Data centers: where are they in the world?

What makes a country a good place for data centers?

The clean way to think about it: yesterday’s data center map followed users and fiber. Today’s AI data center map follows power first, then fiber.

Here are the main criteria that matter.

Criterion	Why it matters	Countries or regions that look strong
Reliable electricity	Data centers need firm, always-on power. For AI campuses, the question is no longer “can we rent a building?” but “can we secure hundreds of megawatts?”	United States, China, Canada, Nordics, Saudi Arabia, UAE
Fast grid connection	A market can look attractive on paper and still be useless if grid connection takes five to ten years. This is now one of the biggest real-world bottlenecks.	Selected U.S. states, China, Malaysia, Gulf countries, parts of Australia
Low-cost power	Electricity can decide whether a compute-heavy site makes economic sense. AI training and high-density workloads make this more important than before.	Nordics, Canada, parts of the U.S., China, Saudi Arabia, UAE
Strong fiber and cable access	Cloud regions, internet exchanges, and submarine cables make a place more useful. This is why Mumbai, Singapore, London, Frankfurt, Tokyo, and São Paulo matter.	United States, UK, Germany, Singapore, Japan, India, Brazil
Big local demand	Finance, cloud, gaming, streaming, government, AI, and enterprise software all create demand for nearby capacity.	United States, China, Europe, India, Japan
Land availability	Large AI campuses need land. Dense markets can keep premium workloads, but a lot of new physical expansion moves outward.	United States, Canada, Malaysia, India, Saudi Arabia, Australia
Political and legal trust	Hyperscalers prefer countries where contracts, assets, regulation, and cross-border operations feel predictable.	United States, Canada, Japan, Singapore, Western Europe
Data sovereignty rules	Local storage and processing rules can force capacity to be built inside a country. This matters more for government, finance, healthcare, and AI.	EU, China, India, Saudi Arabia, UAE
Cooling conditions	Cold air and water availability reduce cost and complexity. This is why Nordic and Canadian markets look better for some workloads than hot, dense cities.	Nordics, Canada, Ireland, parts of Japan, parts of the U.S.
Permitting and community acceptance	Even when power and demand are there, local opposition can slow projects. Noise, water use, land use, and grid stress are becoming political issues.	Mixed: easier in some U.S., Gulf, and Asian markets; harder in Ireland, Netherlands, Singapore, and parts of Europe

Are most data centers still in the United States?

Yes. Today, the United States is still the country at the center of the global data center map.

We saw this in almost every serious signal. The U.S. has more facilities than any other country, it holds more than half of global hyperscale capacity, and it has the deepest mix of cloud providers, AI companies, colocation operators, chip buyers, capital markets, and enterprise customers. Northern Virginia remains the world’s most famous data center hub for a reason: it combines cloud demand, fiber density, government and enterprise workloads, and decades of infrastructure clustering.

But the more interesting signal is what is happening inside the U.S. The market is getting bigger and actually spreading out. North America has more than 35 GW of data center capacity under construction, and a large share of that pipeline is now in frontier markets rather than only the traditional hubs. That tells us builders are not leaving the U.S. They are leaving the most congested parts of the U.S.

If you want more recent data on this point, please see our latest data center market report.

Are we building even more data centers in the US right now?

Yes, massively.

The usual story is “AI is causing a data center boom.” That is true, but it is too vague. We believe the better signal is vacancy. Why? Because when vacancy falls close to zero in major markets, new supply is not optional; it becomes a scramble. Developers are preleasing capacity years ahead because customers know that waiting can mean missing the next available power window.

The second signal is location. Northern Virginia, Dallas, Phoenix, Chicago, Atlanta, Ohio, Oregon, Texas, and Nevada all matter, but not for exactly the same reason. Some are demand hubs. Some are land-and-power markets. Some are becoming AI training locations because they can absorb larger electrical loads than the old city-adjacent clusters.

Is Europe packed with data centers already?

Yes. Europe is already packed, especially in the core hubs, but the pressure is pushing growth into new places.

The old European answer is FLAP-D: Frankfurt, London, Amsterdam, Paris, and Dublin. These markets doubled from roughly 1.8 GW of live capacity in 2019 to about 3.6 GW in 2025. That is a very strong signal because these are not easy markets. They have land issues, grid issues, permitting issues, and political scrutiny, yet they still doubled.

The second signal is concentration. London and Frankfurt alone are expected to account for about half of Europe’s total data center supply by the end of 2025. That tells us customers still pay for proximity to finance, enterprise demand, exchanges, and cloud ecosystems. The expensive markets stay expensive because they are useful.

But Europe’s next phase is more complicated. Grid lead times can stretch for years, and more AI growth is expected to land in Tier 2 and emerging markets. So Europe is not running out of demand. Rather, it is running out of easy places to put that demand.

