Generative-AI-capable smartphones are forecast to reach 45% of global shipments in 2026, up from 36% in 2025 (Counterpoint Research, GenAI Smartphone Share 2026), even as total smartphone volume falls 13.9% to the lowest level on record. The same split runs through PCs: Gartner projected 143 million AI PCs with a neural processing unit (NPU) for 2026, yet its February 2026 update warns that soaring DRAM and SSD prices will cut PC shipments 10.4% and delay the AI PC surge. Silicon is moving faster than demand, with laptop NPUs climbing from Microsoft’s 40 TOPS Copilot+ floor to 80 TOPS on Qualcomm’s Snapdragon X2 Elite. At the very small end of the market, audio and sensor workloads account for more than 80% of TinyML chipset shipments (ABI Research, TinyML 3Q 2025). We aggregated data from Gartner, IDC, Counterpoint Research, Canalys, ABI Research, MLCommons, MarketsandMarkets, Arm Holdings SEC filings, and chipmaker disclosures, building on our broader edge AI statistics roundup.
TL;DR
- GenAI-capable smartphones: 45% of global shipments in 2026, up from 36% in 2025 (Counterpoint Research, 2026)
- IDC’s GenAI smartphone bar: an NPU of 30 TOPS or more at INT8 (IDC, The Rise of Gen AI Smartphones)
- AI PCs: 77.8 million units and 31% of PCs in 2025, forecast 143 million and 55% in 2026 (Gartner, August 2025)
- Copilot+ PC minimum: 40+ TOPS NPU, 16 GB RAM, 256 GB storage (Microsoft, Windows 11 Specifications)
- Snapdragon X2 Elite NPU: 80 TOPS, 78% more than the prior generation (Qualcomm, Snapdragon Summit 2025)
- Audio/sound and sensor analysis: more than 80% of TinyML chipset shipments through 2030 (ABI Research, 3Q 2025)
- Syntiant NDP120: 10x to 100x better energy efficiency than traditional microcontrollers on wake word detection (Syntiant, MLPerf Tiny v1.3)
- Edge AI hardware market: 26.14 billion dollars in 2025 to 58.90 billion by 2030 (MarketsandMarkets)
- Smartphones: 80.5% of edge AI hardware volume in 2024 (MarketsandMarkets)
- Arm fiscal 2026 royalty revenue: 2.61 billion dollars, up 21% (Arm Holdings, Form 6-K, May 2026)
- DRAM plus SSD prices: up 130% by end of 2026 (Gartner, February 2026)
- PC shipments 2026: down 11.3%, with ASPs up 17% (IDC, June 2026)
1. AI PCs: The NPU Becomes Default Silicon
The headline number is not adoption, it is inevitability. Gartner forecast 143 million AI PCs in 2026, 55% of the market, and its definition is simply a PC with an embedded NPU. Once Intel, AMD, Qualcomm and Apple all ship an NPU in mainstream parts, AI PC share converges on total PC share whether buyers ask for it or not. For voice software, that means a local accelerator for noise suppression, transcription and voice effects is quickly becoming a baseline assumption on new Windows laptops rather than a premium feature.
Forecasts have been volatile. Canalys expected 40% AI-capable share in 2025, while Gartner’s later August 2025 number for the same year was 31%, and Gartner itself blamed tariffs and paused purchasing for the slower ramp (Gartner, AI PCs Will Represent 31% of Worldwide PC Market by the End of 2025). Our AI PC statistics page tracks the full shipment picture.
| Metric | Value | Source |
|---|---|---|
| AI PC shipments, 2025 | 77.8 million units (31% of PCs) | Gartner, August 2025 forecast |
| AI PC shipments, 2026 | 143 million units (55% of PCs) | Gartner, August 2025 forecast |
| AI laptops as share of all laptops, 2025 | 51% | Gartner, August 2025 forecast |
| AI-capable PC share, full year 2024 | 17% | Canalys, March 2025 |
| AI-capable PC shipments, Q4 2024 | 15.4 million units (23% of PCs) | Canalys, March 2025 |
| AI-capable PC forecast, 2025 | More than 100 million units (40% of PCs) | Canalys forecast |
| AI-capable PC forecast, 2028 | 205 million units (44% CAGR 2024-2028) | Canalys forecast |
Context: Canalys counts any PC with a dedicated AI block (AMD XDNA, Apple Neural Engine, Intel AI Boost, Qualcomm Hexagon), so its share figures run higher than Gartner’s for the same period. Canalys forecasts are reported via Light Reading.
