Arm’s Custom Chip Push Quietly Corners Qualcomm’s Mobile AI Hold

The Quiet Power Grab in Mobile Silicon

Arm Holdings is not building chips. It never has. But right now, Arm’s business model is doing something far more aggressive than manufacturing – it is quietly pulling the rug from under Qualcomm’s most profitable floor.

How Arm’s Licensing Strategy Became a Competitive Weapon

For years, the relationship between Arm and Qualcomm looked like a clean partnership: Arm licenses its instruction set architecture, Qualcomm builds Snapdragon chips on top of it, and everyone wins. That arrangement has not collapsed, but it has become deeply complicated. Arm’s decision to push its own compute subsystems – pre-designed, pre-validated chip blueprints that go far beyond basic architecture licenses – means that smartphone makers no longer have to depend on Qualcomm’s engineering teams to get a competitive mobile AI processor.

The core of this shift is Arm’s Compute Subsystem program, which bundles together CPU clusters, GPU configurations, and memory interfaces into ready-to-implement packages. A phone maker with enough capital and manufacturing access can license directly from Arm, hand the design to a foundry like TSMC or Samsung, and ship a custom chip without Qualcomm ever entering the conversation. That was not a realistic path for most companies five years ago. Now it is, and several major device makers are actively walking it.

Apple blazed this trail years ago with its A-series chips, but Apple’s approach required building a massive in-house silicon team. What Arm’s compute subsystem packaging does is lower that barrier significantly. A company does not need Apple-scale engineering resources to produce a differentiated chip anymore. The subsystem handles the hardest integration work. What remains – tuning for AI workloads, thermal management, power efficiency – is demanding but no longer prohibitively so for a well-resourced phone maker.

Qualcomm has been aware of this dynamic and has pushed back hard. The company spent years developing its own custom CPU cores under the Oryon brand, explicitly to distance its silicon from vanilla Arm designs. Oryon was Qualcomm’s proof that it could out-engineer the reference architecture, not just customize it. The Snapdragon X Elite chips for laptops were the first showcase, delivering performance numbers that forced genuine comparisons with Apple silicon. That credibility matters, because Qualcomm’s entire premium pricing argument rests on the claim that its engineering adds value that off-the-shelf Arm designs cannot match.

Mobile AI Is the Specific Battleground

The fight is not about general smartphone performance anymore. It is specifically about on-device AI inference – the ability to run large language models, image generation, and real-time translation locally on a phone without cloud connectivity. This is where Qualcomm has staked its near-term identity. The Hexagon NPU inside Snapdragon chips has been Qualcomm’s primary marketing vehicle for the past two years, with the company positioning itself as the default choice for any Android manufacturer that wants to market AI features seriously.

Arm’s answer is the Ethos NPU series, which ships as part of its compute subsystems. The gap between Ethos and Hexagon in raw AI performance has narrowed enough that it is no longer a conversation stopper. For mainstream Android price points – the tier where most global volume actually lives – an Arm-based custom chip with an Ethos NPU can run the AI workloads that matter to consumers: on-device voice assistants, camera scene processing, photo enhancement. Qualcomm’s performance lead is real at the high end, but high-end flagship volume is a small fraction of global smartphone shipments.

MediaTek has already demonstrated this playbook is viable. Its Dimensity lineup is built heavily on Arm’s reference designs, and MediaTek has taken meaningful market share from Qualcomm in the mid-range Android segment. The company’s AI-focused marketing around its APU (AI Processing Unit) has not matched Qualcomm’s brand volume, but it has not needed to – the chips have sold on price-performance, and phone makers have followed. Samsung’s decision to use Dimensity chips in some Galaxy A-series models rather than Snapdragon was a signal the industry registered.

What makes the current moment different is that Arm itself is taking a more active role in shaping where its compute subsystems go and how they are marketed. The company has been more vocal about total solution performance, publishing reference benchmarks for its subsystem configurations and making it easier for smaller chip design houses to compete credibly. This is a deliberate strategy to grow the pool of companies that can build competitive mobile silicon, which directly expands the number of entities that do not need to write Qualcomm a check. Arm profits regardless of who wins – it collects royalties on every chip that uses its architecture – so a world with more chip designers suits Arm’s business even if it complicates Qualcomm’s.

The legal dimension adds another layer of pressure. Qualcomm and Arm have been in active litigation over the terms of the license Qualcomm acquired through its purchase of Nuvia. Arm argued that the Nuvia license did not automatically transfer, and that Qualcomm’s use of Nuvia’s core designs required a fresh agreement under different terms. A jury sided with Qualcomm in early 2025, but the dispute illustrated how willing Arm is to use licensing terms as a commercial lever. For any phone maker watching that case, the message was not subtle: Arm is not a passive infrastructure provider anymore.

What Qualcomm Still Has Going for It

None of this means Qualcomm is losing its position in the near term. The company’s modem technology – specifically the integration of 5G modems into Snapdragon chips – remains a genuine engineering advantage that Arm’s compute subsystems do not address. Building a competitive cellular modem from scratch is orders of magnitude harder than building a CPU or NPU cluster, and no Arm licensee outside of Apple and Qualcomm has managed it at flagship tier. For as long as 5G modem performance differentiates premium phones, Qualcomm has a floor under its value proposition that is difficult to collapse.

But modem integration is a defensive argument, not a growth one. The AI opportunity Qualcomm has been selling to investors – the idea that Snapdragon becomes the platform for a new era of intelligent devices – depends on Qualcomm maintaining differentiation in the NPU and software layer, not just the radio. If Arm’s compute subsystems continue maturing and phone makers continue building custom silicon around them, Qualcomm’s AI narrative becomes much harder to sustain at the valuation multiples the market has assigned it. The company’s next major Snapdragon reveal will carry more weight than usual, because it needs to answer a question that is growing louder with every new Arm subsystem release: what exactly does Qualcomm’s engineering premium still buy you?