NPU equipped enterprise endpoints are forecast to reach 24 percent by 2025. Forty five percent of organizations expect AI to reach mainstream status within two to three years. Global PC shipments totaled 62.7 million units in 2023 with overall volumes down and long term forecasts pointing to 1.9 billion units by 2027.
Key Takeaways
- 62.7 million PCs shipped globally in 2023, representing 5.4% year-over-year growth
- 269.2 million PCs shipped globally in 2023, representing 1.0% year-over-year decline
- 1.9 billion units shipped (forecast) for PCs in 2027 globally (IDC forecast baseline for long-term PC market recovery)
- 45% of respondents in a Gartner survey expect AI to be mainstream in their organizations within 2–3 years (Gartner survey, 2024)
- 34% of organizations plan to deploy AI at the edge within the next 12 months (Gartner survey on edge/IoT, 2024)
- Computer systems on a chip (SoC) shipments for AI PCs are forecast to reach 150 million units in 2026 (Strategy Analytics forecast cited by credible trade coverage)
- 24% of enterprise endpoints are expected to have an NPU by 2025 (IDC forecast on on-device AI readiness)
- 21% of organizations said they use generative AI in production (Gartner, 2024 survey finding cited in press release)
- 1.3 billion people are projected to use at least one AI service by 2026 (IDC forecast of AI users, commonly cited in IDC press releases)
- AI infrastructure spending of $84.9 billion forecast for 2025 worldwide (IDC forecast baseline)
- Energy use from AI inference is estimated at 29,000 MWh in 2023 (Stanford AI Index, 2024 report)
- Global data center energy consumption reached 460 TWh in 2022 (IEA data cited in IEA Data Centres report)
- NPU-based systems can reduce energy per inference by 50% relative to CPU-only execution for common on-device models (peer-reviewed study on edge inference efficiency)
- Top-1 accuracy improvements of 1.5–3.0 percentage points reported for quantization-aware training vs post-training quantization on image classification models (peer-reviewed survey)
- Quantization can reduce model size by 4x to 8x versus FP32 for typical transformer weights (peer-reviewed survey)
With AI PCs accelerating device upgrades, AI readiness and NPU adoption are set to surge, boosting on device intelligence.
Related reading
01 · Category
Market Size4 stats
01
62.7 million PCs shipped globally in 2023, representing 5.4% year-over-year growth
02
269.2 million PCs shipped globally in 2023, representing 1.0% year-over-year decline
03
1.9 billion units shipped (forecast) for PCs in 2027 globally (IDC forecast baseline for long-term PC market recovery)
04
11.6% year-over-year decline in global tablet shipments in 2023 to 139.0 million units (Strategy Analytics), illustrating a weaker broader client device category alongside AI PC momentum.
Interpretation
Market Size Interpretation
For the Market Size view of AI PCs, shipments reached 62.7 million in 2023 with 5.4% year over year growth even as overall PC units fell to 1.0% in 2023 and the market is still expected to rise to 1.9 billion units by 2027, while tablets declined 11.6% to 139.0 million to show AI PC momentum is happening amid weaker broader device demand.
02 · Category
Industry Trends6 stats
01
45% of respondents in a Gartner survey expect AI to be mainstream in their organizations within 2–3 years (Gartner survey, 2024)
02
34% of organizations plan to deploy AI at the edge within the next 12 months (Gartner survey on edge/IoT, 2024)
03
Computer systems on a chip (SoC) shipments for AI PCs are forecast to reach 150 million units in 2026 (Strategy Analytics forecast cited by credible trade coverage)
04
38% of organizations reported using AI/ML in their cybersecurity operations in 2023 (ISC2 Cybersecurity Workforce Study + additional survey findings published by ISC2, 2024 edition), indicating growing AI functionality needs across endpoint security stacks.
05
3.2 billion gigabytes of data were generated per day globally from 2020–2025 projections (EMC/Rapport/IDC-derived industry estimates commonly cited by Statista using IDC/EMC sources), demonstrating data growth pressures that motivate on-device/offload AI patterns.
06
Windows OEMs are required to meet Microsoft’s “Copilot+ PC” hardware criteria, including a neural processing unit (NPU) of 40+ TOPS for classification (Microsoft Copilot+ PC specifications), tying AI PC adoption directly to NPU capability.
Interpretation
Industry Trends Interpretation
With 45% of organizations expecting AI to be mainstream within 2 to 3 years and 34% planning edge AI in the next 12 months, the Industry Trends signal is clear that “AI in the PC” is accelerating beyond pilots toward on device deployment, supported by forecasts of 150 million AI PC SoC shipments in 2026 and the Copilot+ PC requirement of a 40+ TOPS NPU.
