December 11, 2025
Marvell Technology, Inc. (NASDAQ: MRVL) is having one of those “only-in-AI-2015” weeks: the company just posted record data‑center revenue, announced a multibillion‑dollar optical‑interconnect acquisition, and rolled out new connectivity products for hyperscale AI clusters — yet the stock is sliding as investors argue over whether Marvell has lost a critical Amazon AI chip deal.
As of early afternoon on December 11, Marvell shares were trading around $88, down roughly 4–5% on the day and about 31% below their 52‑week high near $127. [1]
Behind that red number is a tug‑of‑war between very strong fundamentals and noisy — and conflicting — headlines about Marvell’s role in the next generation of cloud AI chips.
MRVL stock today: weak price, strong backdrop
Market data from MarketBeat and MarketScreener show Marvell changing hands in the high‑$80s on Thursday, with: [2]
- A 52‑week range of roughly $47 to $127
- A year‑to‑date decline of about 10% and a one‑year decline around 20%, even after a sharp rally since late summer
- A trailing P/E a little above 31x and a modest dividend yield near 0.3%
That puts Marvell in an odd spot for an AI infrastructure name: fundamentally on the upswing, but trading closer to “controversial grower” than “unstoppable AI rocket,” at least for now.
Q3 FY2026: record revenue and an AI‑heavy mix
On December 2, Marvell reported record Q3 FY2026 results: [3]
- Net revenue:$2.075 billion, up 37% year‑on‑year, and slightly above guidance
- GAAP EPS:$2.20, boosted by a gain on a business sale
- Non‑GAAP EPS:$0.76, up ~77% from a year ago and ahead of consensus
- GAAP net income:$1.9 billion, including the gain from selling its automotive Ethernet business
- Operating cash flow: about $582 million in the quarter
The company also closed the sale of its automotive Ethernet business to Infineon for $2.5 billion in cash, generating a pre‑tax gain of roughly $1.8 billion and further strengthening an already solid balance sheet. [4]
Crucially for the AI story, management highlighted that:
- Data center remains the growth engine, with revenue near $1.5 billion, up around high‑30s % year‑on‑year, driven by AI optics, Ethernet switching and custom silicon for hyperscalers. [5]
- The company expects full‑year FY2026 revenue growth to exceed 40%, and guided Q4 for roughly 21% revenue growth and 32% EPS growth at the midpoint. [6]
Marvell’s own longer‑term roadmap is even more aggressive: in commentary highlighted by MarketBeat, the company is targeting data‑center revenue growth of ~25% in FY2027 and ~40% in FY2028, assuming AI infrastructure spending continues to scale. [7]
On top of that, the board has authorized a new $5 billion share repurchase program, equal to roughly 7–8% of shares outstanding, signaling confidence in the company’s trajectory and valuation. [8]
Strategic moves: Celestial AI, AEC “Golden Cable,” and PCIe 6 retimers
Marvell is also leaning hard into the plumbing of AI data centers — where the big money is increasingly in connectivity, not just compute.
Celestial AI: a multibillion‑dollar optical scale‑up bet
Also on December 2, Marvell announced a definitive agreement to acquire Celestial AI, a pioneer in Photonic Fabric™ technology for optical scale‑up interconnects. [9]
Key deal terms and ambitions:
- Upfront consideration: about $3.25 billion (roughly $1 billion in cash plus ~27.2 million MRVL shares)
- Earn‑out: up to an additional $2.25 billion in stock if revenue milestones are hit, for a total potential value near $5.5 billion
- Celestial’s chiplets provide 16 Tbps of optical bandwidth per device — around 10x today’s high‑end 1.6T ports — aimed at the “scale‑up” fabrics that connect hundreds of XPUs within and across racks
- Marvell expects meaningful revenue from Celestial AI starting in the second half of FY2028, with a $500 million annualized run‑rate in Q4 FY2028, doubling to $1 billion by Q4 FY2029, assuming successful customer ramps. [10]
The strategic idea: as AI clusters expand beyond a single rack, copper breaks down on bandwidth, power and reach; Marvell wants to own the transition to all‑optical scale‑up fabrics that tie accelerators together.
