The number that mattered in Qualcomm’s Q3 FY2025 report wasn’t the $10.4 billion top line. It was $984 million, the automotive segment, a record, up 21% year over year. Put IoT’s $1.68 billion next to it and Qualcomm cleared $2.7 billion in non-handset QCT revenue growing around 23%, while the handset business everyone still thinks defines this company crawled at 7%. The phone money is bigger. The phone money is also boring now, and Qualcomm knows it.
The timing is what made the July 30 report worth reading. It landed with a stack of design-win confirmations that read like a thesis statement. Samsung put Snapdragon 8 Elite for Galaxy in the Z Fold7 globally, no Exynos variant anywhere, the first time the regional split that dogged the foldable line just vanished. Xiaomi signed a multi-year agreement that reaches well past phones. BMW’s Neue Klasse picked up Qualcomm’s ADAS stack, the most complicated automotive deployment the company has ever shipped. None of it is press-release filler. It is the diversification story converting into contracts.
The Neue Klasse is BMW’s next-generation vehicle architecture, with first production cars arriving across 2025 and 2026, and Qualcomm isn’t just selling silicon into it. The ADAS software stack hit ISO 26262 ASIL-D certification, the highest functional-safety integrity level the automotive standard defines. ASIL-D demands a random hardware failure probability below 10^-8 per hour on the safety-critical paths, and you cannot get type approval for highway or urban autopilot in Europe without demonstrating exactly that. The compute runs on Snapdragon Ride Flex, a heterogeneous SoC that pulls ASIL-D-capable Cortex-A78AE clusters, the Hexagon DSP for sensor fusion, an Adreno GPU for visualization, the AI Engine NPU for perception, and a lockstep Cortex-R safety island onto one die, handling up to 12 camera inputs alongside radar, lidar, and ultrasonic fusion. Configurations land in the 30 to 60 TOPS range for perception.
The co-developed driving stack is the real role change. BMW and Qualcomm built the software together, which means Qualcomm owns the middleware and the hardware abstraction layer while BMW differentiates at the application layer. It is the Android automotive playbook applied to autonomy: the OEM keeps the UX, the platform vendor owns the compute and connectivity substrate underneath. Because Neue Klasse launches globally instead of as a regional pilot, that stack has to satisfy ISO 26262, SOTIF, UN ECE R157 for automated lane keeping in the EU, and FMVSS in the US all at once. That is a lot of regulatory surface for one platform, and it is the kind of thing that makes a win sticky once it lands.
Those 12 new design wins booked in Q3 will not pay a single dollar of revenue for three to five years. Automotive recognizes at production launch, not at signature. The 50 vehicle launches Qualcomm is counting for FY2025 are wins from roughly 2020 through 2022 finally ramping. So when Cristiano Amon cites a pipeline around $45 billion in total contract value, he is describing revenue largely locked for FY2027 and FY2028 by deals already signed, while today’s wins quietly build the FY2028-to-FY2030 base. The business runs on a four-to-five-year delay between handshake and money, which is great for visibility and brutal if a program gets cancelled.
The 21% looks like a slowdown next to the roughly 81% Qualcomm posted a year earlier, when automotive jumped from about $447 million to about $811 million. It isn’t. The absolute dollar add this year was $173 million on a much bigger base. The segment got large enough that the percentages have to compress even as the raw dollars keep stacking. Anyone reading deceleration into that number is reading the wrong column.
IoT is the messier of the two non-handset lines, and the one I find more interesting, because it is where Qualcomm’s edge-AI thesis turns into actual silicon. The $1.68 billion bundles AI PCs, XR, industrial edge, networking, wearables, and consumer smart-home gear, and the 24% growth came mostly off AI PC ramp and XR volume. The Snapdragon X Elite is the anchor: a 12-core Oryon CPU on TSMC N4P, four prime cores up to 3.8 GHz with eight efficiency cores, an Adreno GPU at 4.6 TFLOPS FP32, and a 45 TOPS Hexagon NPU at INT8 that buried Intel Meteor Lake’s 11 TOPS and AMD Hawk Point’s 16 TOPS at launch. That headroom is what lets it run Llama 3 8B at roughly 20 to 30 tokens per second or Phi-3 Mini around 40 to 60, on-device, no cloud round trip. The silicon was never the problem with Windows on Arm. The app compatibility and the emulation tax were, and that is a software story Qualcomm has been chipping at for two years.
The Xiaomi side of IoT matters too, because the multi-year deal isn’t a phone deal wearing a trenchcoat. It covers AR glasses on Snapdragon AR2 Gen 1, a three-chip design (main SoC, connectivity SoC, always-on SoC) on 4nm with a 12 TOPS NPU tuned to stay under a watt average in a glasses form factor. Wearables, tablets, and Xiaomi’s SU7 cockpit platform all sit inside the scope. That is the version of platform Qualcomm keeps selling, and with Xiaomi it is now contractual across half a dozen device categories.