03 · Category
User Adoption6 stats
01
24% of enterprise endpoints are expected to have an NPU by 2025 (IDC forecast on on-device AI readiness)
02
21% of organizations said they use generative AI in production (Gartner, 2024 survey finding cited in press release)
03
1.3 billion people are projected to use at least one AI service by 2026 (IDC forecast of AI users, commonly cited in IDC press releases)
04
46% of organizations say AI skills gaps are a major barrier to AI adoption (World Economic Forum Future of Jobs report, 2023 data)
05
58% of respondents said they plan to use AI for customer service and support (Salesforce State of Service, 2024), a common enterprise workload that can drive demand for AI-enabled endpoints.
06
55% of knowledge workers said generative AI helps them complete tasks faster (Microsoft Work Trend Index 2024), supporting demand for end-user productivity tooling on AI-capable devices.
Interpretation
User Adoption Interpretation
With only 24% of enterprise endpoints expected to include an NPU by 2025 but 21% of organizations already using generative AI in production, user adoption is accelerating faster than on-device AI readiness, suggesting demand for AI-capable PCs will be pulled forward by organizations and end users, including 55% of knowledge workers saying generative AI helps them finish tasks faster.
More related reading
04 · Category
Cost Analysis8 stats
01
AI infrastructure spending of $84.9 billion forecast for 2025 worldwide (IDC forecast baseline)
02
Energy use from AI inference is estimated at 29,000 MWh in 2023 (Stanford AI Index, 2024 report)
03
Global data center energy consumption reached 460 TWh in 2022 (IEA data cited in IEA Data Centres report)
04
In 2023, worldwide cybersecurity spending was $188.3 billion (Gartner estimate cited in Gartner press release)
05
In 2024, worldwide cybersecurity spending is forecast to total $219.1 billion (Gartner forecast)
06
Global endpoint security software market size is forecast to reach $11.3 billion by 2027 (IDC forecast)
07
NPU-based inference can be 10–100x more energy efficient than CPU for certain on-device CNN workloads (peer-reviewed survey of neural network acceleration energy tradeoffs), supporting energy efficiency rationale for AI PC silicon.
08
Organizations that consolidated vendor tools reported 17% lower breach costs (IBM Cost of a Data Breach Report 2023), implying potential savings from integrated security/control planes that may include AI-assisted detection on PCs.
Interpretation
Cost Analysis Interpretation
Cost analysis shows that while AI infrastructure spending is projected to reach $84.9 billion in 2025 and AI inference energy use is about 29,000 MWh in 2023, AI PC designs that use NPU inference up to 10–100x more energy efficiently can help control operational costs, while the shift to more integrated security tools linked to a 17% lower breach cost and rising cybersecurity spend from $188.3 billion in 2023 to $219.1 billion in 2024 further strengthens the business case for cost-effective AI-assisted detection and prevention on endpoints.
05 · Category
Performance Metrics6 stats
01
NPU-based systems can reduce energy per inference by 50% relative to CPU-only execution for common on-device models (peer-reviewed study on edge inference efficiency)
02
Top-1 accuracy improvements of 1.5–3.0 percentage points reported for quantization-aware training vs post-training quantization on image classification models (peer-reviewed survey)
03
Quantization can reduce model size by 4x to 8x versus FP32 for typical transformer weights (peer-reviewed survey)
04
2.3x faster image classification throughput using optimized mobile inference on dedicated accelerator hardware versus CPU baseline for the same model architecture (peer-reviewed evaluation in the context of on-device accelerators), demonstrating performance motivation for NPUs in AI PCs.
05
Up to 8-bit quantization enables ~4x smaller model storage footprint compared with FP32 for transformer-like architectures (peer-reviewed survey on post-training quantization), aligning with the hardware efficiency push behind AI PCs.
06
Intel’s Core Ultra (Meteor Lake) NPU is specified to provide up to 11 TOPS (Intel product specifications), a measurable accelerator capability used by AI PCs for on-device inference.
Interpretation
Performance Metrics Interpretation
For performance metrics, AI PCs show a clear efficiency and speed trend, cutting energy per inference by about 50% versus CPU-only, delivering up to 2.3x higher throughput on optimized accelerator hardware, and reducing model storage by roughly 4x to 8x through quantization compared with FP32, all while leveraging NPUs like Intel’s up to 11 TOPS capability for on-device AI inference.
Reference
Cite This Report
This report is designed to be cited. We maintain stable URLs and versioned verification dates. Copy the format appropriate for your publication below.
APA
Priyanka Sharma. (2026, February 13). AI In The PC Industry Statistics. Gitnux. https://gitnux.org/ai-in-the-pc-industry-statistics
MLA
Priyanka Sharma. "AI In The PC Industry Statistics." Gitnux, 13 Feb 2026, https://gitnux.org/ai-in-the-pc-industry-statistics.
Chicago
Priyanka Sharma. 2026. "AI In The PC Industry Statistics." Gitnux. https://gitnux.org/ai-in-the-pc-industry-statistics.
Sources & references
32 datasets cited across this report · attribution is report-level
+16 additional datasets cited (not shown individually)