JPMorgan says Marvell stock TPU business reports are false By Investing.com

JPMorgan rejected reports that Marvell has won Google TPU business, but said the company is in more substantive talks on an LPU project and networking support. The note still highlighted strong underlying fundamentals, including 42% trailing revenue growth, 171% share gains over the past year, and expectations for 50% optical-product growth this year and 40% next year. The setup remains constructive, with investors focused on whether Marvell can keep beating estimates and convert Google-related discussions into contracts.

Analysis

The key read-through is not simply “MRVL up or down,” but that the market is conflating three distinct revenue pools: custom silicon wins, networking attach, and optics/interop share. If Google is still shopping across vendors, the value shifts toward whichever supplier can bundle design services, packaging, and software integration — which favors NVDA structurally more than MRVL, because NVIDIA can monetize the full AI stack while Marvell is still largely a component-and-program execution story. That makes the current enthusiasm in MRVL vulnerable if investors are pricing a TPU-style win before the procurement cycle actually closes.

The bigger second-order effect is on the AI supply chain: if hyperscalers diversify custom silicon vendors, the bottleneck migrates from “who wins the chip” to “who can ship the surrounding infrastructure fastest.” That is supportive for optics, networking, and advanced packaging, but it also means the strongest upside may accrue to names with broader platform leverage and pricing power, not necessarily the purest custom-silicon proxy. NVDA benefits from that ecosystem pull, while GOOGL’s leverage improves if it can extract lower unit costs from multiple bidders rather than committing to a single vendor.

The contrarian risk is that MRVL bulls are extrapolating a narrative premium into earnings before the contracts exist. After a large run, even a clean quarter can disappoint if the mix shifts toward lower-margin attach business or if optical growth merely meets, rather than beats, the already-lofty path. Over the next 1-3 months, the stock is most exposed to headline whiplash on Google program selection; over 6-12 months, the real test is whether management converts “discussion” into bookings faster than competitors can compress pricing.

What the market may be missing is that this setup is more about relative winners than outright directional beta. If hyperscaler capex remains strong, the ecosystem keeps working, but MRVL’s multiple is already closer to a successful outcome than a probable one. The better risk/reward may be to own the enablers with broader monetization paths and hedge the highest-expectation beneficiary that still lacks contract visibility.

Key Decisions for Investors

• Reduce or avoid initiating new long MRVL here; if already long, consider trimming 25-50% into strength and use a tight 8-10% trailing stop over the next 2-6 weeks in case Google headlines disappoint.
• Pair trade: long NVDA / short MRVL for 1-3 months, targeting relative outperformance if AI infrastructure spending stays robust but custom-silicon awards remain unsettled; size for a 1.5-2.0x gross exposure with the short capped via calls.
• Buy MRVL downside protection into the next catalyst window: 1-2 month put spreads financed partially by selling upside calls, aiming to monetize elevated implied vol around announcement risk.
• For investors seeking the cleaner AI-infra beta, rotate toward NVDA on dips rather than chasing MRVL; use pullbacks of 5-7% as entry zones and keep a 10% drawdown stop given valuation sensitivity.
• If Google-related award confirmation emerges, trade MRVL tactically for a 5-15% move higher, but treat it as a momentum event, not a fundamental re-rating until orders convert into backlog and margin evidence.