2. Smartphones: GenAI NPUs Move Into the Mid-Range
Phones remain the volume engine of edge AI. Counterpoint expects GenAI-capable models to reach 45% of global smartphone shipments in 2026, up from 36% in 2025, and to pass 50% in 2027 (Counterpoint Research, GenAI Smartphone Share to Rise to 45% of Global Shipments in 2026). The share is rising inside a shrinking market, which says the NPU is protected even when the bill of materials is under pressure.
The more telling shift is price tier. Counterpoint found that 38% of GenAI-capable phones shipped in 2025 were mid-range devices in the 300 to 499 dollar band, triple the 2024 level, while 88% of premium chipsets already supported GenAI (as reported by GSMArena). On-device voice features, from live call translation to dictation, follow the NPU down the price ladder.
| Metric | Value | Source |
|---|---|---|
| GenAI-capable share of smartphone shipments, 2025 / 2026 | 36% / 45% | Counterpoint Research, June 2026 |
| GenAI-capable chipsets as share of all smartphone chipsets, 2025 | 35% (up 74% year over year) | Counterpoint Research, 2025 |
| GenAI chipset vendor share, 2025 | Apple 46%, Qualcomm 35%, MediaTek 12% | Counterpoint Research, 2025 |
| Premium chipsets supporting GenAI, 2025 | 88% | Counterpoint Research, 2025 |
| Mid-range share of GenAI-capable phones, 2025 | 38% (3x the 2024 level) | Counterpoint Research, 2025 |
| GenAI smartphone NPU threshold | 30 TOPS or more at INT8 | IDC, The Rise of Gen AI Smartphones |
| GenAI smartphone shipments, 2024 | 234.2 million (up 364%) | IDC forecast, 2024 |
| GenAI smartphone shipments, 2028 | 912 million (78.4% CAGR 2023-2028) | IDC forecast, 2024 |
Most recent available data for the IDC rows: IDC, The Rise of Gen AI Smartphones (2024 forecast). IDC also projected 73.1% growth in 2025; its forecasts predate the 2026 memory shock and have not been publicly re-cut for the GenAI segment.
3. The TOPS Race: Laptop NPUs Double Microsoft’s Floor
Microsoft set the reference point in 2024: a Copilot+ PC needs an NPU capable of 40+ TOPS, 16 GB of DDR5/LPDDR5 and 256 GB of storage (Microsoft, Windows 11 Specifications). Two years later the flagship laptop NPU has doubled that, and every major x86 and Arm vendor clears the bar. The competitive question is shifting from raw TOPS to sustained performance per watt, which is what decides whether an always-on voice or camera effect drains the battery.
TOPS figures also hide architectural differences. Intel quotes 50 TOPS for the NPU but 180 platform TOPS once the GPU and CPU are added, and Apple now puts a Neural Accelerator in each GPU core of the M5 alongside its 16-core Neural Engine (Apple Newsroom, M5, October 2025). Comparing a single NPU number across vendors is increasingly misleading.
| Metric | Value | Source |
|---|---|---|
| Copilot+ PC NPU minimum | 40+ TOPS | Microsoft, Windows 11 Specifications |
| Qualcomm Snapdragon X2 Elite NPU | 80 TOPS (78% more than prior generation) | Qualcomm, Snapdragon Summit, September 2025 |
| AMD Ryzen AI 400 Series NPU (XDNA 2) | Up to 60 TOPS | AMD, CES 2026 launch, January 2026 |
| Intel Core Ultra Series 3 (Panther Lake) | 50 NPU TOPS; up to 180 platform TOPS | Intel, CES 2026 launch, January 2026 |
| Apple M5 peak GPU compute for AI | More than 4x M4 | Apple Newsroom, October 2025 |
| Apple M5 unified memory bandwidth | 153 GB/s (nearly 30% above M4) | Apple Newsroom, October 2025 |
| Snapdragon 8 Elite Gen 5 Hexagon NPU (phone) | 37% faster, 16% better performance per watt | Qualcomm, September 2025 |
Phone NPU gains come from Qualcomm’s Snapdragon 8 Elite Gen 5 announcement. Vendor TOPS are self-reported peak INT8 figures, not independent benchmarks.
4. Edge Voice and TinyML: Audio Is the Always-On Workload
Voice is where edge AI started and where it still dominates by unit count. Audio/sound processing and sensor data analysis together represent more than 80% of all TinyML chipset shipments through 2030, with shipments growing at a 29% CAGR (ABI Research, TinyML Chipset Shipments 2023-2030). Wake words, keyword spotting and noise suppression must run continuously, so they live on milliwatt-class accelerators rather than on the main CPU or the cloud.
