Note: This write-up includes notes on the earnings report that will come after market close plus the Post-Earnings write-up from last quarter — which was posted on the forum. the Post-Earnings write-up from last quarter — which was posted on the forum.
We expect Marvell to be a late bloomer in AI acceleration compared to Nvidia and AMD. There’s been discussion on the forum lately about Marvell’s AI thesis, and admittedly, it carries some speculation at this point in time. I do think ASICs will take market share from GPUs once the market is more mature, yet who specifically serves the market requires speculation. We think it’s worth the gamble to have a placeholder on Marvell right now. Given the runup in AI, we will use technical analysis to risk manage, as needed.
Revenue and EPS:
Marvell is expected to report revenue of $1.33 billion, representing a decline of (-12.3%). This should mark the bottom with revenue expected to decline (-9.2%) next quarter for $1.4B in revenue. From there, Marvell is expected to return to positive revenue growth with 2024 showing a meaningful rebound by H2.
One thing to note is the company saw some fairly large downward revisions despite the stock rallying in price this year. At the end of last year, the growth was expected to be 15% for FY 2024 ending in January and it’s now at (-7%). This is true for a few semiconductor companies that have exposure to consumers. FY 2025 estimates are improved 200 to 300 basis points to 18% growth.
Marvell is not GAAP profitable, and is expected to report (-$0.16) EPS this quarter. Adjusted EPS for Q2 is expected to be $0.32. The company is expected to return to EPS growth when it returns to revenue growth in the January quarter, with considerable growth on the bottom line beginning in the April quarter.
Per our last write-up: “In summary, as we move forward through this fiscal year, we project our revenue to continue to grow sequentially, gross margin to improve significantly in the fourth quarter, and operating expenses to continue to step down. Our DSO and inventory have already shown signs of improvement. As a result, we are looking forward to driving tremendous operating leverage and significant improvement in cash flow generation over this year, while setting up a strong platform for growth next year.”
Margins:
Numbers stated are from management’s guide for the current quarter unless otherwise indicated.
Overall, we are hoping Marvell is at a bottom this quarter with margins that are not ideal, but are expected to improve over the next few quarters.
- The guide on GAAP gross margin is lower than usual at 45.5% at the midpoint. This compares to a GM of 51% in the year ago quarter. This will lead to GAAP gross profits also being down at $605M compared to $786M in the year ago quarter.
- Adjusted gross margin of 60.5% guided this quarter is 450 basis points lower than year ago quarter.
- GAAP Operating margin of (-6.6%) compares to OPM of +2.60% in the year ago quarter. Note that even the year ago quarter is still quite low compared to semi peers.
- Adjusted operating margin of 26.3% compares to Adjusted OM of 36.5% in the year ago quarter.
- Net margin last quarter was (-13%) for a net loss of ($170 million).
The reason for the lower GAAP margins is due to the amortization of acquired intangible assets. (Marvell acquired Inphi and Innovium in 2021). In the recent quarter amortization of acquired intangible assets that was included in the cost of goods sold was $183.7 million (13.9% of revenue). This led to a 14% difference in the GAAP gross margin and non-GAAP gross margin.
The management expects challenging margins in the near term due to the product mix.
The company’s CEO Matt Murphy mentioned in the Q4 2023 earnings call, “At the same time, we are forecasting very strong sequential growth in revenue from 5G and a number of custom ASICs, but these have gross margins well below Marvell's corporate average. As a result, we expect a challenging gross margin outlook for the next few quarters.” strong sequential growth in revenue from 5G and a number of custom ASICs, but these have gross margins well below Marvell's corporate average. As a result, we expect a challenging gross margin outlook for the next few quarters.” He further said that once the inventory correction situation improves the margins will improve. “However, we are confident that once we emerge from the inventory digestion phase into a more normalized environment, we will be well positioned for our gross margins to recover.”we are confident that once we emerge from the inventory digestion phase into a more normalized environment, we will be well positioned for our gross margins to recover.”
In a further update in the recent Q1 earnings call he said that inventory corrections will be resolved by the end of the year. “We anticipate that inventory corrections will be mostly behind us by the end of this year, and we are excited about the resumption of revenue growth driven by Marvell's specific product ramps. We have been laser focused on a number of cost improvement actions to improve gross margin. As a result, we have confidence in our forecast for our non-GAAP gross margin to return to at least the low-end of our target range in the fourth quarter of this fiscal year.”We anticipate that inventory corrections will be mostly behind us by the end of this year, and we are excited about the resumption of revenue growth driven by Marvell's specific product ramps. We have been laser focused on a number of cost improvement actions to improve gross margin. As a result, we have confidence in our forecast for our non-GAAP gross margin to return to at least the low-end of our target range in the fourth quarter of this fiscal year.”