“Golden Cable” AEC initiative: wiring short‑reach AI links
On December 9, Marvell launched its “Golden Cable” initiative, aimed at accelerating the active electrical cable (AEC) ecosystem for in‑rack and short‑reach inter‑rack AI connectivity. [11]
The program provides:
- Validated cable architectures and firmware
- Reference designs and calibration data for hyperscaler‑grade AECs
- An open framework so cable makers can customize gauge, bend radius and reach
Industry researcher 650 Group expects the AEC market to grow from about $644 million in 2025 to $1.4 billion by 2029, driven by the shift to 1.6T networking and denser AI racks — and Marvell is positioning its silicon at the heart of that growth. [12]
Alaska P PCIe 6 retimers: inside the AI servers
In a companion announcement on December 9, Marvell said its Alaska P PCIe 6 retimers have been adopted by multiple server vendors and interconnect partners for AI and general‑purpose platforms. [13]
Highlights from the release: [14]
- The retimers help maintain signal integrity at 64 GT/s PCIe 6 speeds, across longer on‑board traces and cables inside accelerator‑rich systems
- They are being deployed both on motherboards and inside AEC/AOC cables, and evaluated for storage systems
- Built on Marvell’s 5 nm PAM4 SerDes, they’re designed to be used in copper, optical and hybrid PCIe interconnects, bridging today’s cabling with future optical PCIe links
Taken together — Celestial AI optics, Golden Cable AECs and Alaska P retimers — Marvell is trying to be the connectivity spine of next‑gen AI data centers, from chip‑to‑chip up to rack‑to‑rack.
The Amazon & Microsoft controversy: did Marvell lose Trainium?
That bullish infrastructure story is currently being overshadowed by drama around Marvell’s custom AI ASIC business.
What the bull camp heard: Trainium 3 is in the bag
In its Q3 call, Marvell’s CEO Matt Murphy told investors that next year’s custom revenue forecast includes a transition to a “next‑generation XPU at a large customer” and that the company already has purchase orders covering its entire forecast for that program for the next fiscal year. [15]
Given Marvell’s long history co‑developing Amazon’s Trainium chips, many analysts and investors interpreted that as effectively confirming a Trainium 3 win, helping fuel an ~60% rally in the stock from late August lows into early December. [16]
The bear camp: Benchmark’s “high conviction” loss thesis
On December 8, Benchmark Equity Research pushed back hard. In a widely cited note, the firm: [17]
- Downgraded MRVL from Buy to Hold and withdrew its price target
- Argued, with a “high degree of conviction,” that Marvell lost both the Trainium 3 and Trainium 4 designs at Amazon Web Services to Taiwanese rival Alchip
- Suggested that recent upbeat guidance is driven mainly by older Trainium 2 volumes, not new design wins
The report helped trigger a ~6% drop in pre‑market trading that day, and contributed to the stock’s pullback from the $90–$100 range.
At the same time, there are rumors that Microsoft has been deepening its engagement with Broadcom for custom AI processors, intensifying worries that Marvell could be squeezed out of key hyperscaler ASIC sockets if those reports prove accurate. [18]
CEO, AWS and Stifel push back
The story didn’t end there. A December 11 InvestorsObserver piece details how Marvell’s CEO and several Wall Street firms have publicly countered the loss narrative: [19]
- In a CNBC interview with Jim Cramer, CEO Matt Murphy flatly stated:
“We didn’t lose any business.” He emphasized that nothing had changed in the days between the earnings call and the critical reports. - An AWS spokesperson, in a statement to Barron’s, said Marvell remains “a partner of AWS, including as a valued silicon supplier.”
- Stifel analyst Tore Svanberg called the negative reports “without merit,” reiterated a Buy rating and maintained a $114 price target, arguing Marvell remains one of the only vendors offering a comprehensive, vertically integrated connectivity stack for modern AI systems.
- Raymond James recently initiated coverage with a Strong Buy and a $121 target, calling Marvell well‑positioned for AI‑driven demand and advanced packaging trends.
- UBS raised its target to $110, citing a more bullish view of Marvell’s optics business and the expectation of additional Microsoft‑related ASIC revenue starting in late 2026.
In short: Benchmark believes Marvell’s next‑gen Amazon ASIC business is gone; Marvell, AWS and several bullish analysts say that’s either wrong or at least not supported by the evidence available to public investors.
For now, the market is pricing in uncertainty rather than accepting either side’s story at face value.
Analyst ratings and price targets: still broadly bullish
Despite the recent volatility, the Street remains mostly positive on MRVL.