Benchmarks are catching up. MLCommons added a streaming wake word test to MLPerf Tiny v1.3, built on a 20-minute synthesized recording with 50 instances of the wake word, and its keyword spotting test requires 90% top-1 accuracy (MLCommons, MLPerf Tiny). The leader, Syntiant’s NDP120, handled the streaming workload at roughly a 2% duty cycle, leaving headroom for concurrent noise cancellation and beamforming (Syntiant, MLPerf Tiny v1.3 results). For the software side of the same trend, see our speech-to-text statistics.
| Metric | Value | Source |
|---|---|---|
| Audio/sound plus sensor analysis share of TinyML chipset shipments | More than 80% through 2030 | ABI Research, TinyML report 3Q 2025 |
| TinyML chipset shipment growth | 29% CAGR through 2030 | ABI Research, TinyML report 3Q 2025 |
| MLPerf Tiny streaming wake word test set | 20-minute recording, 50 wake word instances | MLCommons, MLPerf Tiny v1.3 (September 2025) |
| MLPerf Tiny keyword spotting accuracy target | 90% top-1 | MLCommons, MLPerf Tiny |
| Syntiant NDP120 streaming wake word duty cycle | About 2% at 50 MHz | Syntiant, MLPerf Tiny v1.3 |
| Syntiant NDP120 efficiency vs traditional MCUs | 10x to 100x better | Syntiant, MLPerf Tiny v1.3 |
| Syntiant purpose-built silicon deployed | More than 100 million units | Syntiant, September 2025 |
| Phi Silica on a 40+ TOPS NPU | 230 ms first token, up to 20 tokens per second, 56% lower power than CPU per token | Microsoft, Windows Experience Blog, December 2024 |
Most recent available data for Phi Silica: Microsoft, Phi Silica, small but mighty on-device SLM (December 2024). It is Microsoft’s NPU-tuned small language model with 4-bit weights, which shows the same NPU that powers voice effects can also run text generation locally. Real-time audio pipelines like voice cloning software sit in exactly this latency-sensitive class of workload.
5. Market Size: Same Trend, Very Different Totals
Edge AI market sizing depends more on scope than on outlook. MarketsandMarkets puts edge AI hardware at 26.14 billion dollars in 2025, growing to 58.90 billion by 2030 at 17.6% CAGR (MarketsandMarkets, Edge AI Hardware Market), while Precedence Research sizes edge AI chips at just 8.3 billion dollars in 2025. Both count largely the same phones and PCs, and they differ on what share of each device’s value they attribute to AI.
The composition is more stable than the totals. Smartphones are 80.5% of edge AI hardware volume and consumer electronics 81.3%, according to MarketsandMarkets, and devices drawing 1 to 3 watts make up 80.5% of volume. Edge AI is overwhelmingly a battery-powered inference business: inference accounts for 99.8% of volume. Our semiconductor industry statistics cover the wider chip cycle.
| Metric | Value | Source |
|---|---|---|
| Edge AI hardware market, 2025 to 2030 | 26.14 billion to 58.90 billion dollars (17.6% CAGR) | MarketsandMarkets, Edge AI Hardware Market 2025-2030 |
| Smartphones’ share of edge AI hardware volume, 2024 | 80.5% | MarketsandMarkets |
| Inference share of edge AI hardware volume | 99.8% | MarketsandMarkets |
| Asia Pacific share of edge AI hardware volume, 2024 | 41.0% | MarketsandMarkets |
| Edge AI market (BCC scope), 2024 / 2025 / 2030 | 8.7 / 11.8 / 56.8 billion dollars (36.9% CAGR 2025-2030) | BCC Research, Edge AI Market, June 2025 |
| Edge AI chips market, 2025 / 2026 / 2034 | 8.3 / 9.77 / 36.12 billion dollars (17.75% CAGR) | Precedence Research, Edge AI Chips Market |
| Consumer electronics share of edge AI chip revenue, 2024 | 40% | Precedence Research |
| ASIC share of edge AI chips by type, 2024 | 35% (NPUs and AI accelerators fastest growing) | Precedence Research |
Cross-check sources: BCC Research, Edge AI Market to Grow at 36.9% CAGR Through 2030 and Precedence Research, Edge AI Chips Market.
