, I wanted to ask you how you're thinking about your business over the next year or so. I thought the way you framed the AEC opportunity and the optical DSP opportunity, both in the near term and the medium term was pretty consistent with how you framed it three months ago. I'm curious, just given the significant increase in AI spending infrastructure broadly. Have you seen any of your customer projects either get pulled in or the sizing increase over the past 90 days or the way you're thinking about the additional customer and AEC, the engagements under the DSP side, are they pretty consistent with 90 days ago?
Generally speaking, it's pretty consistent with what we've communicated in the past. We're really in the early innings of many of the opportunities, and we're going to see some variability as each one of these things ramp. I can say that we feel great about the fact that we've got more irons in the fire than ever before. I think from a revenue profile standpoint, we've been pretty consistent in saying that we see an inflection point in the second half of fiscal '25. Really no change.
Even if you exclude the increase in NRE revenue, your product gross margins, I thought were pretty good on a sequential basis, they improved quite nicely. What were the drivers there? If you can provide a bridge from 62.2% last quarter to, I guess, 65% at the midpoint for this quarter, that would be helpful. Is there anything onetime in nature embedded in the 65% for this quarter?
Because as we look at things, we thought of the hidden story is really that due to increasing scale and product mix. If you exclude NRE from our product gross margin, it was up 328 basis points sequentially, as you say, from 50%–53%. That's in line with our messaging that we've given historically. We think we're well on track for attaining our long-term gross margin expectation, which is, again, 63%–65% within the next two years. This is a key part of that story is increasing product margins. Now when it comes to Q4. The other thing I mentioned or alluded to, IP in Q3 was only 2%. We said in our prepared remarks that we expect for the full year that IP as a percentage of the overall revenue mix will be at the high end of our long-term range of 10%–15%. If you look year-to-date, IP has only been 9% of our overall revenue mix. You could expect Q4 to be very strong in terms of IP deliverables, which, again, we recognize revenue upon those deliverables of IP databases under ASC 606. IP has been quite variable quarter-to-quarter throughout FY '24. It's been highly variable. It's been very illustrative of that pattern. In a sense, you might say the IP portion contributing to the gross margin being 65% at the midpoint. That's a variable part of it. Having said that, the underlying thing to focus on is really product gross margin, exclusive of NRE and the trend there has been very favorable to us due to product mix and increasing scale.
Our next question comes from Karl Ackerman with BNP Paribas.
You tend to not be dictated by the seasonal variations of the market each quarter. Since you happen to sell primarily into front-end traditional server market applications, I'm curious to hear whether you are seeing a cyclical recovery in traditional servers as new data centers are being built to really ameliorate the power limitation of installing GPUs and therefore, you are seeing that happen and ramp now.
I think that the way view our AEC business is really broken out into two parts, front-end networks and back-end networks. The front-end connections that we're making for general compute as well as AI, they look very similar in a sense. I can tell you that the most history that we've got on the front-end network connections is really with our first AEC customer, Microsoft, and if we think back on history, the number of front-end connections associated with AI clusters is fewer. If you're in an overall volume sense because you've got one front-end network connections for maybe six to eight GPUs, whereas there's one connection for every general compute server in comparison. We saw a big reduction in the forecast, which would clearly imply that they made a huge pivot towards AI. I can say that we've always expected a rebalancing. I think based on forecast, I think we see that coming within the next 12 months, probably back-end weighted on the 12-month forecast. I do think that we're seeing some of that movement. We don't really get super specific visibility on that.
I was hoping you could address how large is your chiplet business today? Is that broadening into multiple customers? Or is it still dominated by a single customer at this point?
