Marvell Stock Price Forecast - MRVL at $83: The AI Data-Path Trade After GPU Excess

NASDAQ:MRVL – AI connectivity and memory at a discounted multiple

Business mix, price context and AI-cycle exposure for NASDAQ:MRVL

Marvell Technology NASDAQ:MRVL is now primarily an AI data-center supplier, not a generic mixed-signal semi. About 73% of revenue comes from data center workloads, so the share price is tied to hyperscaler capex and AI-cluster buildouts.

The stock trades around $83.21, down about 1.7% on the day, versus a 52-week range of $47.09–$127.48 and a market cap near $70.4B. The trailing P/E is roughly 29.0x, price-to-book is about 5.1x, and the dividend yield is a token 0.29%, confirming this is a growth and infrastructure name rather than an income vehicle.

At this level investors are paying below-peak AI multiples for a business that is far more levered to connectivity, memory, and custom accelerators than to older storage or carrier products. The equity now behaves as a high-beta vehicle on the AI infrastructure trade, with fundamentals anchored in bandwidth, latency and power efficiency.

Revenue growth, earnings leverage and free-cash-flow dynamics

In the latest reported quarter to November 2025, NASDAQ:MRVL delivered revenue of about $2.07B, up 36.8% year-on-year on a reported basis. Adjusting for the divestiture of the Automotive Ethernet unit in mid-2025, underlying growth is closer to the low-40% range, which is a strong print at this scale.

Data-center revenue grew around 38% versus the prior year, slower than the previous quarter’s near 69% surge but still firmly in high growth. The company continues to win more than 20 XPU and XPU-attach sockets and meaningful NIC and AEC designs with large cloud operators, which confirms that its solutions remain competitive in new AI racks.

Gross margin came in near 59.7%, down about 80 bps year-on-year, driven by product mix and cost pressures. The crucial offset is operating-expense discipline: total opex rose only 1.2% year-on-year, with GAAP SG&A actually declining. That cost control expanded operating margin by roughly 660 bps, translating modest gross-margin pressure into very strong profit growth.

Net income was reported at about $1.90B, up 381% year-on-year, with the headline net margin above 90% inflated by non-recurring items. Adjusted EPS reached $0.76, up roughly 76–77% versus the prior year and ahead of analyst expectations. Cash from operations was around $582M, while free cash flow hit about $1.42B, up 127% year-on-year, and net cash increased by roughly $1.49B. The business is converting the AI cycle into real cash, not just optical EPS.

AI data-path positioning: monetizing bandwidth, latency and power

The core thesis for NASDAQ:MRVL is its position in the AI data path rather than in GPU cores. Management’s view, supported by industry forecasts, is that the number of custom accelerators (XPUs) will overtake GPUs in units shipped by 2028, with the gap widening after that. Marvell is tied to this transition via custom ASICs, high-speed NICs, optics and switching.

In live AI clusters, bottlenecks appear first in bandwidth, latency and power delivery, not in theoretical FLOPs. Once hyperscalers have purchased and installed GPUs, incremental capex moves into interconnects and memory subsystems that raise utilization on those sunk GPU investments. That is precisely where Marvell’s custom NICs, switches and optical components sit.

Custom NICs already represent more than 30% of NICs inside certain mega-scale data centers, indicating that operators accept higher unit prices in exchange for tighter integration and better performance. As clusters scale and models grow, that willingness to pay for optimized data paths rather than commodity interfaces is likely to deepen.

Memory architecture, HBM and SRAM: squeezing more work from each package

On the memory side, NASDAQ:MRVL is focused on custom base dies and SRAM structures to raise effective capacity and throughput per package.

By tailoring the base die that links compute cores and high-bandwidth memory, Marvell can boost memory capacity by up to 33% within a given footprint. Interface power can be reduced by as much as 70%, freeing data-center power budgets, and silicon area available for cores can rise by about 25%, making each package more compute-dense.

Custom SRAM designs built directly into compute silicon can deliver up to 17x higher data throughput for certain AI workloads. That kind of uplift shortens training and inference cycles, improves hardware utilization and enhances infrastructure returns without requiring a complete GPU refresh.

The result is a structural efficiency layer around GPUs and XPUs. Nvidia captures value on the compute die; Marvell pulls value from the surrounding memory and interconnect stack, where power, latency and bandwidth decide real-world performance.

CXL memory controllers, DRAM inflation and the Structura opportunity

The sharp rise in DRAM spot prices—roughly a 4x jump between August and November—has made memory one of the most painful cost lines for AI infrastructure buyers. That environment is ideal for CXL-based memory controllers such as Marvell’s Structura platform.

Structura controllers let hyperscalers pool and share memory across CPUs, GPUs and XPUs, combining different DRAM generations rather than relying solely on the newest and most expensive nodes. In practice, this gives older DRAM a “second life” while still meeting bandwidth and latency requirements for inference, recommendation and other memory-heavy workloads.

Management expects Structura memory controllers, together with associated NIC products, to generate around $2B of revenue over the next three years, implying roughly $700M of annualized revenue contribution by 2028. Current consensus forecasts for NASDAQ:MRVL near $10B of revenue by CY26/FY27 do not fully reflect that stream. If the ramp tracks guidance, controllers will be a significant incremental leg on top of the existing data-center portfolio.

A key part of the Structura thesis is neutrality. The controllers are being engineered to interoperate across AMD EPYC and Intel Xeon platforms and with DRAM from Samsung, SK Hynix and Micron. That cross-vendor capability positions Marvell as a neutral fabric layer in heterogeneous AI data centers where customers want to avoid lock-in to a single CPU or memory supplier.