6. Arm’s Disclosures: Royalties as an Edge AI Barometer
Arm is the closest thing edge AI has to a public meter, because nearly every phone NPU sits next to Arm CPU cores. Arm reported fiscal 2026 royalty revenue of 2.61 billion dollars, up 21%, and named smartphones, edge AI, physical AI and cloud AI as the growth drivers (Arm Holdings, Form 6-K Exhibit 99.2, fiscal Q4 2026). Growth came from higher royalty rates per chip (Armv9 and Arm Compute Subsystems) as much as from unit volume.
The latest quarter kept the pace: royalties rose 22% to 715 million dollars in the quarter ended June 30, 2026, and Arm described AI as expanding into edge devices such as PCs and smartphones (Arm Holdings, Form 6-K Exhibit 99.2, fiscal Q1 2027). Arm does not break out an edge AI revenue line, so this is an indirect indicator.
| Metric | Value | Source |
|---|---|---|
| Fiscal 2026 total revenue (year to March 31, 2026) | 4.92 billion dollars (record) | Arm Holdings, Form 6-K, May 2026 |
| Fiscal 2026 royalty revenue | 2.61 billion dollars (up 21%) | Arm Holdings, Form 6-K, May 2026 |
| Fiscal Q4 2026 royalty revenue | 671 million dollars (up 11%) | Arm Holdings, Form 6-K, May 2026 |
| Fiscal Q1 2027 total revenue | 1,289 million dollars (up 22%) | Arm Holdings, Form 6-K, Q1 FY2027 |
| Fiscal Q1 2027 royalty revenue | 715 million dollars (up 22%) | Arm Holdings, Form 6-K, Q1 FY2027 |
| Cumulative Arm-based chips shipped | More than 350 billion | Arm Holdings, Form 6-K, May 2026 |
| Developers in the Arm ecosystem | More than 22 million | Arm Holdings, Form 6-K, Q1 FY2027 |
7. Challenges: The Memory Squeeze Meets the NPU
The biggest obstacle to edge AI in 2026 is not compute, it is memory. Gartner expects combined DRAM and SSD prices to surge 130% by the end of 2026, lifting PC prices 17% and smartphone prices 13%, and it warned that the projected AI PC surge will be delayed through 2027 (Gartner, Surging Memory Costs Will Reduce Global PC and Smartphone Shipments in 2026). On-device models need RAM as much as TOPS, which is why Copilot+ PCs require 16 GB.
IDC’s June 2026 update is harsher on PCs, with an 11.3% unit decline and a 20% drop in Q4 alone, and IDC sees no relief before the end of 2027 (IDC, PC Market Enters Volatile Territory as Memory Shortage Persists Through 2027). The likely effect is that NPU features stay concentrated in mid-range and premium devices longer than 2025 forecasts assumed, and that installed bases age.
| Metric | Value | Source |
|---|---|---|
| Combined DRAM and SSD price increase by end of 2026 | 130% | Gartner, February 2026 |
| 2026 shipment change, PCs / smartphones | -10.4% / -8.4% | Gartner, February 2026 |
| 2026 memory price inflation, DRAM / NAND flash | 80% / 202% | Gartner, February 2026 |
| PC unit shipments, 2026 | -11.3% | IDC, June 2026 |
| PC unit shipments, Q4 2026 | -20% year over year | IDC, June 2026 |
| PC average selling price growth, 2026 | 17% | IDC, June 2026 |
| Smartphone shipments, 2026 | -13.9% to 1.08 billion units (lowest on record) | Counterpoint Research, June 2026 |
Gartner also expects PC lifetimes to lengthen by 15% for business buyers and 20% for consumers by the end of 2026, which slows the rate at which older, NPU-less machines leave the installed base.
