AI infrastructure
Chips, data centres, power and rented cloud compute behind AI — data centre power use set to double by 2030 (IEA)
AI infrastructure is the physical and rented stack that trains and runs AI models: accelerator chips (GPUs and custom ASICs) and the memory and networking around them, the data centres that house them, the electricity and cooling they draw, and the cloud providers that rent the compute out. Buyers are hyperscale cloud companies, AI labs, governments and enterprises. The IEA says global data centre investment nearly doubled from 2022 to about half a trillion dollars in 2024. AI is the main driver of growth in data centre electricity demand, and Synergy says GenAI pushed cloud infrastructure revenue up 30% year on year in Q4 2025.
new market · sits inside NAICS 334413, 334111, 518210 · researched 2026-11-14
data centre electricity consumption · world415
↗TWh (about 1.5% of world electricity) · 2024
investment · world500
↗USD billions (stated as 'half a trillion dollars' of global data centre investment) · 2024
global revenue (cloud infrastructure services: IaaS, PaaS, hosted private cloud) · world419
↗USD billions · 2025
data centre electricity consumption · US176
↗TWh (about 4.4% of US electricity) · 2023
G
Growth
Dated and sourced; forecasts are labelled as forecasts.
- 2025-04IEA Base Case forecast: data centre electricity use more than doubles to around 945 TWh by 2030, with AI the most important driver. The US had 45% of 2024 consumption, China 25% and Europe 15%. ↗
- 2025-04IEA Base Case forecast: electricity use by accelerated servers, mainly driven by AI, grows 30% a year, against 9% a year for conventional servers. Accelerated servers account for almost half of the net increase. ↗
- 2024-12DOE/LBNL forecast: US data centre electricity, which rose from 58 TWh in 2014 to 176 TWh in 2023, reaches 325–580 TWh by 2028, or about 6.7–12% of total US electricity. ↗
- 2026-02Synergy: Q4 2025 cloud infrastructure service revenue was US$119.1B, up 30% year on year (ninth straight quarter of faster growth). Synergy calls GenAI 'clearly the primary driver'. Fastest-growing tier-two providers include CoreWeave, OpenAI, Oracle, Crusoe and Nebius. ↗
S
Segments
Each linked segment has its own page with its companies.
- AI chips & accelerators 6 — GPUs and custom AI ASICs (TPUs, Trainium and others) that train and run models
- Memory, networking & chip foundries 4 — HBM memory, switching and interconnect silicon, and the leading-edge fabs and packaging that make the chips
- Data centre colocation 2 — Buildings, land and grid connections leased to or built by hyperscalers and AI labs
- Power and cooling equipment — UPS, power distribution, liquid cooling and thermal systems for high-density racks
- Hyperscale AI cloud 4 — AWS, Azure, Google Cloud and Oracle renting GPU/accelerator capacity and AI services
- Neoclouds (GPU clouds) 2 — GPU-focused specialist clouds (CoreWeave, Crusoe, Nebius) renting AI capacity
C
Leading companies in ai infrastructure
16 companies, with the scale each source states.
| Company | Where | What they do | Scale |
|---|---|---|---|
| Nvidia NVDA ↗ | Santa Clara, United States since 1993 | Leading AI GPU and data centre systems and networking supplier (Hopper/Blackwell, CUDA) | revenue (total company, FY ended 2026-01-25): 215.938 USD billions (2026) ↗ |
| AMD AMD ↗ | Santa Clara, United States since 1969 | Instinct AI GPUs and EPYC data centre CPUs | revenue (total company): 34.639 USD billions (2025) ↗ |
| Broadcom AVGO ↗ | Palo Alto, United States since 1961 | Custom AI ASICs for hyperscalers plus data centre Ethernet switching silicon | revenue (total company, FY ended 2025-11-02): 63.887 USD billions (2025) ↗ |
| TSMC 2330.TW ↗ | Hsinchu, Taiwan since 1987 | Contract foundry that makes most leading AI accelerators, and does their advanced packaging | revenue (total company): 3.81 NTD trillions (US$122.21B as stated) (2025) ↗ |
| SK hynix 000660.KS ↗ | Icheon, South Korea since 1983 | High-bandwidth memory (HBM) stacked alongside AI accelerators | revenue (total company): 97.15 KRW trillions (US$72.49B as stated) (2025) ↗ |
| Intel INTC ↗ | Santa Clara, United States since 1968 | Xeon data centre CPUs, Gaudi AI accelerators and a foundry business | revenue (total company): 52.853 USD billions (2025) ↗ |
| Amazon Web Services AMZN ↗ | Seattle, United States since 2006 | Largest cloud provider; rents GPU capacity and its own Trainium/Inferentia chips | market share (worldwide cloud infrastructure services, Q4): 28 % (2025) ↗ |
| Microsoft Azure MSFT ↗ | Redmond, United States since 2008 | Cloud platform hosting large GPU clusters and AI model services | market share (worldwide cloud infrastructure services, Q4): 21 % (2025) ↗ |
| Google Cloud GOOGL ↗ | Mountain View, United States since 2008 | Cloud platform renting its own TPU accelerators and Nvidia GPUs | market share (worldwide cloud infrastructure services, Q4): 14 % (2025) ↗ |
| Oracle ORCL ↗ | Austin, United States since 1977 | Oracle Cloud Infrastructure, building large GPU clusters for AI labs | revenue (total company, FY ended 2026-05-31): 67.357 USD billions (2026) ↗ |
| CoreWeave CRWV ↗ | Livingston, NJ, United States since 2017 | GPU-focused neocloud renting AI compute; now one of the top ten cloud providers (Synergy) | revenue: 5.131 USD billions (2025) ↗ |
