Two of the most important custom chip makers in AI just delivered back-to-back blowout quarters, and the market is taking notice. The thesis is simple: custom AI accelerators sold by Broadcom and Marvell are believed to be better suited for inference workloads, while training remains NVIDIA’s domain. And with inference demand just beginning to scale, the numbers are getting very big, very fast.
Marvell’s Record Quarter and a Massive Outlook
Marvell Technology (NASDAQ:MRVL | MRVL Price Prediction) posted a record quarter, with revenue of $2.075 billion, up 37% year-over-year, driven almost entirely by its data center business. Data center revenue hit $1.518 billion, representing 73% of total sales and growing 38% year-over-year.
But it’s the forward numbers that are truly jaw-dropping. Marvell should exceed $3 billion in near-term sales, with next year’s sales approaching $11 billion in 2028. Marvell is talking about sales of $15 billion versus the Street’s $13 billion estimate, with EPS well over $5. That’s not a modest beat on guidance. That’s a fundamental re-rating of what this business can become.
CEO Matt Murphy didn’t mince words:
“Looking ahead, we see demand for our products continuing to accelerate, and as a result, our data center revenue growth forecast for next year is now higher than prior expectations.”
Murphy also announced the acquisition of Celestial AI, calling it “a transformational milestone that accelerates our scale-up roadmap for interconnect.” Optical interconnect is the next bottleneck in AI datacenters, and Marvell is moving to own it.
Broadcom’s AI Revenue Is Compounding at an Extraordinary Rate
Broadcom (NASDAQ:AVGO) reported Q1 FY2026 revenue of $19.31 billion, up 29.5% year-over-year, with AI revenue of $8.4 billion growing 106% year-over-year. CEO Hock Tan guided Q2 AI semiconductor revenue to $10.7 billion, with total Q2 revenue expected around $22 billion, representing a 47% year-over-year increase.
That sequential jump from $8.4 billion to $10.7 billion in a single quarter is the kind of acceleration that rewrites analyst models. Broadcom also authorized a new $10 billion share repurchase program through December 31, 2026, returning $10.9 billion to shareholders in Q1 alone.
The NVIDIA GTC Wildcard
NVIDIA (NASDAQ:NVDA) isn’t sitting still. Jensen Huang called Grace Blackwell “the king of inference today,” and at GTC later this month, NVIDIA is expected to unveil a new chip aimed specifically at the inference market. Prediction markets put the probability of Huang mentioning “Vera Rubin” at 96% during the keynote, suggesting a major next-generation architecture reveal is coming.
If NVIDIA can close the inference gap with a purpose-built chip, the competitive dynamics shift meaningfully. That’s the risk Marvell and Broadcom investors need to sit with heading into March 16.
For now, the numbers from both Marvell and Broadcom tell a story of explosive, accelerating demand for custom inference silicon. The GTC keynote is the next test of whether NVIDIA can reclaim that narrative or whether the custom accelerator wave has already become unstoppable.
Top Gaining Stocks
Top Losing Stocks