Connectivity M&A: Celestial AI and XConn expand NASDAQ:MRVL’s reach

Marvell is reinforcing the connectivity and memory strategy via targeted acquisitions.

The company is acquiring Celestial AI for about $3.25B, adding a photonic-fabric specialist to its stack. Celestial’s first-generation chiplet delivers roughly 16 Tbps of bandwidth per device and is designed to link disaggregated compute and memory across packages and modules. Internal projections point to around $500M of annual recurring revenue by FY28 Q4 from Celestial AI products.

Separately, Marvell is buying XConn for about $540M, adding advanced PCIe and CXL switching technology. XConn is expected to contribute around $100M of revenue by FY28. While smaller in absolute terms, this acquisition plugs an important gap in the CXL switching layer that connects pooled memory and accelerators inside and across racks.

Together, Celestial AI and XConn shift NASDAQ:MRVL further toward owning the full connectivity stack in AI data centers, from photonic fabrics through CXL memory pools and PCIe switches to custom NICs and ASICs. That breadth is a differentiator as the industry pushes toward disaggregated architectures.

Balance sheet, capital deployment and insider signaling

The balance sheet supports this investment cycle.

At the end of the latest quarter, NASDAQ:MRVL held about $2.71B in cash and short-term investments, up roughly 213% year-on-year. Total assets were around $21.58B, total liabilities roughly $7.52B, leaving equity of about $14.06B. Return on assets was near 4.4%, and return on capital just under 5%, metrics that should improve as the AI and connectivity portfolio matures and one-off items roll off.

Operating cash flow of $582.3M and free cash flow of $1.42B give the company room to fund R&D, integrate Celestial AI and XConn, and maintain a conservative capital structure without relying on aggressive leverage. Financing cash flows were about -$1.48B, while net cash increased by roughly $1.49B, underscoring that acquisitions and buybacks are being financed from strength rather than desperation.

For tracking management conviction and capital behavior, insider activity on the MRVL stock profile and insider transactions page is relevant. Persistent net buying into AI-cycle volatility would reinforce the long-term thesis, whereas large or repeated disposals near local highs would need to be weighed against the public growth narrative.

Risk set: AI de-rating, capex digestion and China exposure

The risk profile is clear and should not be glossed over.

The first risk is correlation to the broader AI equity trade. If AI and semiconductor names de-rate together, NASDAQ:MRVL will likely sell off regardless of its relative quality. Index flows, factor selling and ETF unwinds can overpower stock-specific fundamentals over short windows.

The second risk is hyperscaler capex digestion. Data-center growth has already slowed from around 69% to the high-30% range. If one or more major AI programs are delayed, resized or cancelled, Marvell’s quarterly revenue could flatten or decline, and the share price will respond quickly.

The third risk is geographic concentration. Roughly 40% of Q3 destination-of-shipment revenue is tied to China. Any tightening of export controls on advanced connectivity or accelerator-adjacent silicon, or a sharp macro slowdown in that market, would hit both volumes and valuation multiples.

The fourth risk concerns execution on the Structura and Celestial AI ramps. The $2B multiyear Structura target and $500M Celestial AI ARR guide assume sustained adoption of CXL memory controllers and photonic fabrics. If customers deploy more slowly, choose rival architectures, or if DRAM prices normalize faster than expected, the incremental revenue legs that underpin medium-term upside will be smaller than planned.

Valuation, target framework and payoff skew for NASDAQ:MRVL

Valuation work around NASDAQ:MRVL converges on a fair-value region near $112.50 by FY28, based on normalized EPS around $5.00 and a forward multiple near 22.5x, slightly below the current forward P/E and below the multiples many AI peers command.

With the stock around $83, that implies roughly 35% upside on earnings delivery. If Marvell hits about $5.00 of EPS and the price stays near today’s level, the P/E on that out-year number would compress toward 16–17x, which is hard to reconcile with mid-20s revenue growth and persistent operating leverage in AI infrastructure.

On a sales basis, the stock changes hands at roughly 9–10x forward revenue, versus mid-teens for some GPU and AI platform names. Consensus revenue near $10B by CY26, plus potential $700M annualized Structura contribution and the Celestial AI and XConn ramps by 2028, make the current revenue multiple look undemanding if execution is solid.

The downside case is a combination of AI bubble deflation, capex pauses and China shocks that cut both earnings and multiples. The upside case is that Marvell proves connectivity, controllers and photonic fabrics are secular growth engines and that its acquisitions hit their $500M + $100M revenue contributions with margins that support continued EPS expansion.

Investment stance on NASDAQ:MRVL – Buy with controlled size

After consolidating the data, Marvell Technology NASDAQ:MRVL is a Buy at the current $83–84 band, with a rational sizing cap around 3–4% of NAV in a concentrated portfolio.

The bull case rests on roughly 73% data-center exposure, high-30s to low-40s revenue growth, mid-70s EPS growth, more than 600 bps of operating-margin expansion, a visible $2B Structura opportunity, and additive Celestial AI and XConn connectivity revenues, all at a sub-30x trailing P/E and around 9–10x forward sales.

The bear case focuses on AI-trade de-rating, hyperscaler digestion and China risk, none of which can be diversified away at the single-stock level. For investors seeking exposure to the AI “plumbing” layer—bandwidth, memory pooling and connectivity rather than just GPUs—NASDAQ:MRVL offers a leveraged but still reasonably valued way to express that view.

That's TradingNEWS