Marvell Raised Its 2028 AI Forecast to $20 Billion. Only 25% of the S&P 500 Is Confirming the Rally.

Marvell Technology's stock climbed as much as 10% in early trading on October 6, its sharpest intraday move in months, before closing up roughly 4%, after the chipmaker told investors at its annual Investor Day that fiscal 2028 revenue will land around $20 billion. That's up from the $18 billion guidance it gave only weeks earlier, and comfortably ahead of the roughly $18.2 billion Wall Street had modeled going into the event.

The headline number matters less than what's actually driving it. Marvell's data center business, already close to 80% of total revenue, isn't accelerating mainly because of the custom AI chips most investors assume. CEO Matt Murphy told analysts that most of the fiscal 2028 increase traces back to connectivity products instead: the 1.6-terabit optical signal processors and scale-up optics that move data between racks of GPUs rather than inside them. "Connectivity is the constraint on AI infrastructure," Murphy said, a framing that shifts where the next bottleneck in the AI buildout actually sits.

[[IMG1]]

What Changed

Marvell wasn't the only company flagging insatiable compute demand this week. AMD climbed nearly 3% the same day, extending a run to fresh highs after CEO Lisa Su said AI chip and CPU demand keeps outpacing supply and that AMD is moving to secure more manufacturing capacity from its Taiwanese partners, the same dynamic that's helped AMD claw server market share away from Intel over the past year. Nvidia, meanwhile, closed within reach of a $6 trillion market cap, a level no public company has approached before. Three different companies, three different parts of the AI supply chain, and the same underlying signal: whatever skepticism investors voice about an AI bubble, hyperscaler spending commitments keep arriving faster than suppliers can fill them.

That long-term target captures the scale of the bet Marvell is making. Management now expects $70 billion to $90 billion in annual revenue by fiscal 2031, roughly four times this year's run rate, with $12 billion of that coming from custom silicon for cloud customers by fiscal 2029 alone. The company pointed to an estimated $400 billion AI infrastructure market opportunity by 2030 as the backdrop, a figure that's becoming something close to a shared assumption among chip and networking suppliers chasing the same build-out.

Why It Matters

The S&P 500 and Nasdaq both closed at record highs on October 6, extending a rally built heavily on names tied to that same AI infrastructure story. Yet by one widely watched gauge, only about a quarter of S&P 500 constituents were trading above their 50-day moving average that same day. The index keeps setting records. Most of the companies inside it aren't participating in them.

That gap matters because narrow leadership tends to be fragile leadership. When gains concentrate in a handful of mega-cap names, a stumble at any one of them (a disappointing earnings call, a delayed chip shipment, a financing wobble at one of the hyperscalers footing these bills) does more damage to the index than it would in a broader rally. Markets have been here before, in 2000 and again in stretches of 2023 and 2024, and the pattern rarely resolves with a calm rotation into the laggards.

It also sits awkwardly next to a separate claim Wall Street has been making all year, that AI capex is finally broadening earnings growth across every sector of the S&P 500. Broader profits and narrower stock prices aren't the same thing, and October 6 is a reminder of how far apart those two trends currently sit. Marvell's guidance doesn't resolve that tension so much as sharpen it. The demand it describes is real, and the company says its backlog visibility now extends into 2029. But real demand concentrated among a handful of suppliers selling to a handful of buyers is still concentration, and concentration is exactly what a 25% breadth reading captures, even on a day the headlines call a record.

What to Watch Next

A few things worth tracking from here. Whether other networking and optical suppliers confirm Marvell's framing when they report would suggest the bottleneck really has shifted from compute itself to the wiring connecting it. Market breadth is worth watching alongside the index level too: a recovery toward 40% or 50% of stocks above their 50-day average would suggest the rally is broadening out, while breadth stuck near current levels into year-end would be a warning sign independent of what any single company reports. And the 10-year Treasury yield matters as much as either; it eased to 5.28% on October 6 after touching a 24-year high of 5.31% the day before, and elevated long-term yields raise the discount rate applied to the distant cash flows these AI infrastructure bets depend on, a calculation that gets less forgiving the longer yields stay this high.

The Pulse24 Take

The AI infrastructure trade keeps clearing a bar that rises every quarter, and Marvell's $20 billion forecast is the latest evidence that hyperscaler spending hasn't slowed down. What's changed is where in the stack that spending lands. A year ago, this story was almost entirely about GPUs. Now it includes the optics and switching gear moving data between them too, which looks more like a buildout maturing than one stalling.

That breadth problem deserves more attention than the headline gives it. A market that sets records on the strength of three or four stocks isn't necessarily wrong to do so, but it is staking a lot on those stocks staying right. Nvidia nearing $6 trillion and Marvell's upgraded targets both argue the AI capex cycle has room left to run. A 25% breadth reading argues most of the market isn't convinced it benefits either way. Both can be true at the same time, and sitting with that tension is more useful than resolving it before the evidence does.