| Equinix EQIX ↗ | Redwood City, United States since 1998 | Global colocation and interconnection data centre REIT | revenue: 9.217 USD billions (2025) ↗ |
| Digital Realty DLR ↗ | Austin, United States since 2004 | Wholesale and colocation data centre REIT leasing capacity to hyperscalers | revenue: 6.113 USD billions (2025) ↗ |
| Vertiv VRT ↗ | Westerville, OH, United States since 2016 | Power, thermal and liquid-cooling equipment for data centres | revenue: 10.23 USD billions (2025) ↗ |
| Cerebras Systems CBRS ↗ | Sunnyvale, United States since 2015 | Wafer-scale AI chips and systems, plus an inference cloud | revenue: 0.51 USD billions ($510 million as stated) (2025) ↗ |
| Tenstorrent ↗ | Santa Clara, United States since 2016 | Toronto-founded AI chip designer (RISC-V); sells hardware to data centre operators and licenses IP | employees: 550 employees ('550-plus', over 120 in Canada) (2024) ↗ |
CA
In Canada
- 2024-12Ottawa launched the Canadian Sovereign AI Compute Strategy with up to $2 billion (from Budget 2024): up to $700M for new or expanded data centres through the AI Compute Challenge, up to $1B for public computing infrastructure, and up to $300M through the AI Compute Access Fund for SMEs. ↗
- 2025-03The federal government finalized up to $240 million for Cohere's $725 million project to build domestic AI compute capacity at a new multi-billion-dollar AI data centre in Canada. Cohere is the first AI Compute Challenge recipient. ↗
- 2024-11Toronto-founded AI chip maker Tenstorrent moved its headquarters to Santa Clara, California in November 2023 and redomiciled to a Delaware holding company. It still had over 120 of its 550-plus employees in Canada. ↗
R
What would have to be true
- Grid capacity and connection queues: DOE/LBNL's 2028 range (6.7–12% of US electricity) is wide, and local power limits could slow builds.
- Supply concentration: leading accelerators depend on a few suppliers (Nvidia GPUs, TSMC fabrication, HBM memory), so a shortage or export restriction at any one ripples through.
- Demand must hold up: today's capex assumes AI services keep earning enough to pay for the buildout. IEA's own scenarios range widely after 2030.
- Canadian talent and companies can move south, as Tenstorrent's 2023 relocation shows, which would weaken sovereign-compute goals.
Quick answers
- How big is the ai infrastructure market?
- data centre electricity consumption (world): 415 TWh (about 1.5% of world electricity), 2024 — International Energy Agency; investment (world): 500 USD billions (stated as 'half a trillion dollars' of global data centre investment), 2024 — International Energy Agency.
- Is the ai infrastructure market growing?
- IEA Base Case forecast: data centre electricity use more than doubles to around 945 TWh by 2030, with AI the most important driver. The US had 45% of 2024 consumption, China 25% and Europe 15%.
- Who are the leading ai infrastructure companies?
- Nvidia (United States), AMD (United States), Broadcom (United States), TSMC (Taiwan), SK hynix (South Korea), Intel (United States), Amazon Web Services (United States), Microsoft Azure (United States), and 8 more in the table above.
- Which ai infrastructure companies are publicly traded?
- Nvidia (NVDA), AMD (AMD), Broadcom (AVGO), TSMC (2330.TW), SK hynix (000660.KS), Intel (INTC), Oracle (ORCL), CoreWeave (CRWV), Equinix (EQIX), Digital Realty (DLR).
- What is happening with ai infrastructure in Canada?
- Ottawa launched the Canadian Sovereign AI Compute Strategy with up to $2 billion (from Budget 2024): up to $700M for new or expanded data centres through the AI Compute Challenge, up to $1B for public computing infrastructure, and up to $300M through the AI Compute Access Fund for SMEs.
- What are the main segments of the ai infrastructure market?
- AI accelerators, Memory, networking and foundry, AI data centres (colocation and self-build), Power and cooling equipment, Hyperscale cloud AI compute, Neoclouds.
- What are the risks in the ai infrastructure market?
- Grid capacity and connection queues: DOE/LBNL's 2028 range (6.7–12% of US electricity) is wide, and local power limits could slow builds.
Sources (9)
- Energy and AI — Executive summary — International Energy Agency
- Energy and AI — Energy demand from AI — International Energy Agency
- DOE Releases New Report Evaluating Increase in Electricity Demand from Data Centers — U.S. Department of Energy
- GenAI Helps Drive Quarterly Cloud Revenues to $119 Billion as Growth Rate Jumped Yet Again in Q4 — Synergy Research Group
- Canada to drive billions in investments to build domestic AI compute capacity at home — Innovation, Science and Economic Development Canada
- Government of Canada finalizes investment to support Canadian-Born AI leader, Cohere — Innovation, Science and Economic Development Canada
- Tenstorrent quietly relocates to the U.S. as AI chip race heats up — The Logic
- XBRL company facts (NVDA, AMD, AVGO, INTC, ORCL, CRWV, EQIX, DLR, VRT) — U.S. SEC
- Company infoboxes: Nvidia, AMD, Broadcom, TSMC, SK Hynix, Intel, Amazon Web Services, Microsoft Azure, Google Cloud Platform, Oracle Corporation, CoreWeave, Equinix, Digital Realty, Vertiv, Cerebras — Wikipedia