Updated: December 8, 2025
Marvell Technology, Inc. (NASDAQ: MRVL) just got a harsh reminder that being an “AI infrastructure winner” doesn’t make you invincible.
On Monday, December 8, Marvell stock fell sharply after a high‑profile analyst downgrade tied to possible custom‑chip design losses at Amazon Web Services and fresh worries that Microsoft may shift some custom silicon work to Broadcom. Intraday, MRVL dropped more than 10% before recovering part of the loss, trading in a roughly $88.9–$92.6 range — about 7% below Friday’s close of $98.91. [1]
Paradoxically, on the very same day, Zacks upgraded Marvell to its top “Strong Buy” rank, citing rising earnings estimates and strengthening fundamentals. [2] Welcome to the 2025 semiconductor market: brutally efficient, wildly opinionated.
Below is a breakdown of what changed today, how it fits into Marvell’s AI data‑center story, and what Wall Street is modeling from here.
Marvell Technology stock today: sharp sell‑off after a big AI rally
After a strong post‑earnings run, MRVL walked straight into a wall.
- Price action (December 8, 2025): Data from StockAnalysis shows Marvell trading between $88.92 and $92.61 during the session, versus a previous close of $98.91, implying an intraday drop of roughly 10% at the low and a loss of about 6.9% based on that end‑of‑day close. [3]
- Context: Over the last 12 months, MRVL has traded between $47.09 and $127.48, with the stock still solidly above its lows but well off its AI‑euphoria peak. [4]
The immediate driver isn’t a macro scare or a disaster in Marvell’s reported numbers. It’s fear that a couple of its most strategic hyperscale customers — Amazon and Microsoft — might not be as locked in as investors assumed.
Why Marvell stock is falling: Amazon design loss and Microsoft–Broadcom chatter
The trigger for the sell‑off is remarkably specific: the market is suddenly doubting Marvell’s grip on high‑value custom AI silicon work at the big clouds.
1. Benchmark downgrade: Amazon’s Trainium designs in doubt
Benchmark analyst Cody Acree cut Marvell from Buy to Hold, and his reasoning hit investors where it hurts. [5]
Key points from Acree’s thesis, as summarized by GuruFocus and Seeking Alpha:
- Benchmark now has a “high degree of conviction” that Marvell lost Amazon’s Trainium 3 and Trainium 4 custom chip designs to Taiwanese rival Alchip. [6]
- That potential loss helps explain why Marvell is only projecting about 20% XPU (accelerator) revenue growth in calendar 2026, despite prior commentary about avoiding a “revenue air pocket” with Amazon. [7]
- Acree argues that Marvell’s upbeat guidance is leaning heavily on ongoing Trainium 2 volumes and Amazon’s Kuiper low‑Earth‑orbit satellite projects, while newer Trainium generations appear more uncertain. [8]
In plain language: the market had been treating Amazon’s future AI chips as a fairly secure pillar of Marvell’s growth. Benchmark is saying “maybe not,” and that’s a big deal.
2. Microsoft may be flirting with Broadcom
Layered on top of the Amazon worry is a report that Microsoft is in talks with Broadcom about future custom chip designs. [9]
According to Investing.com’s coverage:
- A report from The Information suggested Microsoft is exploring new custom chip work with Broadcom, potentially shifting some future design wins away from Marvell. [10]
- Benchmark’s downgrade plus the Microsoft–Broadcom chatter led to Marvell being down roughly 6–8% in early trading, with the stock flagged as one of the day’s notable losers. [11]
Taken together, investors suddenly have to re‑evaluate how much of Marvell’s future AI revenue is tied to a small number of massive customers — and how durable those relationships really are.
Zacks Strong Buy: the oddly bullish counter‑signal
In a plot twist, Zacks Equity Research upgraded Marvell to its highest rank, Zacks Rank #1 (Strong Buy), today. [12]
According to Zacks’ note:
- Marvell’s earnings estimates have been moving higher, with the Zacks consensus EPS for the fiscal year ending January 2026 rising about 2.7% over the last three months. [13]
- Zacks now expects Marvell to earn about $2.84 per share this year, roughly flat with the prior year but with a much healthier earnings mix. [14]
- Zacks emphasizes its long‑running backtest that shows Zacks Rank #1 stocks (top 5% of its coverage universe) have historically outperformed the market. [15]
Zacks and Benchmark are looking at the same company and coming to opposite short‑term conclusions: one focuses on estimate momentum and AI‑driven demand, the other on concentrated customer risk and potential lost designs.