We've got two customers in production now. One is Tesla and one is Intel and both have been publicly discussed. The expectation that we've had -- we've been working on chiplets for several years, and we've been bullish on the space probably before the industry at large was talking about it. I think that the discussion that we've had about this recently is that it has become a nice portion of our business. If I think about long term, and I think about how the different businesses will break down. This is not going to probably rival the AEC business that we've got or the optical DSP business long term. It will be a smaller component of our business, but still very meaningful. To answer your second part of your question, we've got a handful of customers that are engaging with us for chiplet designs, understand that chipless we design will be generally available to the market even though typically, we have an initial sponsor that moves us in a direction on a given spec.
Our next question comes from Tore Svanberg with Stifel.
Bill, I was hoping you could update us on the AEC market. Perhaps a more form factor. It's very clear that the AEC market is heating up. We're starting to see more competitors in the market, but everybody seems to take a different approach on the form factor, system versus chips. I was hoping you could update us on where you see the market evolving over the next 18–24 months. Again, filing a system is still the way to go?
The key for us is that we see the work required to bring an AEC to production with great performance, high quality, high reliability, it's really the same, whether one company is taking responsibility like we are or if many companies are trying to work together and sharing responsibility, our feeling has always been that the most effective approach to bring these products to market quickly is to take full ownership of all aspects of the hardware and the firmware design, the entire process to qualify and bring a product to production and then ultimately the responsible party for directly servicing and supporting during production.There's no lack of clarity for us and our customers as to who owns the responsibility during any stage. As a result, we've been first to samples several times and first qualification now with several different customers. There are separate approaches, but I think there's always going to be some signal loss as it relates to multiple parties trying to respond into urgent needs within the customer base.We're feeling quite good about the efforts that we've made over the last several years in not only developing customer relationships, but also just developing a core capability. I estimate that we have more than 100 people that are dedicated to our AEC business at a system level. If we want to make a direct apples-to-apples comparison, I think it's important to point out that our core SerDes, which is a key point that I want to emphasize, if we're head-to-head apples-to-apples, we're always going to have the most energy-efficient solutions, just given the advantages that we bring with our SerDes technology.
On PAM4 for DSP. You've had the LPO DSP out now for a few months. I'm just wondering what the early reaction has been from customers, assuming you've been able to discuss this in more detail with your customer base?
I think the reaction has been as expected. We've seen certain customers really become very interested in this approach especially for the 800 gig market, there's quite a broad spectrum in the way that customers are thinking about the market. We can go back to OFC a year ago when this whole concept of linear pluggable optics or an optical module without a DSP was really introduced. Although the call to action during last year is really talking about 1.6T modules. Just the fact that the curve is almost unsustainable from a power standpoint. Looking at options. Of course, there are certain people in the market that are investing resources and trying to play out the LPO architecture in the 800 gig space. Of course, in a sense, the 800-gig horse is already out of the barn. Full DSP is what you're seeing in every 800 gig module that's being built right now. I think there's still a desire to lower power and potentially lower cost. The key there is that our solution, which is really deploying DSP on half of the connection within the module versus what we refer to as a full DSP.We're getting a path to maintain industry standards, signal integrity and basically overcoming all the obstacles that people face by eliminating the DSP. Offering a path to much reduced power and potentially cost. It's played out quite well. We'll actually be demonstrating the two optical module partners at OFC. We'll have first sample units from our partners being shipped into the first interest at hyperscaler in the next fiscal quarter.We're pretty satisfied with the efforts that have been made over the last quarter in collaboration with our module partners. I will say that as we look towards the future, we're offering great solutions that are LRO, which is half DSP and full DSP. As we look at 200 gig per lane or 1.6T optical DSPs, we're going to deliver both full DSP solutions and half DSP or LRO solutions at the same time. The design that we're pursuing in 3-nanometer, we've designed in that flexibility. We're really putting the decision in the hands of the customer. Now note that we've chosen 3 nanometer as our process geometry based really on power efficiency. We believe we're going to deliver optical DSP that is capable of being integrated within the current connector specs and within the power ceilings that are already out there with the IEEE standard connectors.
Our next question comes from Matt Ramsey with TD Cowen.