On the handset line itself, the $6.33 billion and 7% growth tell you what the smartphone market is. Global units are tracking around 1.24 billion in 2025, a recovery off the 2022-to-2023 trough but nobody’s idea of a growth market. The 7% is premium-mix ASP lift from Snapdragon 8 Elite and Android volume from Xiaomi, OPPO, vivo, and Honor, set against the slow bleed of Apple’s in-house modem starting to eat into Qualcomm’s biggest single account.
Samsung dropped Exynos entirely and went global on Snapdragon 8 Elite for Galaxy in the Z Fold7. Oryon V3 with two prime cores at 4.47 GHz and six performance cores at 3.53 GHz, an Adreno 830 about 45% faster than the 8 Gen 3, the same 45 TOPS Hexagon NPU, a Spectra ISP good for 320MP single-camera capture and 4K60 HDR, and an X80 modem peaking at 10 Gbps downlink on N3E. The “for Galaxy” designation isn’t cosmetic. Samsung gets custom clocks and thermal profiles tuned for the foldable’s hinge, where heat dissipation is constrained and the SoC sits in a tighter sustained envelope than a slab phone. Foldables are projected at 25 to 30 million units in 2025 with Samsung holding 60 to 65% of that, so a global win captures the entire Samsung foldable volume instead of the 40 to 50% Snapdragon used to split with Exynos. It also quietly says Samsung’s own Exynos roadmap couldn’t keep pace at 3nm in the category where thermal density bites hardest.
DIGITIMES Research had MediaTek at 37% of the 2024 smartphone AP market against Qualcomm’s 27%, and people treat that as Qualcomm losing. It isn’t, because volume share and revenue share are different things. MediaTek’s lead sits on the Dimensity 7000 and 6000 series in the $150 to $400 band, while Qualcomm’s 27% sits in the $500-plus premium tier where a Snapdragon 8 Elite carries an estimated $80 to $120 ASP against a Dimensity 9400’s $60 to $90. Run the math and Qualcomm’s 27% of units is closer to 35 to 40% of AP revenue. The volume crown is real, and it doesn’t touch the part of the market Qualcomm cares about.
MediaTek announced its first cloud AI inference ASIC in 2025, walking straight into the hyperscaler market where Nvidia, AMD, and Google’s TPU already live. Qualcomm went the other way, betting on inference at the edge, across automotive ADAS and AI PCs, with the AI Hub ecosystem deploying models onto Snapdragon hardware. One company is chasing the data center, the other is chasing everything that runs on a battery. Amon’s framing of Qualcomm as the industry platform of choice as AI scales at the edge is corporate, but the technical logic holds up better than the slide deck suggests. Nvidia’s H100 and H200 pull 300 to 700 watts, a non-starter in a phone or a pair of glasses, and Qualcomm’s modem integration plus its ASIL-D automotive certification give it a stack that is awkward for anyone else to replicate quickly.
Guidance for Q4 came in at $10.3 to $11.1 billion with non-GAAP EPS of $2.75 to $2.95, putting full-year EPS growth around 16% over FY2024. The unusually wide $800 million range isn’t about automotive or IoT, both of which Qualcomm forecasts off known production schedules. It is about China handset sell-through and the inventory dance with Xiaomi, OPPO, and vivo. The 30% QCT EBT margin sits at the high end of Qualcomm’s historical band, propped up by premium mix and the fatter gross margins automotive and IoT SoCs carry over mid-range handset parts.
Qualcomm’s Apple modem revenue ran somewhere around $7 to $8 billion in FY2024 across modem supply and royalties, and the C1 modem in the iPhone 16e is the first chunk of that walking out the door, with the iPhone 17 lineup expected to take more. Estimates put the decline at $1 to $2 billion a year through FY2027. The Xiaomi agreement and the Samsung global win are, bluntly, the replacement plan, locking premium-tier volume to backfill the hole. The offset isn’t a clean one-for-one, because Apple modem ASP and Xiaomi SoC ASP aren’t the same animal, but it is the strategy in plain sight. I’m a declared Snapdragon Insider, so weigh my read accordingly. The structural logic here doesn’t need any loyalty to Qualcomm to follow.
I buy the diversification thesis, with one reservation I can’t shake. The whole automotive story rests on lead times Qualcomm doesn’t fully control, and BMW, flagship win or not, ships modest volumes against the mass market. The real automotive prize is the Chinese EV makers, BYD, Geely, Li Auto, NIO, and the rest, and that is precisely where Qualcomm is squeezed between Nvidia Drive and a growing pile of domestic Chinese silicon that doesn’t need an export license to exist. The record automotive quarter is real and the wins are real, but the part of the pipeline that decides whether this becomes a $10-billion automotive business or a nice $4-billion one is the part Qualcomm controls least. I’d bet on Amon over the skeptics. I’d just keep the receipt.
Related reading: the Qualcomm-Xiaomi multi-year deal that anchors this IoT and AR-glasses line.