Consensus from several sources
- MarketBeat tracks 38 analysts covering the stock, with a “Moderate Buy” consensus and an average price target around $111.56, versus a recent price in the high‑$80s. The range runs from about $66 on the low end to $156 on the high end. [20]
- TipRanks shows an even stronger “Strong Buy” consensus, based on 29 analyst ratings, with 22 Buys and 7 Holds. The average 12‑month target of ~$121 implies low‑20s percent upside from recent levels. [21]
Price‑target revisions after the Q3 beat clustered in the $110–$130 range (including J.P. Morgan’s $130 Overweight), reflecting confidence in the AI data‑center roadmap, even as opinions diverge on exactly how much Amazon revenue is locked in. [22]
GARP and growth metrics
Quantitative analysts see Marvell as a “growth at a reasonable price” (GARP) candidate:
- ChartMill ranks MRVL as a top GARP pick, assigning a 9/10 “Expansion Rating”, citing:
- Revenue growth of about 45% and EPS growth above 80% over the last year
- Five‑year average annual growth around 16% for revenue and 19% for EPS
- Forward‑looking expectations for ~22% annual revenue growth and ~35% annual EPS growth over the next few years. [23]
- On valuation, ChartMill notes a P/E near 35x and forward P/E near 27x: not cheap, but modest relative to many high‑growth semiconductor peers once growth is factored in (a favorable PEG ratio). [24]
Valuation check: fair, slightly rich, or quietly cheap?
Different methods give slightly different answers, but they cluster around “not a bargain, not a bubble.”
Discounted cash flow and P/E context
A December breakdown from Simply Wall St found: [25]
- A DCF‑based fair value of $92.09 per share, versus a then‑current price around $92.47 — essentially fairly valued, ~0.4% above fair value by their model
- A trailing P/E of ~31.7x, below the broader semiconductor industry average (~38x) and well below a high‑growth peer basket (around 74x)
- A proprietary “fair” P/E ratio of 35.4x, suggesting MRVL is modestly undervalued on an earnings‑power basis if growth plays out as expected
MarketBeat’s numbers paint a similar picture: Marvell’s forward P/E around 30x sits slightly below its own three‑year average (~33x) and only a bit above the 28x forward P/E for the broader S&P tech sector. [26]
Put differently: MRVL trades at a growth multiple, but not an absurd one given current revenue and EPS trajectories.
Who’s buying (and selling): institutional flows and S&P 500 buzz
Fund flows send a mixed but active signal
Fresh 13F data summarized by MarketBeat show that large institutions are not ignoring Marvell: [27]
- NewEdge Advisors LLCincreased its MRVL stake by 32.8% in Q2, adding 61,877 shares to bring its holdings to 250,253 shares worth about $19.4 million at quarter‑end.
- Frontier Capital Managementreduced its position by 19.9%, selling 158,227 shares and ending Q2 with 638,203 shares worth roughly $49.4 million, or about 0.07% of the company.
Overall, institutional ownership remains high — above 80% of shares — with large positions from Vanguard, State Street, Brown Advisory and others, plus a new stake from Norges Bank. [28]
The cross‑currents suggest professional investors are actively rebalancing rather than abandoning the name.
S&P 500 inclusion: a technical wildcard
MarketWatch recently pointed out that Marvell has finally cleared the S&P Dow Jones Indices profitability requirements and may now be eligible for addition to the S&P 500, given its market cap and positive earnings streak. [29]
There is no guarantee MRVL will be added, or when. But if it is, index funds and closet indexers would likely need to buy millions of shares, providing a potential technical tailwind independent of the Amazon/Microsoft noise.
The bull case: AI connectivity powerhouse at a (relatively) sane price
Supporters of Marvell’s stock argue that the medium‑term AI infrastructure story matters more than short‑term rumor volatility. The bullish narrative typically leans on:
- AI‑centric revenue mix: Data centers, including AI systems, now represent roughly 70%+ of Marvell’s sales — and management expects that slice to keep growing faster than the rest of the business. [30]
- Broad connectivity stack: From optical DSPs, Ethernet switches and PCIe retimers to custom ASICs and now Celestial AI’s Photonic Fabric, Marvell offers a near end‑to‑end portfolio for AI cluster connectivity — something bulls say few rivals can match. [31]
- Long‑dated growth guide: The company is explicitly targeting 25% data‑center growth in FY2027 and 40% in FY2028, numbers that, if achieved, could make today’s P/E multiple look reasonable in hindsight. [32]
- Valuation not at peak hype: With a P/E in the low‑30s and forward P/E around 30x — below its own recent average and much lower than some AI highfliers — bulls see MRVL as a way to play AI infrastructure without paying nosebleed multiples. [33]
For that camp, the current sell‑off is less a verdict on Marvell’s prospects and more a chance to accumulate a high‑quality AI infrastructure name while Wall Street argues over who really owns which hyperscaler socket.