Bill, I wanted to follow up on the tenor of the conversation you were just having in your last answer. I guess with OFC coming up, I think my observation has been developed in the networking and optics space. Two or three polarizing religious technical debate over the last 12 months with InfiniBand and Ethernet with linear drive versus DSP whether to do whole system solutions or just chips. What I wanted to ask you is when you're engaging with your customer base and talking about solutions going forward, are you noticing that hyperscale companies are making bets in all areas here, and there's room and area under the curve for companies like yourselves and your competitors that may take different approaches to all succeed because the pie is big enough? Are the customer base making binary decisions along some of those actions that might open or close opportunities for your company at different customers? It seems like things have got very polarized around several different access over the last 12 months, and I'd just be interested in your thoughts.
From our perspective, it's a little bit more clear. I think the churn that you see, the polarizing opinions in the market, those are really observers that are -- or participants trying to create some momentum. The way that we view it is really the decision-making happens in the conversations that are ongoing directly with hyperscalers. We view each one of them as a separate market. Each one has a different architectural strategy. Each one has a different time line that they're working on. Ultimately, they all have an end objective that they're trying to strive for. By no means does this market have a common cadence or a common road map between the hyperscalers. Our big focus is hey, let's have these conversations and identify the right connectivity solutions for each one of our hyperscale customers. We're agnostic in a sense that if a hyperscale a customer wants to go in a certain direction, we're going to try to find a way to support that direction and deliver the value that they're looking for. From our perspective, yes, we see all of the hyperscalers spending efforts on different ends of the spectrums that you just described. For us, it's all part of our overall pursuit of these customers.
Dan, I just wanted to have you reflect a little bit on the last year or so and compare where things were? I mean, obviously, the AI pivot was way more severe and in magnitude and timing than I think anybody in the market really anticipated 12 months ago. I just want to think your customer base has broadened. I wonder if you could spend a few minutes talking about how you're viewing quarterly revenue visibility. It's been my observation that there are some components in data center builds that are really tight and large hyperscalers want to make sure that some components like you guys sell into certain markets are not the bottleneck of deploying systems. I don't know the confidence in the quarter-to-quarter revenue visibility given the large nature and concentration of some of your customers and maybe contrast that to where we were 12 months ago.
Some of the key observations there are -- if we go back just about a year ago today to when we had that Microsoft reset, I think we've done a very good job at executing to what we had laid out at that point in time. What's really driven that is our product diversification throughout the year and also customer diversification, as you mentioned. As we sit here today and we're preparing to close out our fiscal year '24, our revenue was up just modestly versus our FY '23. Again, if you were to exclude Microsoft from the equation, it's up quite dramatically. Driven by those underlying currents driven by AI spend. A lot of these programs are AI related that we've been given uplift for throughout the year. We're much more comfortable or I'm much more comfortable right now where we stand versus, say, a year ago in terms of our visibility. That's because of a diversified customer base, diversified product base. We're in quite a different position than we had been as we spoke a year ago.
Our next question comes from Vijay Rakesh with Mizuho.
I think you mentioned working with five hyperscalers now. As you go to the back half, do you see the AEC ramp broadening out to others like Google or Open AI. How do you see any of the other partner ramps starting in the back half.
I think we've been pretty consistent in talking about new ramps with two of our existing customers. Both on front-end networks and back-end networks for future deployments. Both of these companies demonstrated last quarter at their respective conferences. I think that as we look at our fiscal '25 and we talk about this inflection point in the second half, it's really related to these first two customers. I would say that as we look at the balance of the U.S. hyperscalers for maybe what will be our third production customer, we have concluded a call qualification with that customer for that first program that we've been working on. There is a possibility that we'll see some contribution in the second half as well in '25, although we don't have good visibility on when they're exactly going to deploy. I don't have an idea as to timing and volume yet, but we have past qualification. The other two hyperscalers, five total, we're in an earlier development stage. When we think about earlier development, we're talking about architecture discussions, spec definition, product development, early samples. Those are the stages that we're working through now. I would say that as it relates to revenue layering in to the first three, I would point to fiscal year '26 and that would be for the initial ramps. Hopefully, that gives some color.
