The AI hardware market reached $182.0 billion in 2025, and NVIDIA took 85% of it. On Global Market Insights' forecast, AI compute alone would outgrow the whole 2025 chip industry by 2035.
AI hardware (compute-only) market size, 2025–2035, US$ billions, against fixed 2025 yardsticks from the chip industry. A diamond marks where the forecast passes each one; the label gives the first published year at or above it. Solid: 2025 estimate; dashed: GMI forecast.
GMI publishes only 2025, 2026 and 2035 values, so the segments between them are straight lines, not yearly forecasts. Yardsticks are held at 2025 levels; SIA projects the chip industry itself will reach roughly $1 trillion in 2026. The scopes overlap but differ: GMI counts accelerator systems at first sale, including the memory and packaging inside them, while SIA counts chip sales by chipmakers.
AI hardware market share by vendor, 2025, % of global revenue. Who leads the AI chip market, and by how much.
GMI publishes shares for NVIDIA, Broadcom and AMD and a 96.8% combined share for the top five. "Marvell + Qualcomm" is derived (96.8% minus the three named shares); "All others" is 100% minus 96.8%. At 85.05% of $182.0B, NVIDIA's implied 2025 AI hardware revenue is about $154.8B (derived).
AI hardware market size by accelerator type, 2025, US$ billions: GPUs vs. custom AI chips (ASICs), NPUs and FPGAs
GPUs ($162.1B) and NPUs ($2.5B) are published in dollars. Custom ASICs (9.07%) and FPGAs (0.27%) are published as shares; their dollar values are derived from the $182.0B total. "Not itemized" is the $0.4B left over, likely rounding. Custom ASICs here are merchant chips sold to buyers; chips a company designs for its own use, such as Google's TPUs, are outside GMI's scope.
AI hardware market size by country, 2025, US$ billions
Japan and the United Kingdom are derived from published shares: Japan is 19.24% of Asia Pacific's $47.3B; the UK is 37.67% of Europe, which is 8.02% of $182.0B. All other values are published in dollars. North America as a whole was $114.7B.
AI hardware market by deployment, 2025, US$ billions: cloud vs. on-premises vs. edge
On-premises is derived from its published 25.00% share of $182.0B. GMI expects edge infrastructure to grow fastest, at 29.2% a year from 2026 to 2035, and cloud to be 65.00% of revenue by 2035.
NVIDIA data center revenue by fiscal year, US$ billions (the biggest seller in the AI chip market)
NVIDIA's fiscal years end in late January (FY2026 ended Jan 25, 2026). Data center revenue includes networking ($31.4B in FY2026), which GMI's compute-only market excludes; data center compute alone was $162.4B. NVIDIA regrouped its reporting from Q2 FY2027, so later periods are not shown.
Data center or AI revenue of the top AI chip makers, latest fiscal year, US$ billions (NVIDIA vs. AMD vs. Broadcom vs. Marvell)
Each company defines its own category, and fiscal years differ. NVIDIA: data center, year ended Jan 25, 2026. Broadcom: AI semiconductor revenue (custom chips plus networking), year ended Nov 2, 2025, stated on its earnings call as a round $20B. AMD: Data Center segment, including EPYC server CPUs, calendar 2025. Marvell: data center end market, year ended Jan 31, 2026.