Qualcomm Pivots Beyond Handsets with Data Center and Silicon Diversification

By The Numbers

• Non-GAAP revenue: $10.44B, down 7% YoY (GAAP $10.32B, down 8%); at the top of guidance and ahead of the $10.1B consensus (Bloomberg, via Quartz)
• Non-GAAP EPS: $2.17, down 28% from $3.00; in line with the $2.16 consensus (Bloomberg, via Quartz)
• QCT: $9.02B, down 8% YoY; EBT margin 24% versus 29% a year ago
• QCT Handsets: $5.14B, down 26% YoY; below the $5.21B StreetAccount estimate
• QCT Automotive: record $1.99B, up 89% YoY
• QCT IoT: $1.89B, up 4% YoY
• QTL: $1.30B, down 7% YoY; EBT margin 70%
• Fiscal 2026: non-GAAP revenue $43.24B (down 2%), non-GAAP EPS $10.54 (down 12%); GAAP EPS of $12.70 includes a $5.7B tax benefit from a valuation allowance release
• Q1 FY27 guide: revenue $10.4B to $11.2B; non-GAAP EPS $2.15 to $2.35; QCT $9.0B to $9.6B, now inclusive of Data Center; QTL $1.3B to $1.5B. Revenue midpoint above the $10.55B consensus and EPS midpoint below the $2.32 consensus (Bloomberg, via Quartz)
• Q1 segment detail: Data Center above $650M; QCT EBT margin 20% to 22%, including about 3 points from moving Data Center into QCT and investing ahead of revenue; handsets down mid-single digits sequentially
• Capital return: about $2.0B in Q4 (roughly $1.0B each in buybacks and dividends); $11.7B for fiscal 2026
• Balance sheet: inventory of $8.64B versus $6.53B a year earlier; fiscal 2026 capital expenditures of $2.49B versus $1.19B in fiscal 2025

The News

Qualcomm reported fiscal Q4 2026 results on November 4, with non-GAAP revenue landing at the top of its guidance range and EPS in line with consensus. The revenue beat came despite handset revenue falling by roughly a quarter year over year, as memory cost inflation shrank the smartphone market and the Apple modem roll-off continued. The key signal is a December-quarter guide that folds in the first meaningful data center revenue but carries an EPS midpoint below the Street. Read the full release inQualcomm's fiscal Q4 2026 earnings announcement.

Analyst Take

Qualcomm entered this call wanting to declare three overhangs closed: licensing renewals, the Apple exit, and the memory shock to handset volumes. The print suggests two of the three are behind it. The QTL renewals with Apple, Huawei, and Transsion held the revenue run rate, and management indicated the next renewal cycle does not begin until fiscal 2030. The Apple roll-off is now a known glide path rather than a guess. The market fixated on an EPS midpoint below consensus. We think that shortfall is the least surprising part of the print. Moving Data Center into QCT explains the lower QCT margin guide, but management said the change leaves company-wide guidance untouched, so it does not explain the EPS gap.

That gap traces to spending ahead of data center revenue and the lower gross margins that come with custom silicon, both of which management signaled last quarter. The bear case sits elsewhere. On the call, Bernstein argued the handset market appears to be exiting the year down more than 20%, which makes management's flat-to-down calendar 2027 planning assumption look generous. Management's response, thin channel inventory and price increases attached mainly to new platforms rather than existing phones, which limits pre-buying, is credible. It is not yet proven.

By leveraging its Oryon CPU architecture and deep packaging expertise across custom hyperscaler ASICs, Qualcomm is transforming its revenue profile from a cyclical smartphone vendor into a high-margin data center infrastructure provider. From our viewpoint, its strategic pivot to co-packaged optics with key players such as Lumentum and Corning positions the company to capture critical interconnect bottlenecks inside AI clusters, establishing an intellectual property moat reminiscent of its legacy RF front-end domination. This enterprise diversification, bolstered by recurring automotive design-win conversions and long-term licensing certainty through 2030, insulates the company’s operating margins against ongoing memory bill-of-materials inflation and the predictable phased roll-off of Apple modem volumes.

What the Numbers Mean

The most useful color came from CFO Akash Palkhiwala's gross margin framework. Management indicated that price increases begin contributing in the March quarter and should return the baseline business to the low end of its historical 48% to 50% gross margin range in the second half of fiscal 2027, with data center custom silicon then pulling the blended figure lower while adding to operating margin. That is an honest trade, and it was stated plainly.

