Marvell Looking To Leverage Expertise In Electro-Optics To Stake Claim In AI Boom (MRVL)

JHVEPhoto This “all AI, all the time” world for chip companies has certainly been a boon for Marvell Technology, Inc. (NASDAQ:MRVL), not only as demand for electro-optics (used for high-speed interconnect) and ramping custom silicon has helped mitigate major cyclical declines in other parts of the business, but as investors have piled into the hot theme of the day. Marvell shares are up about 60% since my last update, lagging names like Broadcom Inc. (AVGO) and NVIDIA Corporation (NVDA) as well as the broader semiconductor space (as measured by the SOX index). While it’s difficult to argue that Marvell is undervalued on anything other than a relative basis (relative, say, to how growth chip stocks have traded in the past), the reality is that the market wants AI-driven plays and Marvell’s strong share in electro-optics and growing custom silicon business are both valuable. With the potential to grow revenue at a 15%-20% rate over the next five years (and possibly above 20%) and ramp margins to new highs, this is a name worth some consideration, but is really only well-suited to more aggressive investors who are less sensitive to valuation. AI Will Likely Be The Engine For A Return To Sequential Growth Marvell’s recent results have been a tale of two businesses – the AI-driven Data Center business has logged sequential growth for four straight quarters heading into this upcoming fiscal Q2 report, while the rest of the business has seen a sharp deceleration. Looking back at FQ1, revenue declined 12% yoy and almost 19% overall, but the non-AI businesses were down 61% yoy and 48% QoQ. Between ongoing strength in electro-optics (Marvell enjoys exceptional market share with its PAM4 DSP product family) and the ramp of custom AI silicon at a major customer (presumably Amazon.com, Inc. (AMZN), though I don’t believe that’s ever been formally confirmed), I believe Data Center growth could approach 90% yoy in the second quarter (around 7% QoQ). That, in turn, should support overall revenue a little above the current sell-side estimate, but revenue is still likely to be down about 7% year over year with ongoing weakness in Carrier, Enterprise Networking, Auto/Industrial, and Consumer. As far as sequential performances go, I expect weak recoveries off the bottom for most of those non-AI units, though I expect a bigger rebound in Consumer after a terrible FQ1 result (down 71% yoy and 70% QoQ). Marvell has been trending at or near sell-side expectations for gross margin for several quarters, and while I do think better electro-optics could offer a little upside, I’m not expecting major surprises on the gross margin side. I’ll discuss the AI/DC business in a moment, but I want to also look at the other segments of Marvell’s business. Channel destocking has been an ongoing challenge in Enterprise Networking, and I’m cautiously optimistic that that process has largely played out, helped in part by Marvell undershipping relative to apparent demand. I’m not that bullish on Carrier; while next-gen DPUs should help grow share, carriers just aren’t spending right now (as seen in the reports from companies like Telefonaktiebolaget LM Ericsson (publ) (ERIC) and Nokia Oyj (NOK)) and an end to destocking is about as good as it’s getting. In the Auto/Industrial business, I do think there is some longer-term leverage to Ethernet content growth, but auto unit volumes just aren’t that strong right now and likely won’t pick up until 2025. Electro-Optics And Custom Silicon Offer A One-Two Punch For AI Growth While Marvell’s leverage to AI growth has only been really appreciated by the Street relatively recently, the company has been sowing the seeds for this growth for some time. The acquisition of Inphi brought in a top-tier electro-optics portfolio and the company has continued to execute and iterate on that portfolio ever since, likely holding more than 70% share of the PAM 4 DSP market despite competing with Broadcom (and MaxLinear, Inc. (MXL) to a much lesser extent). Marvell’s history with custom ASICs goes back much further, though I would imagine some of these efforts got a boost with the acquisition of Cavium years ago. Between electro-optics and custom silicon (which are likely pretty close to 50/50 in the mix), Marvell actually has one of the highest AI exposures of major chip companies behind NVDA – approaching 30% this year and heading higher next year. With its electro-optics business, Marvell boasts an AI customer roster including Nvidia, Alphabet Inc. (GOOG) (GOOGL), Amazon, Meta Platforms, Inc. (META), and Microsoft Corporation (MSFT) and the company has enjoyed a 100% share with Nvidia’s DGX H100. There are reports of Broadcom becoming more significant with the Blackwell product cycle, and Nvidia is reportedly working on its own 1.6Tb PAM4 module, but multiple rivals have found it difficult to beat Marvell in this space. It’s worth taking these reports seriously if only because electro-optics is a major opportunity. While custom silicon is likely to be a bigger grower for Marvell (coming off a smaller base), interconnect is still an attractive growth market as speed demands continue to ramp, and the attachment rate of optical transceivers to GPU could grow from 1.5-2.0 to 4-8 over time, creating a lot more opportunities for Marvell. Looking at custom silicon, Marvell has started to make meaningful inroads into this market, breaking up what had been a near-monopoly for Broadcom for several years. Nvidia holds at least 75% overall share, but Broadcom has leveraged opportunities in custom ASIC with customers like Google and Meta to build what could be a $9B business this year in custom silicon alone. Broadcom has major wins with Google, Meta, and ByteDance, and is rumored to have recently won business from OpenAI for its 2nm/3nm offerings, as well as a fifth unnamed customer. For its part, Marvell has won business with Amazon (already ramping), Google (for ARM CPUs), and Microsoft (ramping in 2026). All of these should be considered “reportedly”, as again there is seldom if ever any specific confirmation of customer relationships. Marvell believes that custom ASIC offerings could become 25% of the market in or by 2028 (versus around 15% today), with as much as $43B in revenue up for grabs. Broadcom will remain a formidable rival, but Marvell’s technology is definitely competitive and there are a lot of customers who won’t work with Broadcom unless they absolutely have no other alternatives. The Outlook Of course, there are valid questions about whether companies like Amazon, Google, and Microsoft will continue to invest such vast sums into AI. Recently, many AI-related stocks sold off on worries about when and how these AI investments would ever pay off, but such concerns are not unusual as growth opportunities ramp up (something similar is happening now with electric vehicles, and it obviously happened with the internet/networking years ago). Given the significant stakes and the risks of being left behind, I don’t think spending is likely to be at serious risk for at least a few years (as Meta has put it, they’d rather overinvest and have excess capacity than risk coming up short). That should be a supportive environment for Marvell. I’m expecting AI/DC to be the major driver for revenue growth over the next three to five years (though recoveries in Enterprise Networking and Auto/Industrial will help), and I’m looking for revenue growth over the next five years of around 16% and longer-term annualized growth of over 12%. I’ve seen projections of 15% to 20% growth “over the long term”, but with Street analysts that could mean as little as three to five years. While the downturn in Marvell’s businesses has hit margins hard, dragging them from the mid-30s percent back into the low-20s percent, the ramp of high-margin electro-optics and custom silicon, as well as improved leverage from a broader recovery, should drive margins back into the low-to-mid-30%’s in FY’26 and toward 40% over the next three to five years. That, in turn, should support a strong ramp in free cash flow and high-20%’s annualized free cash flow growth. Whether or not AI is a bubble, it’s definitely a hot space, and you don’t go into hot spaces looking for bargains. Marvell’s valuation already more than anticipates strong revenue and margin growth, so forget about using traditional benchmarks for valuation. The best I can do is look at past chip growth stories (names like Altera, Cavium, Inphi, Nvidia, Qualcomm, Silicon Labs, and Xilinx) and see what the market has been willing to pay. Doing so, I think a 35x P/E multiple on calendar 24-exit EPS (around $2.50-$2.60) is at least reasonable, and that would support a fair value close to $90. The Bottom Line Given the relative performance, Marvell could have some appeal as a “catch up” trade, particularly given what should be a strong operating environment for the key electro-optics and custom silicon businesses. I’m not suggesting that Marvell is a bargain, nor am I saying that AI expectations aren’t inflated, but if you want to play in this neighborhood, this is a name worth considering.

Buy cryptocurrency

Source link

Refer And Earn Demat Account – Get ₹300 | Referral Program

Open Demat Account In Angel One For FREE