Marvell Promised Wall Street $90 Billion by 2031. The Fine Print Is the Interesting Part.

The number

On Tuesday, at an investor day in New York, Marvell CEO Matt Murphy put a number on the wall that made the room go quiet before the applause: $70 billion to $90 billion in annual revenue by fiscal 2031. Marvell booked $8.2 billion in revenue in fiscal 2026, the year that ended in January. The target means growing roughly 55 to 60 percent a year for five straight years — a pace the company has never sustained, for a business analysts had modeled at roughly $47 billion by then.

The stock closed up 5.8 percent at $287.01. It's tripled this year, and now trades at about 50 times forward earnings. Wall Street either just bought into the most audacious growth story in semiconductors, or it priced in a promise that five years of flawless execution will have to honor. Both things can be true at once.

What Marvell actually said

Strip out the applause and the investor day laid down specifics. Fiscal 2028 revenue guidance goes to $20 billion, raised from $18 billion in August and above the $18.2 billion Wall Street consensus. Roughly $18 billion of that is expected to come from the data-center business alone. Custom chip revenue — silicon designed for a specific customer's AI systems rather than sold off the shelf — is now targeted above $12 billion in fiscal 2029, up from a prior target of $10 billion.

Further out, Marvell pegs the total AI addressable market at roughly $400 billion by 2030 and targets adjusted earnings above $30 a share in fiscal 2031, against $2.84 in fiscal 2026. The nearer-term numbers are concrete; the 2031 figures are a declaration of intent. That distinction matters, and Marvell's own CFO reportedly said the latest targets could prove conservative as custom-silicon programs with major hyperscalers ramp up — which is exactly what a confident management team says, and exactly what a management team with a 50-times-earnings multiple needs to be right about.

The thesis: the GPU era is going bespoke

Marvell's bet has one engine: custom silicon. The hyperscalers — Google, Amazon, Microsoft — are designing their own AI processors in-house so they're not renting Nvidia's margin forever. Marvell's strategy, set back in 2021 and centered on custom and cloud-optimized chips, has made it the prime contractor for that de-Nvidia-ification.

It named custom-silicon programs at all three major US cloud providers plus Meta. Google is the flagship relationship: Marvell granted Alphabet a warrant in August to buy nearly 59 million shares, most of it vesting in 240 tranches — one tranche for every $500 million Google spends on Marvell's custom chips, through fiscal 2033. Amazon got a smaller warrant last December tied to photonic fabric purchases. And in the most delicious irony of the whole saga, Nvidia itself invested $2 billion in Marvell in March, partnering on custom XPUs and scale-up networking for customers building semi-custom AI infrastructure. When the company that sells the GPUs invests in the company helping customers buy fewer of them, the hedging is no longer subtle.

Even Jensen Huang seemed to acknowledge it: at a public event in Taipei this June, he reportedly called Marvell "the next trillion-dollar company."

Read the fine print before the headlines

Here's the part of the story the headlines keep flattening: the $120 billion figure attached to the Google deal is not a purchase commitment. It's a vesting ceiling. Google earns shares by buying; if its buying falls materially short, most of the warrant simply expires in 2033 with the tranches unvested, and Marvell has given up far less equity than the $12.18 billion headline number suggests. Google, the Motley Fool's coverage noted dryly, "hasn't promised to buy that much." The $120 billion is a measure of how big the relationship could get, not a contract.

That's not a criticism — it's just how these instruments work, and understanding it is the difference between reading the news and understanding it. The honest read: Google put a mechanism in place that only rewards it if the silicon actually ships and sells. A tranche vesting on $500 million of qualifying purchases is Wall Street's way of saying "show me the revenue, not the roadmap." Of the projected $70–90 billion, Marvell expects roughly $30 billion at the midpoint to come from the custom business — the rest from interconnect and connectivity, the optical links that move data between AI servers, where demand for 1.6-terabit optics is already accelerating.

The catch

Now the counter-argument, because it deserves one clean paragraph. This target requires everything to go right: hyperscaler capex keeps climbing, custom programs convert from design wins into recognized revenue on schedule, Broadcom — which holds an estimated 80 to 85 percent of the custom ASIC market today — doesn't squeeze Marvell's share, and the AI buildout doesn't hit a demand or power-grid wall before 2031. The stock has run 250 percent this year; analysts note execution slips on hyperscaler ramps, "not just headline misses," could trigger outsized downside at this valuation.

President and COO Chris Koopmans told Barron's the company "has a track record of doing what we said we were going to do" — and to be fair, the Q2 numbers (revenue $2.739 billion, up 37 percent year over year; data center up 46 percent) back him. But a 55–60 percent CAGR target is not a track record. It's a dare.

Why this matters if you write software

None of this is really about Marvell's stock. It's about who designs the machines your code will run on. When the chip architecture stops being "buy Nvidia GPUs" and becomes "co-design accelerators with your cloud provider," the cost and performance of AI inference — the API bills, the latency, the model you can afford to run in production — start depending on procurement deals struck years in advance by people who never commit code.

The trend to watch isn't the $90 billion. It's the conversion rate: whether hyperscaler design wins turn into scheduled, recognized revenue. The Google warrant vesting pace will publish the answer quarterly, in public, without anyone having to speculate. If the tranches keep vesting, the bespoke-silicon era is real. If they stall, you'll have learned that before the next investor day — for free.

References

• Barron's — Marvell Stock Rises After CEO Announces Huge Long-Term Revenue Target
• Reuters (via WIXX) — Marvell raises 2028 revenue forecast on strong AI data center demand
• Motley Fool — Marvell Stock Has More Than Tripled This Year. I Think It Can Double Again by 2031.
• Motley Fool — Prediction: This Nvidia-Backed AI Stock Could 10X by 2033
• Terrain Labs — Nvidia's $108B Lease Guarantees, Marvell's Google Warrant (Aug 2026)
• TS2 — Marvell Shares Gain $12 Billion After Google Agreement Eases Pressure
• Talk Markets — Why Is Marvell Technology Stock Rallying On Tuesday?