That tension is exactly why MRVL is trading like a tug‑of‑war rope today.
Under the hood: record Q3 results and a big bet on photonics
Today’s sell‑off lands barely a week after Marvell posted arguably its best quarter yet and doubled down on its AI data‑center roadmap.
Q3 FY 2026: AI data‑center engine in full gear
For the quarter ended November 1, 2025 (Marvell’s Q3 FY 2026), the company reported: [16]
- Revenue: $2.075 billion, up 37% year‑over‑year, a new record and slightly ahead of guidance.
- Non‑GAAP EPS: $0.76, up from $0.43 a year earlier.
- Data center revenue: about $1.52 billion, up 38% YoY and representing ~73% of total revenue. [17]
- Strong double‑digit growth in enterprise networking (+57% YoY) and carrier infrastructure (+98% YoY), while automotive/industrial declined after the sale of the automotive Ethernet business. [18]
CEO Matt Murphy described the quarter as driven by “strong demand for our data center products” and guided for full‑year revenue growth above 40% with robust Q4 growth still to come. [19]
Independent research from Futurum Group framed the quarter as evidence of “accelerating AI‑driven demand across custom silicon and optics,” with management pointing to sequential growth into next year and improved order visibility. [20]
Celestial AI acquisition: betting on photonic fabrics
On top of earnings, Marvell announced a sizeable deal to acquire Celestial AI, a private company building photonic fabric for next‑generation AI data centers. [21]
From Marvell’s and Futurum’s disclosures:
- Deal value: about $3.25 billion (roughly $1.0B cash + $2.25B stock), plus up to $2.25B in contingent stock tied to revenue milestones through FY 2029. [22]
- Celestial’s photonic fabric chiplets deliver up to 16 Tbps of bandwidth per chiplet — roughly 10× current 1.6T optical ports in scale‑out applications — and are designed to be co‑packaged with AI accelerators and switches. [23]
- Marvell expects Celestial AI to reach a $500 million annualized run rate by Q4 FY 2028, and $1 billion annualized by Q4 FY 2029, with revenue becoming meaningful from the second half of FY 2028. [24]
- Management outlined a path toward roughly $10 billion in organic revenue by FY 2027, driven by custom silicon and optical interconnect, with data‑center revenue growth projected around 25% in FY 2027 and ~40% in FY 2028. [25]
In other words, the long‑term AI infrastructure story still looks very aggressive on paper. Today’s sell‑off is about whether Amazon and Microsoft threaten that slope.
What Wall Street is modeling for Marvell now
Despite today’s drama, the consensus view on Marvell remains bullish — just not unanimous.
Analyst price targets and ratings
Different data providers tally slightly different universes of analysts, but the picture is consistent:
- TipRanks:
- Average 12‑month price target:$121.04
- High / low:$156 / $85
- Implied upside: about 22% from a reference price of $98.91
- Consensus rating:Strong Buy, based on 22 Buy and 7 Hold ratings, 0 Sells. [26]
- MarketBeat:
- Average target: about $111.56, with a similar high end in the mid‑$150s and a consensus rating of “Moderate Buy”. [27]
- StockAnalysis:
- Lists an average rating of “Buy”, with 13 Strong Buy, 8 Buy, 11 Hold recommendations. [28]
Drilling into individual calls after Q3 earnings, Benzinga and TipRanks highlight a wave of raised targets from major firms: [29]
- Needham: $95 → $120, Buy
- Wells Fargo: $90 → $135, Overweight
- Evercore ISI: $122 → $156, Outperform
- Deutsche Bank: $90 → $125, Buy
- Rosenblatt, KeyBanc, Piper Sandler, Stifel, Roth MKM and others also pushed targets into the $110–$135+ range with bullish language around AI data‑center momentum.