The company’s CFO, Willem Meintjes said, “We have also put in place multiple cost reduction efforts to improve gross margin. Internally, we are optimizing headcount and further streamlining operations. Externally, we continue to partner with our strategic suppliers to drive more efficiency in the supply chain. We are confident that as a result of mix improvement and our cost reduction efforts, our non-GAAP gross margin will start to improve.”
Higher R&D and SG&A:
Marvell has higher R&D expenses and SG&A expenses compared to some of its peers like Broadcom, Microchip, and Analog Devices.
Marvell had R&D expenses as a percentage of revenue of 30.14% for the year ending Jan 2023 compared to:
- 14.81% for Broadcom for the year ending Oct 2022
- 13.25% for Microchip for the year ending Mar 2023
- 14.15% for Analog Devices for the year ending Oct 2022.
Marvell had SG&A expenses as a percentage of revenue of 14.25% for the year ending Jan 2023 compared to:
- 4.16% for Broadcom for the year ending Oct 2022
- 9.45% for Microchip for the year ending Mar 2023
- 10.54% for Analog Devices for the year ending Oct 2022
Marvell also has slightly higher stock-based compensation than its peers at 9.33% compared to:
- 4.62% for Broadcom for the year ending Oct 2022
- 2.02% for Microchip for the year ending Mar 2023
- 2.69% for Analog Devices for the year ending Oct 2022
Cash Flow:
- Last quarter, operating cash flow of $208.4 million was up on a year-over basis but down quite a bit sequentially. Last year, OCF was $195 million and was $351 million in the previous quarter.
- Last quarter, free cash flow of $105.8M compared to $156M in the year ago quarter.
- Marvell is also different from its peers in that the company is highly leveraged with $1B in cash and $4.7B in debt. This should garner a lower valuation, and it’s also likely Marvell sees more volatile price action depending on the FED’s actions.
- Marvell has to repay $1.5B in debt over the next year and has $1B in cash. Of this, $500M was just repaid and Marvell is expecting to resume buybacks in Q3. This is a line item we need an update on in the upcoming earnings call.
The operating cash flow included $40 million for a long-term capacity payment. Management expects minimal additional payments for FY2024 and beyond. Capex was also higher in the recent quarter as it was $102.6 million compared to $56 million in the January quarter and $38.50 million in the same quarter last year.
Key Metrics:
The following numbers are for the previous quarter YoY, ending in April:
- Data center was down (-32%) to $436M – The sequential decline was due to storage and the management expects the storage business to grow sequentially in the next quarter and grow in the second half of the year. The data center revenue is expected to be flat sequentially in the next quarter (more below).
- Carrier infrastructure was up 15% to $290M — helped by strong demand for wireless products due to 5G adoption. The management expects revenue to decline mid-single digits sequentially due to the ongoing inventory digestion in the wired end market.
- Enterprise networking was up 27% to $365M — Management expects revenue to decline more than 10% sequentially in the next quarter due to the ongoing inventory corrections that also negatively impacted the recent quarter.
- Consumer was down (-20%) for $142M — Management expects revenue to grow mid-30% range sequentially due to the strong seasonal growth expected for custom SSD controllers.
- Automotive/Industrial – (Flat) for $89M — Due to the weakness in the industrial business. The management expects the automotive and industrial end market to grow sequentially in the low teens in the next quarter.
Storage:
Marvell’s data center segment has exposure to the deep trough in storage. We recently covered Lam Research’s exposure to NAND. This is not unique to Marvell, and many quality companies are affected. We ultimately think it’ll lead to a buying opportunity for FY2024.
Here was a question on the call about where storage might stabilize in terms of revenue:
Toshiya Hari:
Good point. Maybe one last question on data center, and then I’ll certainly open it up to the crowd. On your storage business and you gave great context in your prior comments. But I think pre-pandemic you were doing $200 millionish in storage revenue within data center. I think at the peak, you were closer to $300 million?
Ashish Saran:
Higher – it’s a lot higher.
