AI On-Devices

AI on-device is artificial intelligence that runs locally on phones, PCs, cars, and edge hardware rather than in the cloud, using the neural processing units built into the device’s chip. It cuts latency, cost, and privacy exposure, and it runs through Qualcomm, Apple, Arm, Samsung, and Micron.

CategoryThematic / Semiconductors — edge & on-device AI
Representative companies5
Last updated2026-06-15
Key takeaways
  • AI on-device is AI that runs locally on phones, PCs, cars, and edge hardware rather than in the cloud, using the neural processing units (NPUs) built into the device's chip.
  • The case for on-device is latency, cost, privacy, and offline use: a query answered on the device needs no round trip to a data center and exposes no data.
  • Qualcomm is the central merchant supplier: its Snapdragon X Elite NPU delivers 45 TOPS and can run models with more than 13 billion parameters locally.
  • Apple is the flagship deployer: Apple Intelligence runs a roughly 3-billion-parameter on-device model across more than 2.5 billion active devices, with Private Cloud Compute for larger requests.
  • The enablers are Arm (the CPU and GPU IP in nearly every device SoC), Samsung (Exynos and Galaxy AI plus LPDDR memory), and Micron (low-power LPDDR for edge AI).

What is AI on-device?

It is AI that runs on the hardware in your hand or on your desk, not in a remote data center. A modern phone or laptop chip includes a neural processing unit (NPU) sized to run a compact model locally, so a summary, a translation, or an assistant reply can be produced without sending data to the cloud. The concept holds the layers that make that possible: Qualcomm as the merchant chip supplier, Apple as the flagship deployer, Arm as the architecture inside most device chips, and Samsung and Micron for the SoCs and low-power memory.

Why run AI on the device instead of the cloud?

Four reasons drive on-device AI: latency, because there is no round trip to a server; cost, because the device does the work; privacy, because the data never leaves the device; and offline use. The hardware is now capable of it. Qualcomm’s Snapdragon X Elite NPU is rated at 45 TOPS and can run models with more than 13 billion parameters locally (Qualcomm Snapdragon X Elite), and Apple runs a roughly 3-billion-parameter on-device model, reserving Private Cloud Compute for larger requests (Apple Machine Learning Research). At WWDC 2026 Apple pushed the software further:

“We’re delivering the next generation of Apple Intelligence across our platforms; introducing Siri AI, a profoundly more intelligent, knowledgeable, and capable Siri.”

— Craig Federighi, SVP Software Engineering, Apple (Apple, Jun 8, 2026)

Who leads on-device AI silicon and software?

Qualcomm leads the merchant market: its Snapdragon platforms put on-device AI in premium Android phones and a growing share of AI PCs, and it is extending the same engines into cars. Apple leads deployment by sheer reach, running Apple Intelligence across more than 2.5 billion active devices (Apple, Jan 29, 2026). Underneath both sits Arm, whose CPU and GPU IP is in nearly every device SoC, with fiscal Q4 2026 revenue up 20% to a record $1.49 billion (Arm, May 6, 2026). Samsung adds Exynos and Galaxy AI plus LPDDR memory, and Micron supplies the low-power LPDDR that holds models on the device.

Who should consider AI on-device?

It suits an investor who wants AI exposure through large, liquid consumer-tech and chip names with a steadier profile than the GPU trade, sized as part of a broader allocation. The companies are profitable and dated: Apple’s fiscal Q2 2026 revenue grew 17% to $111.2 billion (Apple 8-K, Apr 30, 2026), and Qualcomm posted record automotive revenue of $1,326 million (Qualcomm 8-K, Apr 29, 2026).

The caveat is that on-device AI is bounded by the memory and power of a phone or laptop, so the largest models still run in the cloud, and the split is unsettled. Adoption also depends on software people want, where Apple’s more capable Siri slipped and was delayed in the EU (Apple, Jun 8, 2026). For an institution this works as a lower-beta AI sleeve that diversifies an accelerator-heavy book.

Which companies represent AI On-Devices?

CompanySectorWhat it does
Qualcomm (QCOM) Information Technology · Semiconductors (Mobile & Edge AI) The central merchant supplier of on-device AI. Snapdragon platforms carry NPUs that run generative AI locally, with the Snapdragon X Elite rated at 45 TOPS, spanning phones, AI PCs, and cars, with a data-center inference re-entry via AI200 and AI250.
Apple (AAPL) Information Technology · Consumer Electronics & On-Device AI The flagship on-device AI deployer. Apple Intelligence runs a roughly 3-billion-parameter on-device model on custom A-series and M-series silicon across more than 2.5 billion active devices, with Private Cloud Compute for larger requests.
Arm Holdings (ARM) Information Technology · Semiconductors (IP) Arm's CPU and GPU IP is inside nearly every smartphone and AI-PC SoC, so on-device AI runs on Arm-designed cores. Fiscal Q4 2026 revenue rose 20% to a record $1.49 billion as device and data-center royalties grew.
Samsung Electronics (005930) Information Technology · Semiconductors & Consumer Electronics Samsung runs on-device AI through its Exynos SoCs and Galaxy AI features on its phones, and supplies the LPDDR memory that feeds edge AI. It posted record Q1 2026 revenue of 133.87 trillion won on the AI memory boom.
Micron Technology (MU) Information Technology · Semiconductors (Memory) Micron supplies the low-power LPDDR memory that holds models and data for on-device AI in phones and AI PCs, alongside the HBM used in data centers. Fiscal Q2 2026 revenue nearly tripled to $23.86 billion on AI memory demand.