Today’s twist is that Benchmark — which just raised its target to $130 last week — has now downgraded to Hold after sensing Amazon design losses, and Barclays/Goldman sit more cautiously in the $90–$105 range with Hold ratings. [30]
Revenue and earnings forecasts
StockAnalysis’ aggregated forecasts sketch a very fast‑growing but lumpy story: [31]
- Revenue FY 2026 (current year): about $8.32 billion, up 44% from $5.77B.
- Revenue FY 2027 (next year): about $9.66 billion, another 16% growth.
- EPS FY 2026: roughly $2.87 (swinging from a GAAP loss last year).
- EPS FY 2027: about $3.44, implying ~20% growth.
Futurum’s analysis broadly aligns with management’s message: a multi‑year AI data‑center build‑out with rising contributions from custom accelerators and optics, and a goal of maintaining sequential revenue growth every quarter next year. [32]
The catch: a lot of that is predicated on large hyperscaler programs landing and ramping as expected. Today’s headlines are investors stress‑testing that assumption.
Who is actually buying and selling MRVL?
Price is where the votes get counted, but it’s useful to see who’s voting.
Big institutions: mixed signals, but heavy ownership
Two fresh 13F‑driven stories on December 8 shine some light on institutional positioning: [33]
- Temasek Holdings (Singapore’s sovereign fund)
- Trimmed its stake by 3.5% in Q2, selling about 42,387 shares.
- Still holds 1,156,814 shares, worth roughly $89.5 million and representing about 0.13% of the company.
- Oxbow Capital Management HK Ltd
- Initiated a new position of 127,000 shares in Q2 (~$9.83M).
- MRVL is now about 2.5% of Oxbow’s portfolio and its 9th‑largest position.
MarketBeat notes that institutional investors own roughly 83.5% of Marvell’s float, with large stakes reported from Norges Bank, Nuveen, Amundi, Franklin Resources, and others. [34]
Insiders and capital returns
On the insider and capital‑allocation front: [35]
- Insiders reportedly bought about 27,200 shares over the past 90 days, including a 6,800‑share purchase by COO Chris Koopmans around $78 per share.
- Insider ownership is modest (~0.33%), but the direction — buying, not selling — generally reads as a vote of confidence.
- Marvell has authorized a $5.0 billion share repurchase plan, equivalent to up to 7.8% of shares outstanding, and continues to pay a $0.06 quarterly dividend (about 0.2% yield at recent prices).
That combination — massive buyback firepower, tiny dividend, heavy institutional ownership — is pretty classic “high‑growth tech trying to signal it believes its own story.”
Valuation snapshot: expensive or just AI‑priced?
With AI hardware names, the question is rarely “is it cheap” and more “is the growth big enough to justify expensive.”
According to MarketBeat’s and company data: [36]
- At recent prices in the low‑ to mid‑$90s, Marvell’s market cap is around $84 billion.
- Trailing P/E: roughly 34–35×.
- PEG ratio: about 0.96, implying that on some analyst models, earnings growth is roughly in line with the valuation multiple.
- Forward P/E on consensus is in the low‑30s for this year and high‑20s for next year, assuming forecasts hold.
Morningstar recently flagged Marvell as one of a handful of “new 4‑star stocks” — their way of saying the stock has slipped into undervalued relative to fair value territory after the recent pullback. [37]
So MRVL isn’t cheap on traditional metrics, but relative to its own history and AI peers, the argument is that you’re paying a slightly below‑normal multiple for above‑normal growth — if the growth sticks.
Key bull and bear arguments investors are weighing today
Strip away the noise, and Marvell’s investment debate now hangs on a few core questions.