Are data centers still going to Frankfurt, London, Amsterdam, Paris and Dublin?

Yes, but those cities cannot carry the whole next wave alone.

We found two signals that give us interesting information.

First, FLAP-D still captures most new European supply, and demand there remains strong.

Second, the same markets are exactly where power, land, and planning constraints are most painful. That combination creates a strange market: the best locations are still the hardest places to build.

Amsterdam is a good example. It has been politically cautious about new data centers, but developers still fight for capacity there because the connectivity value is real. Dublin shows the other side of the same story: huge cloud demand, but local grid stress makes each new project more controversial.

So the future European map will probably look like a barbell. Premium, latency-sensitive workloads stay in the classic hubs, while heavier AI and cloud expansion spills into Milan, Madrid, Warsaw, the Nordics, Marseille, Berlin, Manchester, and other secondary locations with better power or land.

If you want more recent data on this point, please see our latest data center market report.

Is China building data centers like crazy now?

Yes. China is one of the few places building data centers at a scale that can be compared with the U.S.

The most striking signal is capacity. China already has tens of gigawatts of installed data center capacity, and recent forecasts show it adding roughly another 28 GW by 2030. That is not a normal market expansion. That is infrastructure planning at national scale.

The second signal is policy. China’s “East Data, West Compute” strategy basically tries to move computing from expensive, power-constrained coastal cities toward western regions with more energy resources. That is an important difference from the U.S. model. In the U.S., the buildout is heavily driven by private hyperscalers chasing available power. In China, the state is much more directly shaping where compute should go.

The third signal is experimentation. China is already trying things like offshore and underwater data centers powered by wind. That does not mean underwater data centers become the mainstream model tomorrow, but it shows how aggressively the country is looking for ways around land, cooling, and energy pressure.

So yes, China is building fast.

Does Japan actually have a lot of data centers?

Yes. Today, Japan is a real data center country, but it is a high-cost, high-quality one.

Tokyo and Osaka are the obvious anchors. Japan has deep enterprise demand, finance, gaming, manufacturing, cloud adoption, and strict reliability expectations. That creates a strong reason to build locally, because not every workload can sit in Singapore, Korea, or the U.S.

The numbers show a market still expanding: Japan has more than 100 existing facilities, dozens more upcoming, and more than 500 MW of additional power capacity expected in upcoming projects. Tokyo is still the center, but Osaka matters because it gives the country a second major hub and a resilience option.

The interesting part here is how Japan is adapting to constraints. There are experiments with modular data centers under railway infrastructure, and developers are paying premiums for sites with secured power. That tells us Japan’s problem is not demand but rather how to fit more compute into a dense, expensive, earthquake-conscious country.

Are we building more and more data centers in India these days?

Yes, definitely. India is one of the clearest growth markets right now.

The strongest signal is that India’s data center capacity has crossed about 1.5 GW, with Mumbai alone holding roughly 53% of national capacity. That is a huge concentration. It shows that Mumbai is not just “one Indian market”; it is the country’s digital gateway because it has submarine cables, internet exchanges, finance, cloud customers, and enterprise demand in the same place.

The second signal is that growth is spreading beyond Mumbai. Hyderabad, Chennai, Delhi NCR, Pune, and Bengaluru all matter, but for different reasons. Hyderabad’s pipeline has become especially interesting because it moved from a secondary IT market into a serious hyperscale and AI location. When a city more than doubles its live IT capacity in a few years and builds a gigawatt-scale pipeline, it is no longer just catching overflow.

So it looks like India is underbuilt relative to its digital economy.

The population is huge, cloud adoption is still rising, AI workloads are coming, and data localization makes domestic infrastructure more valuable. The constraint is execution: power, heat, land, and grid reliability will decide which Indian cities actually convert demand into working capacity.

If you want more recent data on this point, please see our latest data center market report.

Are we building more and more data centers in Malaysia right now?

Yes. Malaysia, and especially Johor, is one of the hottest data center stories in the world right now.

Johor is the cleanest example of Singapore spillover. Singapore has the connectivity, customers, legal trust, and regional headquarters. Johor has more land, more room to build, and a location close enough to serve the same regional ecosystem. That is why Johor moved from a secondary market to a hyperscale magnet so quickly.

The scale is the real signal. DC Byte tracks more than 6 GW of total supply in Johor across live, under-construction, committed, and early-stage projects. Even if not all early-stage capacity becomes real capacity, the order of magnitude matters. This is no longer a small overflow market.

But the boom has a ceiling. Johor now has to prove that power, water, and local planning can keep up. The interesting story here is that Malaysia became the physical expansion layer for a Singapore-centered digital region, and now it has to mature from opportunistic buildout into managed infrastructure planning.