It's very intresting sing to see how fast you came to market with the resist DSP and now it seems like you're ramping with the hyperscale in the next quarter. Can you talk to how big that business could be? Do you see the optical DSP ramping a whole lot faster Obviously, it's much better margins as well.
I'm not sure if that came across correctly. Optical DSP design has a long development cycle associated with it in a sense that the first step you've got to do is build a module with a module partner. Ultimately, there's a qualification cycle that they've got to go through before they can even be considered for qualification at a hyperscaler.I look at the time line from start T0 to being in production as being a little bit longer, maybe significantly longer than some of the AEC experience that we've had. As we're looking at the near-term for optical DSP business, we're in the early innings of ramping our first U.S. hyperscale end customer through our module partner. We're seeing continued maybe come back in spending with the customers that we've got already qualified in China, there is a second U.S. hyperscaler that we're in the stages of competing for a next-generation ramp that could impact our fiscal '25. Then beyond that, I would say that there are other opportunities that we're pursuing, but from a significant volume ramp perspective, probably more in fiscal '26.
Our next question comes from Thomas O'Malley with Barclays.
In the quarter, you saw a big engineering services contribution. I think, Dan, you mentioned product NRE. If you look historically, those have come in on a quarterly basis like $2 million –$3 million. Could you maybe give us some color why this one was so big? Is it one customer, multiple customers? Is that volume related? Maybe a little color if it's AEC or DSP, anything that would give us a little color just given how significant was in the January quarter?
As you correctly identify, historically, our product NREs range for maybe, I'd call it, $2 million–$4 million per quarter. Q3 was higher than historically. Again, this really speaks to the value of the innovative solutions that we're bringing to market. This particular case, it was largely for a single customer in the development, as Bill mentioned, for our next-generation chip lit. Now the other piece of that, of course, as we're going down the process node geometry. This particular development was for 5-nanometer. The engineering resources required for that was more engineering heavy historically than what we've had in 7 nanometer and prior geometries. Those are the things that are impacting that. Now having said all that, I don't think this sets a new bar or level of NRE engineering services. I would expect it to trend back to the mean of our historical averages.
You said IP license for the full year is going to be at the high end of your range. You're saying product engineering and services goes back to the historical levels, which is like $2 million–$3 million. You've spoken pretty positively on the AEC business. When you got to take all those pieces together in April, other products seems to be down a little bit. Could you give us a little color on where you may be seeing some sequential weakness?
I don't know if I would characterize anything as sequential weakness. If you put all those pieces together, if you exclude NRE from product, it's flat to up quarter-over-quarter, and that's based off of a very strong sequential performance in Q3. As programs ramp, they become large. There might be pauses in certain things, just speaking generically around programs of this nature. No seasonal weakness, I wouldn't say.
[Operator Instructions] Our next question comes from Vivek Arya with Bank of America Securities.
The quarterly product revenues are running in the $40 million, $45-ish million range. I was hoping you could give us a sense of how much of that is AEC run rating right now? How do you think about that quarterly trend over the next few quarters?
We don't break it out specifically by product line, but what you'll infer when we file our Q and you look through that is one of our 10% customers is going through the initial stages of a production ramp of AEC. This is our second hyperscaler that we've talked a lot about. AEC drove a lot of the sequential growth from Q2 to Q3. In the upcoming quarters, again, as these programs start to ramp, it's a little hard to predict linearity of it, there may be some variability quarter-to-quarter until we hit this second half of fiscal '25 inflection point. Hopefully, that gives you some color.
My bigger question is how should we get the confidence that AEC will be a preferred choice by your customers? Because they deployed a lot of GPUs last year. They are already deploying a lot of GPUs. They are deploying a lot of optical transceivers. So far, they have not deployed as much AEC. What will change in the second half for them to start deploying more AEC and consider that a more mainstream choice as opposed to a one-off or niche choice in a handful of deployments. I think that's really the key question.