The bear case: concentration, rumor risk and execution challenges
Skeptics don’t dispute that AI is a secular tailwind — they worry more about who captures the spoils and how much is already priced in.
Key points in the cautious camp:
- Customer concentration: A heavy reliance on a handful of hyperscalers (especially Amazon and Microsoft) means that winning or losing a single custom ASIC program can swing billions in future revenue. The Trainium saga itself is Exhibit A for that vulnerability. [34]
- Signal‑to‑noise problem: When one research house claims, with “high conviction,” that Marvell lost Amazon to Alchip, and the company plus AWS say the opposite, it’s hard for investors to know whom to trust. In that environment, volatility itself becomes a risk. [35]
- Integration and ramp risk: The Celestial AI acquisition is strategic but large; integrating a bleeding‑edge optical startup and hitting the $500 million / $1 billion revenue run‑rate targets for FY2028–29 will require flawless execution in a very competitive optical‑interconnect market. [36]
- Valuation downside if growth disappoints: Even a “reasonable” 30x forward P/E can compress quickly if revenue growth slows from the 30–40% range to something more pedestrian, especially if the broader AI trade cools off. [37]
For bears, MRVL is a high‑beta AI infrastructure proxy where small changes in hyperscaler capex or socket wins can have outsized effects on both fundamentals and sentiment.
What to watch next for Marvell stock
For investors tracking MRVL into 2026, several catalysts and questions stand out:
- Clarification on Amazon and Microsoft ASIC programs
- Future disclosures (or lack thereof) around Trainium 3/4 and other custom accelerators will likely determine whether Benchmark’s thesis or the CEO’s reassurances end up closer to the truth. [38]
- Progress on Celestial AI and optical scale‑up
- Design‑win announcements, especially with large hyperscalers, and evidence that revenue is tracking toward the FY2028–29 targets will be key to justifying the acquisition price. [39]
- Adoption of Golden Cable AECs and Alaska P retimers
- As AI racks move from 800G to 1.6T and beyond, Marvell’s success in AECs and PCIe 6 retimers will show whether it can maintain leadership in the messy, hardware‑heavy part of AI infrastructure. [40]
- Any news on S&P 500 inclusion
- An announcement that MRVL is joining the index could create a mechanical bid from passive funds, potentially smoothing volatility in the short term. [41]
- Macro AI capex trends
- Ultimately, Marvell’s fate is tied to whether hyperscalers maintain their aggressive AI capex plans through 2026–2028 or decide to hit the brakes.
Marvell Technology today sits at the intersection of very real AI infrastructure demand and very loud headline risk. Whether MRVL at ~$88 proves to be a temporary waystation or a fair long‑term price will depend less on this week’s Trainium drama and more on how consistently the company can convert its broad connectivity roadmap into durable, diversified AI data‑center revenue.
References
1. www.marketbeat.com, 2. www.marketbeat.com, 3. investor.marvell.com, 4. investor.marvell.com, 5. futurumgroup.com, 6. investor.marvell.com, 7. www.marketbeat.com, 8. www.marketbeat.com, 9. investor.marvell.com, 10. investor.marvell.com, 11. investor.marvell.com, 12. investor.marvell.com, 13. investor.marvell.com, 14. investor.marvell.com, 15. www.marketbeat.com, 16. www.marketbeat.com, 17. www.tipranks.com, 18. www.marketscreener.com, 19. investorsobserver.com, 20. www.marketbeat.com, 21. www.tipranks.com, 22. www.marketbeat.com, 23. www.chartmill.com, 24. www.chartmill.com, 25. simplywall.st, 26. www.marketbeat.com, 27. www.marketbeat.com, 28. www.marketbeat.com, 29. www.marketwatch.com, 30. www.marketscreener.com, 31. investor.marvell.com, 32. www.marketbeat.com, 33. simplywall.st, 34. investorsobserver.com, 35. www.tipranks.com, 36. investor.marvell.com, 37. www.marketbeat.com, 38. www.tipranks.com, 39. investor.marvell.com, 40. investor.marvell.com, 41. www.marketwatch.com