Toshiya Hari:
And now you’re sub $100 million. And I think you talked about ultimately getting back to $200 millionish. Is that sort of the steady state if there is such a thing in this business?
Willem Meintjes:
Yes. I think – when you look at it – yes, so exiting this year, we don’t think we get all the way back to $200 million, maybe three quarters of the way.
According to management, they are returning to growth soon. Per the last earnings call:
“As expected, storage was responsible for the majority of the overall sequential decline in our data center revenue in the first quarter, although we are forecasting sequential growth to start in the second quarter and our data center storage business continue to grow in the second half.
Looking ahead to the second quarter for our overall data center end market, we expect cloud revenue to grow over 10% sequentially. However, we are expecting the enterprise on-premise portion of our data center end market to decline an offset growth from cloud. As a result, we expect revenue from our overall data center end market to be flat sequentially in the second quarter.”
AI-related Revenue:
Below are excerpts from the forum post following last quarter’s earnings results. excerpts from the forum post following last quarter’s earnings results.
If you listen closely to Marvell’s call, you’ll see there was a slight pivot to how management pitched AI and the data center. This pivot is important to understand.
In the past, the company’s data center segment was benefiting from the Inphi acquisition. We’ve written many times about this acquisition, and in fact, we owned both Inphi and Marvell when it was acquired by Marvell.
The COLORZ silicon photonics technology from Inphi allows data centers located in the same region to function like a mega data center. This helped Marvell become a critical supplier for data center interconnects. At one point, following the Inphi acquisition, the COLORZ 400-gig ZR datacenter interconnect products were driving data center revenue growth of 100%.
On the call, management is referring to networking and connectivity products when they stated:
“We created the industry's first pluggable module for DCI, and we are now providing 400 gigabits per second in our latest DCI product line. We already seen that AI cloud data centers are driving a significant increase in demand for 400 ZR solution. Another demand driver for our DCI products is that the next generation AI implementations are planning on clustering accelerators across different sites.”
Networking and connectivity are not to be underestimated in terms of growing importance, especially as AI will drive a need for more bandwidth. However, up until FY2023 ending in January, the revenue was $200 million. This revenue is expected to double in the current fiscal year 2024 to $400 million. Compare this to Marvell’s overall revenue of $5.52 billion. That’s only 7% of revenue from AI – a far cry from Nvidia’s $4 billion data center beat & raise for one quarter for a data center segment that will be $8 billion per quarter, purely from GPUs and AI acceleration.
Maybe Marvell is an AI bubble stock? I don’t think so … instead, Marvell is a serious contender.
Here’s why.
What Marvell later communicated is quite important, which is that the company plans to compete against GPUs with custom silicon. To be clear, this is a leap from relying on network connectivity to now relying more heavily on compute to drive AI revenue. This will be done through ASICs, which stands for application-specific integrated circuit (ASIC). As soon as Marvell started talking compute, the revenue for networking and connectivity is no longer the driver for AI revenue.
Here is what management said:
“Today, we see cloud customers enhancing their AI offerings by building custom accelerators of their own designed to address their specific needs. This is a core part of Marvell's cloud optimized silicon strategy, and we now see a much larger and faster growing opportunity for custom compute and AI infrastructure. When we previously discussed our cloud optimized silicon opportunities and revenue ramp expectations at our Investor Day in October 2021, we projected revenue from the first set of design wins to grow to $800 million annually once all the programs were in production.”
Management indicated $800M is a starting point when the CEO stated: “The continuing increase in demand from AI, we see annual revenue from those same set of cloud optimized design wins, well exceeding the prior 800 million projection as these programs ramp over time. In fact, we have a number of custom silicon products tied to AI expected to ramp into volume production next year. As an example for one of these programs, initial samples are already up and running at our customer's lab and qualification is proceeding well. And in other case, we expect to take out this quarter and deliver first silicon before the end of the calendar year.”
In terms of trying to quantify it further, management also stated: “In other words, we are forecasting an AI revenue growth CAGR over 100% over the fiscal 2023 to 2025 time frame.”
Later, it was stated: “And then what I updated on the call was that that peak revenue was actually going to exceed the 800 million, just given that the total lifetime volume now of those same design wins has actually increased pretty significantly. And then the percentage of those has moved meaningfully towards AI.”
What are ASICs?
