Marvell's 102.4 Tbps AI Switch Just Fired a Shot at Broadcom — And Wall Street Is Catching On

Here's something that should make you sit up straight: Marvell Technology quietly turned an $885 million loss into $2.67 billion in profit, won all three hyperscalers as customers, shipped the world's fastest AI switch, and then the market handed it an 8% selloff. That last part? Pure noise. And the smart money knows it.

The story starts with the Teralynx T100, announced June 1 — the industry's first 102.4 Tbps switch silicon built from the ground up for AI. Not adapted from some enterprise switch that was already collecting dust in the engineering backlog. Purpose-built. No legacy baggage. Lower power, lower latency, designed specifically for the bandwidth bottlenecks choking large AI clusters right now. When you're stitching tens of thousands of GPUs into a single training run, the network fabric isn't a nice-to-have — it's the difference between a model that trains in weeks and one that takes months. Marvell just made that fabric four times faster than anything else on the market.

Here's why that matters beyond the spec sheet. Marvell has been known as the custom ASIC player — the company that builds chips for Amazon's Trainium, Microsoft's Maia, and, as of April this year, Google's Axion and TPU programs. That last one? Marvell stole it from Broadcom. Directly. Google looked at both suppliers and chose Marvell. That's not just a win — it's a signal that the custom silicon market, where Broadcom has historically held a 70% share, is suddenly a two-horse race. And Marvell is the one gaining ground.

But the Teralynx T100 moves Marvell into entirely new territory. Now they're not just designing custom chips for the hyperscalers — they're competing with Broadcom in the networking layer too. The AI data center needs both: custom compute silicon and ultra-fast switch fabric to connect it all. Marvell just became the only company besides Broadcom that can credibly offer both. That's a duopoly narrative, and duopoly narratives tend to command premium valuations.

Wall Street is catching on. RBC Capital reiterated Outperform on July 7 with a $360 price target — 53% upside from where the stock sits today. BofA is at $365. UBS raised its target to $340. The consensus is 31 Buy ratings, 6 Holds, zero Sells. These aren't reflexive upgrades either. RBC's Srini Pajjuri sees 40%+ revenue growth sustained for three years, with data center revenue growing 50% or more this year and next. That's a compounder trading at 38x forward earnings — not cheap, but reasonable when you're growing revenue at 35% and actually turning a profit.

Because here's the part that doesn't get enough attention: Marvell is profitable now. GAAP net income of $2.67 billion in FY2026, compared to an $885 million loss the year before. Free cash flow of $2.27 billion over the trailing twelve months. Net debt of just $1.12 billion against a $211 billion market cap. This isn't some pre-revenue AI story burning cash on the promise of future compute. This is a real business that crossed over from losses to profit while quadrupling down on the single biggest technological shift of our lifetimes.

The 8% selloff on July 7-8 that everyone panicked about? Samsung earnings disappointed. DeepSeek chip rumors surfaced. Geopolitical jitters flared up. None of it was Marvell-specific. By July 9, the stock had already rebounded 5-7%. Sector rotation in a stock that's 28% below its 52-week high of $329.88 — that's not a thesis breaker. That's a gift for anyone paying attention.

Marvell sits at the intersection of three massive AI trends. Custom ASIC design for the hyperscalers, where it now has Amazon, Microsoft, and Google as customers plus a strategic partnership with Nvidia that included a $2 billion investment. AI networking via the Teralynx T100, a category expected to generate $600 million in revenue in FY2027 and over $1 billion in FY2028. And the broader data center buildout, which now makes up 74-76% of Marvell's total revenue at $6.1 billion growing 46% year-over-year.

Jensen Huang called Marvell the "next trillion-dollar company" at Computex. That might sound like hype until you look at the trajectory. A company that went from an $885 million loss to $2.67 billion in profit while winning every major AI customer on the planet. A stock 28% off its highs with a 53% upside target from the most recent analyst upgrade. A new product category that positions it as the only real alternative to Broadcom in AI networking. The pieces are all there. The market's just taking its time to connect the dots.

Disclosure: The Signal holds no position in MRVL. Positions may change. This is not financial advice.