AI data centers are proliferating rapidly. And they’ve quickly become a flashpoint of both business and political conversations.

One problem lies with defining what counts as an “AI data center.” Many are similar to traditional data centers, which can make differentiating the two tricky.

There are some differences. AI data centers tend to be larger. They often have more intense power and cooling requirements. And some are constructed by companies on the forefront of AI, specifically to support their AI ambitions.

This article tracks the growth, impact, and future of AI data centers.

Editor’s Picks

Number of AI Data Centers

There are roughly 3,900 AI-optimized data centers worldwide (AllAboutAI)

Approximately 33% of the world’s 11,800 data centers are now optimized for AI workloads. This is expected to grow to 70% by 2030.

33 countries have data centers capable of powering AI (SSRN)

Some governments consider control over AI to be core to global strategic competition, national security, and economic innovation. This is beginning to raise questions of digital sovereignty, the control over the digital systems and data that AI depends on.

While these countries have data centers capable of powering AI within their territories, the ability to obtain AI capabilities is further limited by countries’ dependence on the U.S. and China to provide the necessary cloud or chip infrastructure.

132 of the world’s 225 cloud regions have AI hardware (SSRN)

Cloud regions are clusters of large, interconnected data centers worldwide.

These 132 regions are situated in only 33 different countries, predominantly in North America, Europe, the Middle East, and Asia. This concentration gives significant technological and geopolitical influence to a relatively small part of the world.

AI-optimized servers account for 31% of data center power consumption (Gartner)

Data centers with AI-optimized servers, those designed or configured to efficiently run compute-intensive AI workloads, are projected to consume 84% more electricity in 2026 than in 2025. They will account for 31% of total data center electricity consumption in 2026.

With demand for compute-intensive AI driving unprecedented growth in data center power, AI capacity is now largely constrained by power availability.

The U.S. has 8.2 GW of AI-dedicated data center capacity (ABI Research)

In 2026, the U.S. holds the largest active AI-dedicated data center capacity at 8.2 GW, while its nearest competitors, China and Europe, hold less than half that capacity at 3.1 GW and 3.5 GW, respectively.

This leaves the U.S. with 40% of total installed capacity.

One of the drivers of this growth may come from the Stargate Project, a company intending to invest $500 billion to build new AI infrastructure for OpenAI in the U.S.

AI represents 23% of data center workloads (JLL)

While AI represented about a quarter of all data center workloads in 2025, it is projected that this could double by 2030.

The type of workload may also differ. In 2025, workloads were mostly driven by AI training—building or updating models—whereas workloads in 2030 are expected to focus on inference, using the models to provide information, generate content, and perform other tasks.

The global computing capacity of AI chips is 17.07 million H100-equivalents (Stanford)

Google alone accounts for around 25% of the global AI-compute capacity. It is estimated that 70% of the global AI compute is owned by the five U.S. hyperscalers.

There are 1,360 large data centers operated by hyperscale companies (Synergy)

These hyperscale data centers account for 48% of global data center capacity. By 2031, hyperscale operators will account for two-thirds of total capacity.

Hyperscale data centers aren’t necessarily dedicated to AI. But AI data centers do tend to be larger, and hyperscale companies are investing heavily into AI.

Growth of AI Data Centers

Investment in AI data centers could total $232 billion in 2026 (Goldman Sachs)

This annual figure is estimated to almost double by 2031.

Investment in data centers is one layer of the baseline estimates that anticipate $7.6 trillion of cumulative capital expenditure between 2026 and 2031. This figure includes cumulative investment in compute, data centers, and power infrastructure.

The four biggest hyperscale companies will invest $700 billion on AI this year (Fortune)

Alphabet, Amazon, Meta, and Microsoft are the four biggest hyperscale companies.

These companies were previously expected to spend around $600 billion on AI in 2026, but this forecast has risen to $700 billion after the latest earnings updates.

Major U.S. tech companies invested $420 billion in AI infrastructure in 2025 (Congressional Research Service)

These companies included Amazon, Google, Meta, and Microsoft.

In 2026, AI related investment in the U.S. is projected to be close to $581 billion.

Companies around the world will invest over $1 trillion in AI infrastructure in 2026 (Goldman Sachs)

This includes AI investment by U.S. hyperscalers as well as investment by other private companies and companies outside the U.S.

