MRVL — Marvell Technology, Inc.

Wall Street is firmly constructive: 60 Buy/Strong Buy vs 12 Hold and 1 Sell, consensus mean target around $271, and Piper Sandler initiated Overweight with a $270 target on 2026-09-11. The Q2 FY27 beat (revenue $2.739B, non-GAAP EPS $0.94), data-center revenue up 46% to $2.17B, raised FY27/FY28 guidance, and the Google custom-AI framework signed 2026-07-29 support the bull case. The pushback is valuation: as of 2026-09-13 the stock trades at roughly 56x next-FY EPS and 17x forward sales, so the 3-month -15.6% drawdown shows how quickly sentiment cools when expectations are this high.

Revenue has inflected from the FY24 trough: $1.161B in Q1 FY25 to $1.516B (Q3 FY25), $1.817B (Q4 FY25), $1.895B (Q1 FY26), $2.006B (Q2 FY26), $2.075B (Q3 FY26), $2.219B (Q4 FY26), $2.418B (Q1 FY27) and $2.739B (Q2 FY27). That is a clear acceleration, with the latest quarter +36.5% YoY and +13.3% QoQ, driven by data-center/AI. Gross margin has recovered from a charge-distorted 23.0% in Q3 FY25 to 53.1% in Q2 FY27; operating margin rose to 16.7% from 14.5% a year ago. GAAP net income is noisy (Q3 FY26's $1.901B included a large one-off gain; Q2 FY27 was $308M/$0.33), but operating cash flow and FCF are consistently strong at $478.8M FCF in Q2 FY27 and $483.1M in Q1 FY27. The trend is structural AI-driven growth, not a blip.

Expensive, even after the pullback. As of 2026-09-13, MRVL's TTM P/E is 77.4, forward P/E on next-FY EPS of $4.198 is about 56.2x at $236.10, P/S is 21.9x TTM (17.2x forward sales of $12.04B), P/B is 11.4x, and EV/EBITDA is 46.0x. That compares with a 5-year average P/S of 10.95 and a 1-year-used average P/E of 25.46. The PEG ratio of 0.027 is not meaningful given the explosive growth base. The premium is justified only if AI/data-center growth persists and margins expand; otherwise multiple compression is a real risk.

The market is pricing in the AI ramp, not waiting for it. At $236.10, MRVL trades at roughly 56x next-FY EPS of $4.198, so the stock needs the custom silicon and optical networking pipeline to convert into the raised FY27/FY28 guidance. Analyst targets ($271 consensus, $293 recent avg) imply upside if estimates move higher, but the 3-month -15.6% move shows that any sign of order digestion, delayed custom-ASIC ramps, or warrant dilution can hit the multiple fast. The set-up favors buying on estimate-confirming weakness, not chasing strength unless the October Investor Day raises multi-year visibility.

As of 2026-09-13, RSI(14) is 55.47 - neutral, neither overbought nor oversold. Price is 3.67% above the 20-day SMA, 8.88% above the 50-day, and 52.5% above the 200-day, so the long-term uptrend remains intact. But the 3-month return is -15.59% versus +168.72% over six months, showing a sharp correction within a powerful uptrend. The stock is above all major moving averages with room before overbought; a hold above the 20/50-day would be constructive for entry, while a break below the 50-day would warn the pullback is deepening. Beta is 2.253, so position sizing matters.

• AI infrastructure capex cycle: hyperscaler spending on custom accelerators, optical networking, and scale-up switches is the primary demand driver; data center was 79% of Q2 FY27 revenue.
• Interest rates and valuation sensitivity: MRVL's high multiple (~56x forward EPS) makes it vulnerable to rising discount rates or rotation out of long-duration growth equities.
• Semiconductor supply chain: fabless model relies on TSMC and advanced packaging; capacity constraints or geopolitical disruptions could limit shipments.
• US-China trade restrictions: export controls on high-end AI hardware could affect addressable market and customer demand.
• Competition and customer concentration: Broadcom, Nvidia, and in-house hyperscaler silicon compete for custom AI/ASIC sockets; Google framework is a major but concentrated relationship.

• Investor Day on 2026-10-06: management expected to outline long-term AI strategy, margin targets, and custom silicon roadmap.
• Q3 FY27 earnings late Nov 2026: first major test of raised guidance and custom silicon ramp.
• Google custom AI accelerator/storage/networking framework signed 2026-07-29; further design-win or deployment updates.
• Nvidia NVLink Fusion ecosystem integration and potential new hyperscaler design wins.
• Hyperscaler capex announcements and AI networking demand updates.

• Valuation/multiple compression: 56x forward EPS and 17x forward sales leave little room for execution misses.
• Customer concentration and framework dependence: Google deal includes warrants that may dilute shareholders; revenue is tied to a few hyperscalers.
• Execution risk on custom silicon and advanced packaging ramps, including timing and yield issues.
• Competitive pressure from Broadcom, Nvidia, and hyperscaler in-house silicon.
• Macro/geopolitical supply-chain risk: TSMC reliance and US-China export controls.
• High beta (2.253) and 52-week range $66.33-$329.88 mean sharp drawdowns are normal.

Marvell is a high-quality AI infrastructure growth story with accelerating revenue, data-center dominance, strong FCF, and a clean balance sheet. But at roughly 56x next-FY EPS and 17x forward sales as of 2026-09-13, the stock already discounts a lot of that success. Cautiously bullish: own/accumulate on pullbacks or after the October Investor Day confirms multi-year visibility, but avoid chasing at current levels without a margin of safety.

I applied a 58x forward P/E to the next-FY consensus EPS estimate of $4.198, yielding about $243.5, rounded to $245. The multiple is a premium to the 25.46x 5-year average P/E but justified by roughly 40% EPS growth, 37% YoY revenue growth, and data-center momentum; it is below the 65x+ bull-case multiple to reflect warrant dilution, customer concentration, and execution risk.