The non-handset growth claim deserves arithmetic. Data center sat outside QCT in fiscal 2026 with no reported revenue line. So if automotive, IoT, and data center together grow more than 60%, and data center lands at the $5B target, automotive and IoT need to grow at least about 20% on their own. Healthy. But the headline leans on two hyperscaler custom silicon programs, one of them Amazon. Management called the at-the-money warrants in the Amazon agreement an exception and said future data center deals should follow traditional semiconductor terms, weighed case by case. We will watch whether that holds once other hyperscalers ask for the same equity sweetener. Management also framed upside beyond $5B as supply-limited rather than demand-limited, which for a company exiting a demand-constrained handset year is a notable reversal of its usual problem.

Connectivity is the next-largest data center contributor, and the more interesting long-term thread. Qualcomm added a second optical connectivity customer and demonstrated co-packaged optics with Lumentum and Corning at ECOC. Amon compared the coming restructuring of the optics industry to what happened in RF front end and pointed to conversations with laser companies about strategic partnerships. Qualcomm has run that play before.

The SpaceX agreement adds a different lever. Qualcomm will supply radio silicon including massive MIMO, digital-unit acceleration, and Layer 1 software for Starlink Mobile's nationwide terrestrial 5G network. Deployment starts in 2027, with revenue landing mostly in fiscal 2028. For Dragonwing, that is a meaningful entry into US radio access networks, a market long served by dedicated equipment.

Automotive appears structural rather than pulled forward. Amon tied the record quarter to content gains, citing roughly eight times more silicon in fifth-generation platforms than in third-generation vehicles, and to share converting from the design pipeline.

On handsets, the Samsung commentary may matter more than the guide. Management reset its planning baseline at Samsung from roughly half of flagship sockets to north of 70%, and is expanding into Galaxy wearables, which partially cushions the Apple roll-off. The longer argument is content per device. Amon cited industry goals of running 100B-parameter models locally on phones by 2028, a step change in on-device compute that favors the premium tier, where Qualcomm's Android share concentrates.

The balance sheet carries one item the call did not dwell on. Inventory rose by roughly a third year over year, and capital expenditures roughly doubled, to $2.49B in fiscal 2026 from $1.19B in fiscal 2025. We read the build as deliberate positioning ahead of input-cost inflation and the data center ramp, though a softer handset year would make that read less charitable.

Market Analysis

The September-quarter constellation tells a consistent story about who is paying for the memory supercycle. Memory has become the landlord of the smartphone bill of materials, and it raised the rent mid-lease. Samsung offers the cleanest illustration, posting record AI-driven memory profits while its mobile and networks businesses ran an operating loss on higher component costs. The same company is collecting the rent and paying it.

Apple sits at the other end. iPhone revenue grew 17% to a September-quarter record despite supply constraints and currency headwinds, with strong double-digit growth in the US and Greater China. That suggests the premium tier absorbs cost inflation far better than the mass market. It is mildly encouraging for Qualcomm, whose Android exposure skews premium, and less so for anyone selling into entry-tier volume.

Arm's print echoes Qualcomm's handset read. Royalty growth slowed to 8%, with smartphone memory pricing offsetting a richer Armv9 and CSS mix, while edge AI royalties ran roughly flat and data center royalties doubled. The two reported the same afternoon, with Qualcomm citing a jury finding that Arm breached its architecture license agreement, which is about as much as they share these days. Qualcomm said the pending remedies phase carries no implications for its Oryon CPU roadmap, and it treats any CPU customer beyond Meta as upside to its data center targets rather than a dependency.

MediaTek is the more instructive comparison, and the less comfortable one. It posted record revenue, up 18% year over year, began shipping its first AI accelerator in September, and expects about $2.5B from that program in the December quarter alone, with 2027 tracking toward the high end of a roughly $16B target. MediaTek also flagged gross margin dilution from data center against operating accretion, and raised prices on its non-data-center lines. The playbook rhymes. The scale does not, yet. Qualcomm's case rests less on matching program for program and more on breadth across connectivity, High Bandwidth Compute, and Oryon CPUs layered on top of its custom work.

Looking Ahead

The key trend we'll be monitoring is the shape of the data center ramp across fiscal 2027. Moving from a first quarter above $650M to a $7B exit run rate implies a fiscal fourth quarter of roughly $1.75B, about 2.7 times the first-quarter floor, gated by supply rather than orders if management's framing holds. That cadence will be visible quarter by quarter, and it will likely shape how investors read the Qualcomm story more than any handset beat. Broadcom's fiscal fourth quarter, which ends November 1 with results expected around December 10 though not yet confirmed, and Marvell's October-quarter print are not apples-to-apples with September. Still, they should help indicate whether hyperscaler custom silicon demand is broadening or concentrating. We will also watch whether the High Bandwidth Compute results, which management said are beating their power-efficiency design target and gaining traction for decode workloads, convert into a named merchant accelerator customer before the fiscal second half. Moving from slides to silicon is where accelerator narratives tend to either harden or quietly soften.