Of the $1 trillion, 58% is expected to be invested in AI in the U.S.

Spending on AI in 2026 will equal 1.8% of U.S. GDP (Goldman Sachs)

AI-related data center construction, compute hardware and networking made up about 0.8% of U.S. GDP in Q1 2026.

All computing infrastructure reached about 1.5% of GDP, compared with an average of about 0.7% from 2015 to 2022. Epoch says the AI boom has “more than doubled” that share.

Companies added 18 GW in AI data center capacity in 2026 (McKinsey)

This increased the proportion of data center capacity dedicated to AI from 54% to 60%.

Global demand for total data center capacity could almost triple by 2030.

The power consumption of AI servers will grow 8x between 2023 and 2028 (New York Times)

AI server electricity use is expected to grow as much as 44% annually from 2023 to 2028 and may account for 50% of data-center electricity demand by 2030.

The EU is investing $23.2 billion in AI gigafactories (European Commission)

At the Artificial Intelligence Action Summit in Paris, Commission President Ursula von der Leyen launched InvestAI, an initiative to mobilize €200 billion in AI, including a new European fund of €20 billion for AI gigafactories.

InvestAI will finance four future AI gigafactories across the EU. The gigafactories will have around 100,000 latest-generation AI chips, around four times more than the AI factories being set up right now.

Over the same period, the value of obligated funds increased from $261 million to $675 million (over 150% growth).

By 2026, this increased to $7.2 billion, a 966% increase since 2024.

The global computing capacity of AI chips has grown 3.3x per year since 2022 (Stanford)

This is the annual growth of AI compute. To understand the scale of this growth, cumulative global AI-chip computing capacity was equivalent to more than 15 million Nvidia H100 GPUs by early 2026.

The Federal Reserve’s tracking of AI supercomputers estimates that they represent 10-20% of existing global computing capacity.

Impact of AI Data Centers

Investment in AI contributed 0.97 percentage points to real GDP growth in the first 9 months of 2025 (St. Louis Federal Reserve)

The contribution to GDP growth of information processing equipment, software, R&D and data centers has normalized despite the ongoing AI investment boom.

This doesn’t indicate a drop in the level of investment in these categories, rather their contribution fell because investment growth slowed.

Global AI demand will account for up to 6.6 billion cubic meters of water withdrawal by 2027 (arXiv)

Many millions of liters of freshwater are consumed for cooling the servers and for electricity generation needed for AI.

The total annual on-site water consumption by U.S. data centers in 2028 could double or even quadruple the 2023 level, reaching approximately 150 – 280 billion liters and further stressing water infrastructures.

The power capacity of AI data centers is 29.6 GW (Stanford)

AI chip power accounted for roughly 11.8 GW of the total, with the remainder attributed to cooling, networking, and other data center infrastructure.

For context, this is comparable to New York state at peak demand.

Future of AI Data Centers

There will be at least 16 million AI servers in the U.S. by 2030 (ETA)

The deployment of these accelerated servers led to a 14% increase in U.S. data center electricity use between 2023 and 2024. U.S. data center electricity use was projected to increase by 22% in 2025 and by another 29% by 2026.

Increases in energy use are primarily driven by growth in both the quantity and rated power of accelerated servers shipped each year.

Demand for AI data centers will rise 33% per year between 2023 and 2030 (McKinsey)

By 2030, Gen AI, currently the fastest-growing advanced-AI use case, may account for around 40% of the total demand.

AI inference capacity demand is expected to rise from 20.9 GW in 2025 to 93.3 GW in 2030. By then, inference is expected to make up more than 40% of total data-center demand.

AI servers will consume up to 326 TWh by 2028 (New York Times)

AI servers are estimated to rise from 22% of U.S. data-center electricity use in 2023 to 50–55% by 2028. In 2023, inference servers accounted for 60% of AI server electricity. That share is expected to fall to 50-55% by 2028.

AI servers may account for 55% of total data center energy use by 2030 (ETA)

In 2030, AI servers may dominate energy consumption, accounting for 84% of total server energy use.

The ever-increasing computing capabilities of accelerated servers are expected to increase U.S. data center electricity use by a compounded annual growth rate of 21% after 2026.