What are the risks of AI On-Devices?

On-device AI is real, but it is not where the largest models run.

  • Hardware ceiling. Phone and laptop memory and power cap on-device model size, so cloud still wins the frontier, and the on-device-versus-cloud split is unsettled.
  • Software adoption. Value depends on features users want; Apple’s Siri rollout slipped and was delayed in the EU (Apple, Jun 8, 2026).
  • Consumer cycle. Revenue tracks the smartphone and PC replacement cycle, not just AI.
  • Customer concentration. Qualcomm faces the loss of Apple’s modem business over time, a known headwind.

Related concepts, securities & terms

Frequently asked questions

What is AI on-device?

It is AI that runs locally on the device, a phone, PC, car, or edge appliance, using a neural processing unit (NPU) in the device's chip, instead of sending the request to a cloud data center. The basket holds the merchant chip supplier (Qualcomm), the flagship deployer (Apple), the architecture inside most device chips (Arm), and the memory and SoC makers (Samsung, Micron).

Why run AI on the device instead of the cloud?

Four reasons: latency, because there is no round trip to a server; cost, because the device does the compute; privacy, because data stays on the device; and offline use. Apple's design runs a roughly 3-billion-parameter model on-device, with Private Cloud Compute only for larger requests (Apple Machine Learning Research), and Qualcomm's Snapdragon X Elite NPU runs 13-billion-parameter models locally (Qualcomm).

Which companies represent AI on-device?

Five names: Qualcomm (QCOM), the merchant Snapdragon supplier; Apple (AAPL), which runs Apple Intelligence on 2.5 billion-plus devices; Arm (ARM), whose IP is in nearly every device SoC; Samsung (005930), with Exynos, Galaxy AI, and LPDDR memory; and Micron (MU), supplying low-power LPDDR. Qualcomm's Snapdragon X Elite NPU delivers 45 TOPS (Qualcomm).

What are the main risks of AI on-device?

On-device AI is constrained by the memory and power budget of a phone or laptop, so the largest models still run in the cloud, and the split between on-device and cloud is unsettled. Adoption depends on software that users actually want, where Apple's Siri rollout slipped and was delayed in the EU (Apple, Jun 8, 2026). The names are also exposed to the smartphone and PC replacement cycle.

Is AI on-device a good fit for an individual investor?

It suits an investor who wants AI exposure through large, liquid consumer-tech and chip names rather than data-center pure-plays, with a steadier profile than the GPU trade. The companies are profitable and dated: Apple's revenue grew 17% to $111.2 billion (Apple 8-K, Apr 30, 2026) and Arm grew 20% to a record $1.49 billion (Arm, May 6, 2026). Always check fees, holdings, and risk before investing.

How does AI on-device fit an institutional mandate?

As a lower-beta AI sleeve anchored by mega-cap consumer tech (Apple) and a key IP and chip layer (Arm, Qualcomm), with optionality if on-device inference takes share from the cloud. The thesis rests on installed base and silicon, not data-center capex, so it diversifies an AI book concentrated in accelerators. The allocator should weigh smartphone-cycle exposure and the unsettled on-device-versus-cloud split.

Is there an ETF for AI on-device?

There is no dedicated on-device-AI ETF. The names are large caps, Apple, Qualcomm, Arm, Samsung, and Micron, that appear in broad technology and semiconductor funds rather than a single pure-play product. Always check fees, holdings, and risk before investing.

Sources & references

  1. Qualcomm Announces Second Quarter Fiscal 2026 Results (SEC 8-K, Exhibit 99.1) · Qualcomm Incorporated / SEC EDGAR, 2026-04-29
  2. Snapdragon X Elite (45 TOPS Hexagon NPU) · Qualcomm Incorporated, 2026-06-15
  3. Apple reports second quarter results (SEC 8-K, Exhibit 99.1) · Apple Inc. / SEC EDGAR, 2026-04-30
  4. Updates to Apple's On-Device and Server Foundation Language Models · Apple Machine Learning Research, 2025-06-09
  5. Apple unveils next generation of Apple Intelligence, Siri AI, and more · Apple Inc., 2026-06-08
  6. Arm Holdings plc Reports Results for the Fourth Quarter and Fiscal Year Ended 2026 · Arm Holdings plc, 2026-05-06
  7. Samsung Electronics Announces First Quarter 2026 Results · Samsung Electronics, 2026-04-30