Marvell’s $177B AI Bet: Q1 Beat, 102.4 Tbps Switch, NVLink

Marvell Technology (NASDAQ: MRVL) has quietly become one of the most important — and most expensive — AI infrastructure stocks on the tape. As of the most recent close the shares trade near $202, valuing the custom-silicon and connectivity vendor at roughly $177 billion in market cap and putting them up about 138% year-to-date. That makes Marvell one of the strongest large-cap performers of 2026, with the move backed by a record Q1 fiscal 2027 earnings print, a flurry of AI-infrastructure product launches, and a deepening partnership with Nvidia that has investors calling Marvell a credible trillion-dollar candidate.

Q1 FY27: a record print, then a big raise

Marvell reported Q1 fiscal 2027 results on May 27, 2026, with revenue of $2.418 billion, up 28% year over year, and non-GAAP diluted EPS of $0.80. Data center was again the engine: that segment delivered $1.833 billion of revenue, up about 27% YoY, and now accounts for roughly three-quarters of the company’s top line. Non-GAAP gross margin came in at 58.9% (GAAP 52.1%), reflecting the heavier custom-silicon mix in the data-center business.

CEO Matt Murphy said the company is seeing “exceptional AI-related bookings,” driving “significant” upward revisions to both fiscal 2027 and fiscal 2028 outlooks — demand he attributed to optical solutions, Ethernet switches, and custom accelerator products. The Q2 guide reflects that confidence: revenue of $2.7 billion plus/minus 5% (about +35% YoY at the midpoint), non-GAAP EPS of $0.93 plus/minus $0.05, and non-GAAP gross margin of 58.25% to 59.25%.

Metric Q1 FY27 Actual Q2 FY27 Guide YoY (Q1)
Total revenue $2.418B $2.70B ±5% +28%
Data center revenue $1.833B +27%
Non-GAAP gross margin 58.9% 58.25–59.25%
Non-GAAP EPS (diluted) $0.80 $0.93 ±$0.05
GAAP EPS (diluted) $0.04 $0.37 ±$0.05
Source: Marvell Q1 FY27 press release, May 27, 2026.

The 102.4 Tbps switch, and why it matters

On June 1, 2026 Marvell announced the availability of the industry’s first 102.4 Tbps switch purpose-built for AI and cloud data-center infrastructure. The number sounds abstract; the use case is not. As AI training clusters scale into the hundreds of thousands of accelerators, every additional rack of GPUs creates exponentially more east-west traffic between accelerators. The bottleneck for next-generation models is increasingly not compute itself but the fabric that moves activations and gradients between them.

A 102.4 Tbps single-chip switch roughly doubles the bandwidth density of the prior generation. For hyperscalers building AI factories, that translates directly into fewer switches per row, lower power per bit moved, and shorter optical reach — all of which compound when you are trying to wire 100,000+ GPUs into a single training job. It is the same arms race that has rewarded Broadcom’s Tomahawk and Jericho lines; Marvell is now publicly competing at the very top of it.

Why investors are circling the “next trillion-dollar” label

The bullish framing on Marvell rests on three pillars that the Q1 print and June product cadence reinforce:

  • Custom silicon for hyperscalers. Marvell designs custom AI accelerators (sometimes called XPUs or ASICs) on behalf of cloud customers who do not want to depend exclusively on Nvidia merchant GPUs. Murphy has flagged multiple large custom-silicon programs ramping into FY27 and FY28.
  • Optical and electrical connectivity. Marvell owns category-leading positions in DSPs for optical transceivers and in PAM4 SerDes — the analog plumbing that makes 800G and 1.6T optical links work. Every additional GPU in a training cluster pulls more of this content.
  • Nvidia ecosystem integration. In March 2026 Marvell joined Nvidia’s NVLink Fusion ecosystem, the program that allows third-party silicon to interconnect directly with Nvidia accelerators over NVLink rather than only over PCIe. That places Marvell’s custom XPUs and switches inside Nvidia-anchored AI factories rather than competing with them.

CEO Matt Murphy is also delivering the COMPUTEX 2026 keynote on June 2 in Taipei, titled “The Future of AI Scaling Depends on Connectivity.” The theme — that data movement, not raw compute, is the next bottleneck — is exactly the thesis the stock is pricing.

Performance in context

Marvell 52-week range as of June 2, 2026 Bar chart showing Marvell Technology’s 52-week low, recent close, and 52-week high as of June 2, 2026. MRVL: 52-week range and recent close $250 $170 $90 $0 $58.61 52w low $202.31 Recent close $218.26 52w high
Source: Yahoo Finance, intraday June 2, 2026. YTD performance: +138.06%.

At the current quote Marvell is up roughly 245% from its 52-week low of $58.61 and sits about 7% below the 52-week high of $218.26. The market is paying for accelerating data-center growth, rising custom-silicon mix, and a Q2 guide that calls for revenue growth to re-accelerate from 28% to 35% YoY. That is unusual at this revenue scale.

What to watch — and what could go wrong

The bear case is concentration and competition. Marvell’s custom-silicon programs are anchored by a small number of hyperscale customers; any one of them building more in-house, shifting to a competitor, or pausing a program would have outsized P&L impact. On the merchant connectivity side, Broadcom remains a formidable competitor in switch silicon and DSPs, and Nvidia itself is vertically integrating more of the fabric (NVLink, Spectrum-X, BlueField). The very ecosystems that lift Marvell today could compress its margin profile if competitive intensity rises.

Valuation is the second worry. At roughly $177 billion in market cap on a forward revenue run-rate near $11 billion, the stock screens expensive on traditional metrics. The thesis only works if hyperscale capex stays elevated, custom-silicon mix continues to climb, and AI-driven optical / switch content per GPU keeps growing — all of which require management execution on a quarterly cadence.

Near-term catalysts include the COMPUTEX keynote on June 2, the next earnings print in late August (Q2 FY27), and any incremental hyperscaler custom-silicon program disclosures. For now, the read on the print and the product roadmap is simple: connectivity is the next bottleneck of the AI build-out, and Marvell is selling the picks and shovels.

Sources

Disclosure: This article was produced with AI assistance and reviewed before publication. It is for informational purposes only and is not investment advice.

Leave a Comment