Key findings
- • Data center liquid cooling is the fastest-growing physical AI market we track: US$3.9 Bn in 2025 to US$71.7 Bn by 2035 (33.8% CAGR).
- • The AI data center market is estimated at US$55.0 Bn in 2025, reaching US$264.4 Bn by 2035.
- • GPU capacity sold as a service grows at 26.1% a year - faster than the servers themselves (18.0%).
- • The median physical AI infrastructure market grows at 17.8% a year.
AI data center infrastructure markets, 2025 vs 2035
| Market | 2025 (US$) | 2035 (US$) | CAGR 2025-2035 |
|---|---|---|---|
| AI Data Center | $55.0 Bn | $264.4 Bn | 17.0% |
| AI Server | $70.0 Bn | $366.4 Bn | 18.0% |
| GPU Server Manufacturing | $124.0 Bn | $488.7 Bn | 14.7% |
| GPU-as-a-Service | $14.8 Bn | $150.5 Bn | 26.1% |
| Data Center Construction | $68.3 Bn | $169.3 Bn | 9.5% |
| Data Center Colocation | $65.7 Bn | $167.3 Bn | 9.8% |
| Data Center Liquid Cooling | $3.9 Bn | $71.7 Bn | 33.8% |
| Immersion Cooling | $1.8 Bn | $16.6 Bn | 24.9% |
| GPU Cooling | $6.1 Bn | $41.7 Bn | 21.2% |
| Data Center Heat Recovery | $312.0 Mn | $3.6 Bn | 27.7% |
| Photonic Data Centers | $1.5 Bn | $18.4 Bn | 28.5% |
| Orbital Data Center | $64.8 Mn | $994.3 Mn | 31.4% |
Source: Market-Reports.com estimates. Base year 2025, forecast to 2035.
Heat is the constraint
Each new generation of AI accelerators packs more power into the same rack, and air cooling runs out of headroom. That is why liquid and immersion cooling sit at the top of the growth table: they move from niche to default in AI halls. Heat recovery follows the same logic - waste heat becomes a resource once there is enough of it in one place.
Capacity is increasingly rented
GPU capacity sold by the hour grows faster than the hardware it runs on. Many companies would rather rent accelerators than build data centers, which shifts spending from equipment budgets to operating budgets - and creates room for specialised providers alongside the largest clouds.
Where the next bottleneck is
Power, cooling and interconnect now decide where AI capacity can be built. Photonic interconnects and even experimental orbital data centers are on the map because the conventional answers are reaching physical limits. For suppliers in electrical, thermal and construction markets, AI is a growth driver in its own right.
About the data
All figures are Market-Reports.com estimates of global market revenue in US dollars, with 2025 as the base year and forecasts to 2035; growth rates are compound annual growth rates (CAGR) for 2025-2035. They come from our catalogue of more than 4,600 market reports across 19 industries, in which every market is sized on its own definition. Markets overlap - an AI server is also part of AI infrastructure - so the figures in this study must not be added together. Medians are calculated across all reports whose market name matches the theme, as published in October 2026.
Use this data
Journalists, analysts and bloggers may quote this study and reproduce the table, with credit to Market-Reports.com and a link to this page:
Source: Market-Reports.com, "The Physical Bill for AI: Cooling, Servers and Data Center Markets to 2035" - https://www.market-reports.com/insights/ai-data-center-physical-cost-2035
Need a figure for a different market or region? Email contact@market-reports.com.
Key takeaways
- • Liquid cooling grows fastest among physical AI markets.
- • Rented GPU capacity grows faster than the servers themselves.
- • Power and heat now decide where AI can be built.