Marvell beat earnings and fell 7%: the payoff is FY2029
Marvell posted record $2.739B revenue and raised guidance, then dropped 7% because the $120B Google chip deal lands in fiscal 2029. What that timeline means for compute costs.
Marvell reported a record quarter, raised full-year guidance twice over, and the stock fell about 7% on Friday. The reason is a date. The $120 billion Google custom silicon deal does not move the revenue line in a serious way until fiscal 2029, and the market repriced accordingly. If you buy inference tokens, that date is worth understanding.
What actually happened
Per Marvell's Q2 FY2027 release, revenue was $2.739 billion, up 37% year over year and above the midpoint of its own guidance. Non-GAAP net income was $865.9 million, or $0.94 per diluted share. Data center revenue hit $2.17 billion, up 46%, now 79% of the company.
The company raised its FY2027 outlook to roughly $12 billion from $11.5 billion, and FY2028 to roughly $18 billion from $16.5 billion. CEO Matt Murphy said AI-related bookings remain exceptionally robust.
Then the timing. The Google agreement — inference accelerators, storage controllers, NICs, memory interface controllers, near-memory compute, with a warrant for up to 7% of Marvell shares tied to revenue milestones — carries roughly $120 billion of cumulative revenue potential across six years. Murphy put the meaningful financial impact in fiscal 2029. Shares fell 7-8% in early trading, per 24/7 Wall St., against a 185% year-to-date gain.
Why the AI compute timeline matters for your business
Cheaper inference is a 2029 story, not a 2027 one. The second-source silicon that should pressure token prices is still being taped out. Budget your AI features against today's per-token cost with no assumption of relief, and treat any price cut you do get as margin, not as a plan.
Capacity commitments are getting longer than your vendor relationships. Six-year supply agreements mean your model provider's cost structure is set by contracts signed before you picked them. That is an argument for portability — abstract your model calls behind one interface so switching providers is a config change, not a rewrite.
A beat that trades down is a signal about expectations, not execution. Marvell grew 37% and got punished for a calendar. Read that as the market marking AI infrastructure to a delivery schedule rather than a narrative. Apply the same discipline internally: when a vendor pitches you a roadmap, ask which quarter the thing you need actually ships.
Key takeaways
- Q2 FY2027 revenue $2.739B, up 37%; data center $2.17B, up 46%, at 79% of total
- FY2027 outlook raised to ~$12B, FY2028 to ~$18B
- The ~$120B Google deal spans six years; material revenue impact begins fiscal 2029
- Stock fell 7-8% on the timing despite the beat and the raise
- Plan AI feature economics at current token prices — the competing silicon is not shipping yet
Your AI costs should survive a vendor change. We build model-agnostic pipelines where the provider is a config value, so a price hike or a capacity crunch is a swap instead of a rebuild. Run the numbers on your automation, or see how we architect for portability.
Sources: Marvell investor relations, 24/7 Wall St..
- #marvell
- #ai-infrastructure
- #compute-costs
- #custom-silicon
- #vendor-risk
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Hidden HTML hijacked AI email summaries in 10 of 10 runs
Forcepoint X-Labs hid prompt injection in zero-font-size HTML. The AI email summary reported a EUR 46,200 invoice instead of the real one — every single run.
Read itTencent Hy4 preview: 770B open weights, Apache 2.0
Tencent open-sourced a 770B-parameter MoE flagship with a 1M context under Apache 2.0. The open-weights fallback in your cost model just got a lot more credible.
Read it