Scaling AI data centers will require $5.2 trillion through 2030 (McKinsey)

This scaling requires investment in the entire compute power value chain, which includes the real estate developers that build data centers, the utilities that power them, the semiconductor firms that produce chips, and the cloud service hyperscalers that host trillions of terabytes of data.

The largest share of investment, 60 percent ($3.1 trillion), will go to technology developers and designers, which produce chips and computing hardware for data centers.

70% of global data center capacity demand will come from AI workloads by 2030 (McKinsey)

This is a 16 percentage point increase from 2025.

While AI is about 70% of capacity demand, it is about 78% of investment. So each unit of AI capacity costs more than a unit of traditional capacity.

Data center capacity demand for AI workloads will increase 3.5x by 2030 (McKinsey)

This reflects growth in AI workload demand from about 44 GW in 2025 to 155 GW in 2030. This is about a 252% change over five years.

The U.S. will have 26.4 GW of AI-driven data center capacity by 2031 (ABI Research)

This is 54% of total installed capacity.

U.S.-based hyperscalers continue to lead capacity additions, but neocloud and sovereign providers are scaling faster in regions where pricing transparency, Graphics Processing Unit (GPU) access, and data control matter more.

Investment in AI will equal 2.8% of U.S. GDP in 2028 (Goldman Sachs)

In 2026, AI investment is expected to make up 1.8% of GDP in the U.S., increasing to 2.5% in 2027 and 2.8% in 2028.

Estimates suggest that AI investment will make up 1.4% of GDP worldwide in 2028.

Global investment in AI infrastructure will increase by 27.7% in 2027 (Gartner)

While AI investment is projected to grow by 27.7% between 2026 and 2027, the increase between 2025 and 2026 is expected to be 41.6%. This shows that, while spending continues to rise, the growth rate is expected to slow.

Global annual investment in AI data centers will increase 87.9% by 2031 (Goldman Sachs)

This represents a change from $232 billion in 2026 to $436 billion in 2031.

But predictive ability is limited by uncertainty around innovation. Because of the massive scale at which data centers are built, even modest changes in cost per megawatt—for example through innovations in cooling architecture or new power delivery paradigms—could compound quickly.

The world will spend $7.6 trillion on AI infrastructure through the end of the decade (Pimco)

AI infrastructure investment is expected to be 192% higher than spending on energy and grid infrastructure, 217% higher than defense spending, and 443% higher than spending on reshoring and supply chains.

At least 60% of AI workloads in the EU and U.S. will be hosted on hyperscale infrastructure by 2030 (McKinsey)

To address increasing demand, cloud service providers, which currently own more than half of the world’s AI-ready data center capacity, are rapidly constructing state-of-the-art facilities.

Impending supply constraints have also led to partnering with colocation providers (known as “colos”) that are similarly expanding their infrastructures.

AI could represent half of data center workloads by 2030 (JLL)

Though expected to rise to 50% in 2030, AI only represented a quarter of all data center workloads in 2025, with training driving most of the demand. However, a significant shift is anticipated in 2027, when inference workloads could overtake training as the dominant AI requirement.

Challenges Facing AI Data Centers

AI infrastructure becomes economically obsolete in 5 years (Syzygy)

Gross spending on AI infrastructure may not mean lasting capital accumulation.

It is plausible that more than half of projected 2026 hyperscaler capital expenditure replaces economically obsolete hardware rather than expanding productive capacity.

NVIDIA designs 80-95% of AI chips (SSRN)

Aside from China, the other 32 countries worldwide that have data centers capable of powering AI within their territories depend on US-owned NVIDIA accelerators to power their AI data centers.

Nvidia accounts for more than 60% of global AI-chip computing capacity, with Google and Amazon supplying much of the remainder and Huawei holding a small but growing share.

Spending on power for AI data centers could increase 87% by 2031 (Goldman Sachs)

In 2026, spending on power is expected to be around $39 billion, and this could reach $73 billion in 2031.

However, this remains a relatively small share of total investment in AI infrastructure compared to silicon, data-center construction, and supporting systems.

Air-based cooling systems are effective for AI chip racks with energy demands up to 50 kW (McKinsey)

50 kW of power may be adequate for AI inferencing workloads, computing work used to run a finished model, but not for training workloads from building AI models.

The need for additional cooling is prompting innovations such as using liquids to remove heat directly from racks, which is significantly more effective in absorbing and transferring heat than air.