On Monday, semiconductor heavyweight Qualcomm (QCOM) surged by roughly 12% on the day following a major strategic announcement in the rapidly growing artificial-intelligence infrastructure market.
Qualcomm announced two new AI-inference-optimized chips, dubbed the AI200 (for 2026) and AI250 (for 2027). This breaks ground for Qualcomm on two fronts: The first being that inference-first chips are aimed at the processing power required to run AI models (answer queries, generate video, etc.) rather than train AI models (the GPU-intensive workload that’s been largely behind the Nvidia (NVDA) move over the last few years).
Second, it’s a departure from Qualcomm’s long-standing mobile-first mindset and represents the company’s first foray into the high-growth AI data center market.
[text_ad]
On top of the chip announcement, Qualcomm disclosed that they already have a customer lined up, with Saudi-based HUMAIN planning to deploy 200 megawatts (MW) of AI200/AI250 server racks starting in 2026.
For an established chipset and mobile-first company like Qualcomm, the shift into AI infrastructure is a big deal.
Qualcomm’s Shift to AI Infrastructure
Historically, its business has been tied heavily to smartphones and wireless connectivity, a space that is maturing and highly competitive. But with the surge in generative AI applications, data center expansion and AI deployment, Qualcomm is betting that the “next frontier” for semiconductors is increasingly in high-end inference and model deployment, and not just phone chips.
Of course, these developments don’t preclude Qualcomm from benefiting from growth in AI-enabled phone chips. Should models continue to evolve to the point that local AI (meaning running on your own hardware and not on a data center you interact with via the cloud) becomes more feasible for most consumers, the inference specialization should pay dividends for Qualcomm in that market as well.
In fact, the leap to data centers was partially fueled by experience developed on the humble Windows PC, as the company has cited its Hexagon neural processing unit (NPU) as enabling them to learn and scale for the new AI chips; there’s no reason the company can’t then take its data center expertise and redirect (and minify) it for next-generation smartphone chips.
Plus, by expanding into the high-demand data center space, Qualcomm gets the added benefit of (at least partially) decoupling itself from the smartphone upgrade cycle, something that Cabot’s Tom Hutchinson has written about in regards to Qualcomm before.
What’s particularly noteworthy about this development is how companies will be able to deploy these new chips, per Yahoo! Finance:
“According to Durga Malladi, senior vice president and GM for technology planning, edge solutions, and data centers at Qualcomm, they’ll be able to choose either individual chips, portions of the company’s server offerings, or the entire setup.
“And those customers, Malladi said, could include the likes of Nvidia and AMD, making the companies both rivals and potential partners.”
It’s an interesting development for sure, and the success or failure of the gambit will, as seems to so frequently be the case these days, hinge on the eventual uptake of AI tools themselves.
After all, these inference-first chips aren’t building new AI models, they’re delivering processing power for their actual use, meaning there’s less room for speculative buildout (it’s one thing to sell server capacity to hyperscalers as they develop bleeding-edge models, it’s another thing entirely to sell server capacity to a company that’s trying to commercialize those models once they exist).
As for the investing case, I’ll defer to Cabot’s resident Qualcomm expert, Tom Hutchinson, who laid out his bull case for Qualcomm last year. To learn the names of Tom’s newest big (dividend-paying) bets in the tech realm, I highly recommend subscribing to his Cabot Dividend Investor newsletter.
[author_ad]