I think the question is maybe related to architectures for the back-end networks and whether there are top of rack switches in the AI plant racks. I think that if we look at the work that we're doing with our customers, I think the long-term preference is to, first of all, deploy Ethernet networks. I think that's pretty commonly understood amongst the five U.S. hyperscalers. I think that each one will have a different time line on when they go into high volume with Ethernet networks. Some of them are deploying with InfiniBand today. I think from the standpoint of designing with top-of-rack switching versus pulling all of the connections from the AI appliance using optical connections to the leaf-spine dedicated switching network for the back end. I think it's around operational efficiency, and it's around how do you build and deploy these? If you were to look at not implementing top-of-rack switching, you're talking about not being able to build known good racks and then those known good racks being forklift installed in the data center. You would be looking at having to assemble these clusters almost on site, there's a real trade-off on operational efficiency. Nonetheless, all of our discussions that we're having are directly associated with feedback from customers. I don't think anything that we're talking about is a real one-off. As I think customers come to execute to their long-term plans -- this will come more into view.
Our next question comes from Quinn Bolton with Needham.
Following up on the next question a little bit. Last quarter, you had four top customers each represented a different product line. This quarter, you talked about three 10%-plus customers. I was wondering if you can give us those 10% customers, what set of revenues they were? Were they all representing different product lines? Are you starting to see AEC coming to the top and representing the majority of revenue from at least two-year, three top customer?
As I mentioned in my prepared remarks, we had three 10% end customers in the quarter. Again, it's important to realize that when our Q is filed in the next day or so, you'll see our end customer disclosure in our MD&A. To give you color on what they were, the largest customer was our first AEC hyperscale customer that we've talked about, they came in at 28%. That's followed by a lead chiplet customer, much of which was NRE driven at 23%. Then the final of the top three, which you'll be able to figure out who that is based on the warrant, was our second AEC hyperscaler customer at 19%. Two of the top three then would be AEC driven, the other being chiplet driven. There's going to be variability in that as we go through. Last quarter was a bit interesting and unique in that the top four customers all represent four different product lines.
Bill, you talked a lot about the U.S. hyperscalers wanting to deploy Ethernet-based back-end network instead of relying on InfiniBand. Obviously, to date, most of those GPU networks that have been deployed have been Nvidia-based and therefore, InfiniBand certainly logical choice. You've got a second GPU customer AMD that's beginning to ramp in meaningful volumes. I'm just wondering, can you talk about from a high level, what are you seeing across your hyperscale customer base for the back-end connections in AMD, GPU networks? Are they deploying Ethernet? Are they deploying InfiniBand? Is there some other fabric because it certainly seems like there'd be a pretty good opportunity for AECs with that AMD and MI300 deployments going forward.
I think that generally speaking, you're right. I would say that the visibility we get into the GPUs that are actually being used, it's really somewhat masked by the NIC. We see a certain number of lanes of Ethernet at a certain speed. Really, our solutions are the same. I'm sure we're connecting to Nvidia GPUs and internally developed GPUs as well. I don't think that we would be the best group to comment on the decision-making around Ethernet versus InfiniBand. I can just tell you, based on the discussions that we've had, it seems very clear that the U.S. hyperscalers have a strong intention to deploy Ethernet. You can go each hyperscaler and do a bottoms up on it. I think you'll see that Microsoft is clearly out there as the lead hyperscaler that is deploying with InfiniBand. I think after that, the list becomes pretty short. I think everything that we're seeing is Ethernet is going to be preferred.
I'm showing no further questions at this time. I would now like to turn it back to Bill Brennan, CEO, for closing remarks.
Thank you very much for the questions. We really appreciate the participation, and we look forward to following up on the callbacks. Much appreciate it. Thank you.
This concludes today's conference call. Thank you for participating. You may now disconnect.