The bull case
Supporters of Marvell stock point to:
- Explosive AI data‑center demand: Q3 showed 38% data‑center revenue growth and record total revenue, with management guiding for >40% full‑year revenue growth and strong Q4. [38]
- Deep custom‑silicon relationships: Multiple hyperscaler XPUs (accelerators) already in production, with 3‑nm designs secured through next year and 2‑nm designs in development, plus another major XPU customer expected around FY 2028. [39]
- End‑to‑end optics and interconnect: A portfolio spanning DSPs, retimers, AECs, low‑power optics, switches, and now Celestial’s photonic fabric, positioning Marvell as a “one‑stop shop” for AI rack‑scale connectivity. [40]
- Multi‑year growth roadmap: Internal and external analyses sketch a path to around $10B in revenue by FY 2027, with data‑center growth in the mid‑20s to ~40% and Celestial potentially adding $0.5–1B annual run‑rate by FY 2028–29. [41]
- Supportive analyst and quant signals: Strong Buy / Buy consensus, rising estimates, Zacks Rank #1, and a long list of double‑digit‑upside price targets. [42]
If you believe this trajectory, today’s drop looks like a volatility tax on a high‑beta AI infrastructure leader.
The bear case
Skeptics (or just cautious types) highlight:
- Hyperscaler concentration risk: A lot of Marvell’s upside is tied to a small handful of customers. Potential loss of Amazon’s Trainium 3 & 4 designs and any shift at Microsoft toward Broadcom would hit the custom‑chip growth engine hard. [43]
- Management vs. reality on “no air pocket”: Management has repeatedly talked about avoiding “air pockets” in custom revenue, but today’s downgrade explicitly questions whether that’s realistic if newer Amazon designs are gone. [44]
- Valuation risk in a cyclical industry: Even after the pullback, a mid‑30s P/E and high‑20s forward P/E leave little room for disappointment in an industry where capex cycles can and do turn. [45]
- Execution risk on Celestial AI: The photonics bet is big, expensive, and back‑loaded — meaningful revenue only starts appearing in FY 2028 under current plans, with a lot of integration and technology risk in between. [46]
Today’s sell‑off is the market saying: “We still like the story, but we want a bigger discount for these risks.”
Bottom line: what today’s move really means for MRVL stock
Zooming out, nothing in today’s headlines says “Marvell’s AI data‑center thesis is dead.” The quarter it just reported, the guidance it issued, and the long‑term roadmap around custom XPUs and photonics are all still intact on paper. [47]
What changed on December 8 is the confidence interval:
- The street just got a plausible, sourced narrative that Marvell may have lost some high‑profile Amazon work and could face more intense competition for future cloud programs. [48]
- The Microsoft–Broadcom chatter reminds everyone that hyperscalers like to diversify suppliers and bargain hard. [49]
- Meanwhile, the fundamental numbers and long‑term AI demand still look very strong, which is why consensus remains firmly bullish and research firms like Zacks are leaning in rather than bailing out. [50]
References
1. stockanalysis.com, 2. www.nasdaq.com, 3. stockanalysis.com, 4. www.marketbeat.com, 5. www.gurufocus.com, 6. www.gurufocus.com, 7. www.gurufocus.com, 8. www.gurufocus.com, 9. www.investing.com, 10. www.investing.com, 11. www.investing.com, 12. www.nasdaq.com, 13. www.nasdaq.com, 14. www.nasdaq.com, 15. www.nasdaq.com, 16. investor.marvell.com, 17. investor.marvell.com, 18. futurumgroup.com, 19. investor.marvell.com, 20. futurumgroup.com, 21. investor.marvell.com, 22. futurumgroup.com, 23. investor.marvell.com, 24. investor.marvell.com, 25. futurumgroup.com, 26. www.tipranks.com, 27. www.marketbeat.com, 28. stockanalysis.com, 29. www.benzinga.com, 30. stockanalysis.com, 31. stockanalysis.com, 32. futurumgroup.com, 33. www.marketbeat.com, 34. www.marketbeat.com, 35. www.marketbeat.com, 36. www.marketbeat.com, 37. www.morningstar.com, 38. futurumgroup.com, 39. futurumgroup.com, 40. futurumgroup.com, 41. futurumgroup.com, 42. www.tipranks.com, 43. www.investing.com, 44. futurumgroup.com, 45. www.marketbeat.com, 46. investor.marvell.com, 47. futurumgroup.com, 48. www.investing.com, 49. www.investing.com, 50. www.nasdaq.com