I’ve written about ASICs a few times in the past, including in this article about Google. Google is well known for TPUs, which is customized silicon, known as ASICs. The two companies that supply leading edge ASICs are Broadcom and Marvell. The reason it makes sense to outsource custom chips is because a Big Tech company trying to tackle too many things can often work against them, especially with hardware of complexity.
“Just as Google was one of the first to need automated orchestration for containerization of cloud-native apps, the company was also one of the first to require low-power machine learning workloads. The compute intensive workloads were running on Nvidia’s GPUs for both training and inferencing until Google made their own processing unit called Tensorflow (TPUs) to perform the workload at a lower cost and higher performance.
Performance between TPUs and GPUs is often debated depending on the current release (A100 versus fourth-generation TPUs is the current battle). However, the TPU does have an undisputed better performance per watt for power-constrained applications. Notably, some of this comes with the territory of being an ASIC, which is designed to do one specific application very well whereas GPUs can be programmed as a more general-purpose accelerator. In this case, the benchmarks where TPUs compete are object detection, image classification, natural language processing and machine translation – all areas where Google’s product portfolio of Search, YouTube, AI assistants, and Google Maps, for example, excels.
Notably, TPUs are used internally at Google to help drive down the costs and capex of its own AI and ML portfolio and they are also available to users of Google’s AI cloud services. For example, eBay adopted TPUs to build a machine learning solution that could recognize millions of product images.”
In a previous Marvell analysis, I stated the following about ASICs:
“ASICs:
In May, Marvell announced plans to acquire Avera Semiconductor for $650 million in cash plus $90 million if the business does well within the next 15 months. The deal is expected to close at the end of fiscal year 2020. Avera is a player in the ASIC market, which will help diversify Marvell as 5G has begun to favor ASICs over FPGAs due to costs and power consumption. ASICs, which stands for application-specific integrated circuit, are customized to perform one very specific function repeatedly rather than general-purpose chips – hence the “application specific.” Broadcom could potentially be challenged by this acquisition. Avera will add $300 million per year to Marvell’s top line.
In contrast to ASICs, the traditional FPGA chips are high in cost and power consumption, according to critics. Marvell is attempting to offer end-to-end network infrastructure with baseband DSPs, Arm multi-core SoCs (system on chips), purpose-built hardware accelerators, Ethernet connectivity engines and system-level security solutions. Although Marvell aims to offer specific-use ASICs and semi-custom ASICs, the 5G platform that Marvell offers will be adaptable for many use cases to expand on any ASIC limitations.
The primary SoC competitor is Broadcom. NXP Semiconductors and Qualcomm also compete with Marvell. Xilinx is a competitor on FPGAs.”
AWS Potential Customer on ASICs:
There was a report out of Taiwan that Marvell and Amazon may be working together on custom silicon for AWS. Since then, the report has been questioned but this is certainly the size and type of customer that would contract Marvell for a custom design.
Per the report:
“Marvell Technology (NASDAQ:MRVL) shares trade nearly 5% higher in pre-open Wednesday after Taiwan-based media outlet Liberty Times reported the chipmaker won Amazon’s (NASDAQ:AMZN) second-gen ASIC chip design. The production will begin in the second half of 2023, according to the report.
Some had speculated MSFT would be a leading MRVL ASIC customer. AMZN would make sense as they have some of the biggest custom in-house silicon development (Gravitron). And NVDA CEO has oddly NEVER INCLUDED AMZN as a direct GPU / AI partner in recent presentations, only MSFT, ORCL, GOOG,” they added.”
Conclusion:
The timing for custom silicon to show up in Marvell’s financials is 9 to 12 months out, and reaching full potential in two years. Many of our Members have been patiently waiting for our other semiconductor stocks to play out, what’s another two years? 😀
Maybe then, we will be needing to reallocate Nvidia anyways away from being such a large position! This stuff takes a LONG time to play out. I'm writing this Marvell post right now about custom silicon for a 24-month outlook, minimum.
Here is what management said:
“I mean – and those are the projects now that are coming to fruition. But I mean, just to be very clear, since that time, we have layered in a significant number of incremental new design wins. And those are – many of those are now underway, probably not as much revenue from those next year, but then those would be, sort of the year after. And I would just also add that our design win funnel, and this is data I was just reviewing at the company level, who's just expanded dramatically with the biggest portion of that coming from our cloud design win funnel.”
I didn’t address CXL in this post but this is an added bonus should Marvell be a frontrunner with CXL Memory. You can read about this new product line here.
Additional Reading: