Qualcomm: From Pocket to Data Centre

Qualcomm's entire history is a masterclass in doing more with less power, the discipline required to run advanced computing on a phone without melting it or draining the battery by lunchtime. For years that looked like a niche strength. Now it is the central problem in AI, where data centres are constrained by how much electricity they can draw and cool. Qualcomm is pitching its new Dragonfly data-centre chips squarely on this point, claiming far better performance per watt than rival server processors. If "tokens per watt" is the metric that matters, Qualcomm is built for it in a way few competitors are.

For two years the question was whether Qualcomm could win even one major cloud customer. At its 2026 investor day it answered emphatically: Meta signed a multi-generation deal for its new server CPU, Microsoft committed to its memory technology, and two further hyperscalers signed custom-silicon agreements, each projected to generate over $1 billion. Management now guides data-centre revenue from almost nothing today to roughly $5 billion in 2027 and over $15 billion by 2029. The strategic logic is telling: companies that design their own chips do not quietly buy someone else's unless that supplier offers something they cannot build themselves.

The data centre is the headline, but it sits alongside two other growth engines. Automotive has hit record revenue, crossing a $5 billion annual run-rate for the first time, backed by a $65 billion pipeline of design wins with carmakers from Volkswagen to Mercedes, and is targeted at $10 billion by 2029. The "internet of things" arm, spanning robotics, smart glasses, and AI laptops, adds more. Crucially, the existing business throws off enough cash to fund all this while still returning capital, with a fresh $20 billion buyback and a raised dividend. Handsets are projected to shrink to about a third of chip revenue.

• Data-centre revenue ramping: Hitting the roughly $5 billion 2027 data-centre target would turn a speculative catalyst into a quantified growth engine.

• On-device AI and the OpenAI tie-up: Confirmation as a lead chip partner for new AI-native devices positions Qualcomm at the centre of agentic AI on phones.

• Delivering the 2029 roadmap: Reaching $40 billion in non-handset revenue and over $18 in earnings per share would cement the shift from cyclical to compounder.

• Closing the valuation gap: Trading at a fraction of Nvidia's and AMD's multiples, a credible re-rating toward peers is the prize if execution holds.

• Early hyperscaler shipments: First custom-silicon deliveries to a major cloud customer are due later this year, the first hard proof the data-centre story is real.

• China handset recovery: Management expects Chinese phone demand to bottom and recover, which would lift the part of the business the market has written off.

Signing four hyperscalers is one thing; building a $15 billion cloud-infrastructure business by 2029 is another. The chips underpinning that target largely do not ship in volume until 2028, so investors are underwriting a roadmap, not current sales. Qualcomm is entering a market dominated by entrenched incumbents with deep customer relationships and mature software. If the revenue ramp disappoints against these ambitious targets, the re-rating could stall and the stock risks handing back its investor-day gains, as some of it did within days of the announcement.

The bear case the market has clung to is real. Apple, once a meaningful modem customer, is moving to its own in-house chips and is expected to fall to zero by 2027. The Chinese Android market has been soft, and global smartphone shipments face one of their steepest projected declines on record. The whole thesis depends on new revenue from data centres, automotive, and IoT arriving fast enough to outrun these losses. If the timing slips, the company could de-rate further before the new engines take over.

Qualcomm is the newcomer in AI infrastructure. Nvidia dominates and is pushing hard into the inference market Qualcomm is targeting, AMD competes aggressively on price, and the hyperscalers themselves are building their own chips, a "build versus buy" threat that could limit demand. Qualcomm's answer is a roughly $4 billion acquisition of software firm Modular, meant to offer an alternative to Nvidia's sticky software ecosystem. But integrating that deal while standing up three new businesses at once is itself a meaningful execution risk.