Qualcomm has signed agreements to supply AI chips to three of the largest cloud-service providers, marking the company’s most serious entry into a market dominated by Nvidia, according to people familiar with the deals. The agreements cover low-power inference chips — the processors that run AI models once they are trained — a segment where Qualcomm argues its efficiency gives it a real advantage.
The deals are a bet on how AI computing evolves. Training models requires enormous, power-hungry clusters of the kind Nvidia dominates. Running those models in production, the inference stage, is different: it happens constantly, at massive scale, and the cost of electricity becomes the binding constraint. That is where Qualcomm says its expertise lies. The company has spent decades designing chips that squeeze performance out of limited power budgets, first for phones, now for servers.
Analysts said the agreements position Qualcomm as a credible alternative in the inference market, which Nvidia has not fully captured. “Nvidia owns training. Inference is still up for grabs, and power efficiency is the battleground,” one analyst said. Cloud providers are eager for alternatives, both to control costs and to avoid dependence on a single supplier whose products are in short supply. Qualcomm’s pitch — comparable performance at a fraction of the power draw — is designed for exactly those concerns.
The deals also represent a strategic pivot for a company long identified with smartphones. Qualcomm’s phone business has been its cash engine for decades, but the market has matured, and the company has been looking for its next act. Data-center AI chips would be the largest new market it has entered since the smartphone era began. Executives have described the AI inference opportunity in terms that suggest they intend to compete seriously, not merely to dabble.
The technical details matter. Qualcomm’s inference chips are built around its experience with efficiency: they are designed to run AI models with lower power consumption than the data-center chips of its rivals, which matters both for operating cost and for the heat and cooling that large deployments require. The company has said its products are suited to the smaller, distributed AI workloads that are multiplying across the industry — the recommendation systems, search functions and chatbots that companies run every day.
The competitive response from Nvidia has been to push its own efficiency improvements and to argue that its software ecosystem keeps customers locked in. Qualcomm’s answer is that the inference market is large enough for multiple winners and that its power advantage will decide a growing share of the business. Cloud providers, for their part, are treating the arrival of a new supplier as an opportunity to renegotiate terms across the board.
The bigger question is whether Qualcomm can convert these agreements into a durable business. Signing three cloud customers matters in an ordinary sense — it validates the product — but the revenue will take time to build, and the company will be measured against expectations set by its new position. If the inference market grows as forecast, Qualcomm has a chance to become what analysts have begun calling it: the underrated AI competitor. If the market consolidates around Nvidia’s platform, the agreements will matter less than the momentum behind them. For now, Qualcomm is in the game, and the cloud providers have a reason to keep it there.
The timing of the deals coincides with a broader shift in how cloud providers buy chips. For years, the largest clouds designed their own custom processors or bought almost exclusively from Nvidia; now they are spreading purchases across multiple suppliers, both to manage supply risk and to bring down the cost of inference at scale. Qualcomm’s entry gives them another option, and its efficiency claims give procurement teams a concrete basis for comparison.
Qualcomm’s internal reorganization has been preparing for this moment. The company has reallocated engineering resources toward data center products, hired executives with server-industry backgrounds and said publicly that AI inference is its biggest growth opportunity. The agreements with the cloud providers are the first evidence that those preparations are paying off. The company’s shares have responded, with investors beginning to price in a future in which Qualcomm is more than a phone-chip maker.
The engineering challenge should not be understated. Data-center customers demand reliability, software support and performance at scale — qualities that take years to build and that Nvidia has perfected through a decade of dominant market share. Qualcomm’s advantage in power efficiency is real, but it must be paired with the software tooling and integration support that cloud providers expect. The company has said it is investing in both, and its new customers will be the judges.
The deals also have implications beyond the three named clouds. If Qualcomm’s inference chips perform in production, other buyers — from enterprise data centers to the companies building specialized AI appliances — become available markets. The inference segment is expected to grow faster than training over the coming years as deployed AI applications multiply. Qualcomm’s position at the start of that growth, with signed customers in hand, is the strongest hand the company has held in the data center market since it began trying to enter it.