Summary: Edge AI Voice and NPU by the Numbers
| Metric | Value | Source |
|---|---|---|
| GenAI-capable smartphone share, 2026 | 45% (from 36% in 2025) | Counterpoint Research, June 2026 |
| GenAI-capable smartphone chipsets, 2025 | 35% of all chipsets, up 74% | Counterpoint Research, 2025 |
| GenAI smartphone NPU threshold | 30+ TOPS at INT8 | IDC |
| AI PC shipments, 2026 forecast | 143 million (55% of PCs) | Gartner, August 2025 |
| AI-capable PC share, Q4 2024 | 23% | Canalys, March 2025 |
| AI-capable PC forecast, 2028 | 205 million units | Canalys |
| Copilot+ PC NPU minimum | 40+ TOPS | Microsoft |
| Snapdragon X2 Elite NPU | 80 TOPS | Qualcomm, September 2025 |
| Intel Core Ultra Series 3 platform AI | Up to 180 TOPS (50 TOPS NPU) | Intel, CES 2026 |
| Apple M5 peak GPU compute for AI | More than 4x M4 | Apple, October 2025 |
| TinyML shipments from audio and sensor workloads | More than 80% | ABI Research, 3Q 2025 |
| TinyML chipset shipment growth | 29% CAGR through 2030 | ABI Research, 3Q 2025 |
| Wake word chip efficiency vs MCUs | 10x to 100x better | Syntiant, MLPerf Tiny v1.3 |
| Edge AI hardware market, 2030 | 58.90 billion dollars | MarketsandMarkets |
| Edge AI market (BCC scope), 2030 | 56.8 billion dollars (36.9% CAGR) | BCC Research, June 2025 |
| Smartphones’ share of edge AI hardware volume | 80.5% | MarketsandMarkets |
| Arm fiscal 2026 royalty revenue | 2.61 billion dollars, up 21% | Arm Holdings, Form 6-K |
| DRAM plus SSD price surge by end of 2026 | 130% | Gartner, February 2026 |
| PC shipments, 2026 | -11.3% | IDC, June 2026 |
Methodology and Sources
We ran targeted searches across AI PC and smartphone shipments, NPU specifications, TinyML and voice benchmarks, market sizing and semiconductor filings, then traced each figure to the publishing organization. Forecasts are labeled as forecasts, vendor performance claims are labeled as self-reported, and stats older than 2025 are flagged in place.
- Gartner, AI PCs Will Represent 31% of Worldwide PC Market by the End of 2025 (August 2025)
- Gartner, Surging Memory Costs Will Reduce Global PC and Smartphone Shipments in 2026 (February 2026)
- Canalys AI-capable PC forecasts, as reported by Light Reading
- Counterpoint Research, GenAI Smartphone Share to Rise to 45% of Global Shipments in 2026 (June 2026)
- Counterpoint Research GenAI chipset data, as reported by GSMArena (2025)
- IDC, The Rise of Gen AI Smartphones
- IDC, PC Market Enters Volatile Territory as Memory Shortage Persists Through 2027 (June 2026)
- IDC, Worldwide Smartphone Market to Decline 13.9% in 2026 (May 2026)
- Microsoft, Windows 11 Specifications (Copilot+ PC requirements)
- Microsoft, Phi Silica, small but mighty on-device SLM (December 2024)
- Qualcomm, Snapdragon 8 Elite Gen 5 press release (September 2025); Snapdragon X2 Elite figures from Qualcomm’s Snapdragon Summit 2025 announcement
- Apple Newsroom, M5 (October 2025)
- Intel Core Ultra Series 3 and AMD Ryzen AI 400 Series figures: vendor launch announcements at CES 2026 (January 2026)
- ABI Research, TinyML Chipset Shipments 2023-2030 (3Q 2025)
- MLCommons, MLPerf Tiny benchmark
- Syntiant, NDP120 MLPerf Tiny v1.3 results (September 2025)
- MarketsandMarkets, Edge AI Hardware Market 2025-2030
- BCC Research, Edge AI Market to Grow at 36.9% CAGR Through 2030
- Precedence Research, Edge AI Chips Market
- Arm Holdings, Form 6-K Exhibit 99.2, fiscal Q4 2026 (May 2026)
- Arm Holdings, Form 6-K Exhibit 99.2, fiscal Q1 2027 (quarter ended June 30, 2026)
- Data watch: Gartner’s 143 million AI PC forecast for 2026 predates its February 2026 memory-cost update, which says the AI PC surge will be delayed, so read it as a ceiling. AI PC share differs by firm because definitions differ (Gartner counts any embedded NPU; Microsoft’s Copilot+ bar is 40+ TOPS). Smartphone 2026 volume is 1.08 billion units at Counterpoint and 1.09 billion at IDC, both a 13.9% decline. Edge AI market sizing ranges from 8.3 billion dollars (Precedence, chips) to 26.14 billion (MarketsandMarkets, hardware) for 2025, with CAGRs from 17.6% to 36.9% (BCC), reflecting scope, not disagreement on direction. IDC’s GenAI smartphone forecasts date from 2024 and Microsoft’s Phi Silica numbers from December 2024. Vendor TOPS are peak self-reported figures. Edge AI and Vision Alliance developer survey results and Semico Research forecasts are member- or paywall-gated and could not be verified publicly, so they are not used.
Last updated: October 3, 2026. We update this roundup quarterly, and the next refresh is expected when Arm reports fiscal Q2 2027 results and Gartner publishes its next AI PC forecast.