Is Singapore still a data center hub, or is it too full now?

Singapore is still a hub, but it is now a selective one.

It has almost everything data centers want: submarine cables, regional headquarters, finance, cloud demand, low corruption, legal stability, and world-class connectivity. The issue is space and power. Singapore cannot behave like Texas or Johor because it is a compact city-state.

That is why the recent 300 MW additional capacity target matters. It shows Singapore is not “closed” to data centers, but it is filtering growth. New capacity has to be cleaner, more efficient, and more strategically useful. In normal conversation, we would say: Singapore still wants data centers, just not any data center at any cost.

This makes Singapore and Johor one combined story. Singapore keeps the premium, regulated, latency-sensitive, headquarters-driven layer. Johor gets much of the physical scale. Together, they form one of Asia’s most important data center zones.

If you want more recent data on this point, please see our latest data center market report.

Is the Gulf really becoming a data center region?

Yes, especially Saudi Arabia and the UAE, but the reason is strategic more than organic.

In Europe, data centers follow enterprise demand. In India, they follow digital growth. In the Gulf, they are also part of national strategy. Saudi Arabia and the UAE want AI infrastructure because compute is becoming an economic asset, the way ports, airlines, and energy infrastructure were strategic assets before.

Saudi Arabia is the most aggressive signal. Its data center megawatt load is projected to grow around 29% annually from 2024 to 2030, and some forecasts point to more than 1 GW of new capacity by the end of the decade. That is very fast for a market that was not historically in the same tier as the U.S., Europe, Japan, or Singapore.

The UAE has a different profile: smaller population, but strong capital, strong connectivity ambitions, cloud regions, and a clear AI positioning. The Gulf’s challenge is obvious: heat and water. But for AI training workloads that do not need to sit next to every end user, cheap energy, government coordination, and capital can matter more than climate.

Does Russia have many data centers?

Russia has data centers, but it is not a major global data center hub.

The domestic market is real. Russia has cloud, telecom, government, banking, and data sovereignty demand, and the market is still expected to grow. Moscow remains the main center, and low-cost power helps.

But the scale is much smaller than the U.S., China, Europe, India, Japan, or even several fast-growing secondary markets. One forecast puts Russia’s data center power capacity at only around 138 MW by 2031. That is tiny compared with gigawatt-scale pipelines in places like North America, China, India, Johor, or Saudi Arabia.

The real issue is not whether Russia can host servers. It can. The question is whether global hyperscalers, hardware supply chains, and international customers want to build there.

Are we seeing more data centers in Canada now?

Yes, but Canada is more of a power-and-climate complement to the U.S. than a replacement for it.

Canada has a strong pitch: cooler weather, political stability, land, renewable-heavy power in several provinces, and proximity to U.S. users. Montreal, Toronto, and Alberta are the main clusters, and tracked capacity is now very large when planned AI and GPU-focused projects are included.

The 5.6 GW Wonder Valley project in Alberta is a good signal because it shows what Canada is trying to sell: not just “we have a few cloud buildings,” but “we can host very large AI infrastructure if power and land are secured.” That is a different category of ambition.

The limitation is that Canada does not have the same cloud ecosystem density as the U.S. So the best read is: Canada wins when the workload cares about power, climate, and North American proximity more than being in the densest U.S. cloud hub.

Are data centers moving to the Nordics now?

Yes, actually, especially the compute-heavy ones.

The Nordics are attractive for a very practical reason: cool climate, abundant renewable power, lower electricity costs than many European hubs, and more land. Sweden, Norway, Finland, Denmark, and Iceland all benefit from the same basic shift: Europe needs more compute, but Frankfurt, London, Amsterdam, Paris, and Dublin cannot absorb everything.

The signal that makes this more than a “green data center” talking point is the AI buildout. Norway’s Stargate project is expected to start with hundreds of megawatts and target 100,000 GPUs, powered by renewable hydropower. Sweden also has a large upcoming capacity base, with more than 500 MW of upcoming IT power capacity reported.

The caveat is latency. The Nordics are excellent for AI training, batch processing, backup, and compute workloads that can sit farther from users. They are less perfect for ultra-low-latency finance or enterprise workloads that need to be right next to London or Frankfurt. So the Nordics are not replacing Europe’s core hubs; they are becoming Europe’s power-friendly compute layer.

Is Latin America getting real data centers, or is it still small?

It is getting real data centers, but the region is still much smaller than North America, Europe, and Asia.

The action is concentrated in Brazil, Mexico, Chile, and Colombia. Brazil is the biggest because São Paulo has the region’s deepest enterprise and cloud demand. Mexico is getting more interesting because Querétaro can serve as a major data center cluster with proximity to U.S. demand and Mexican enterprise growth. Chile stands out for stability, connectivity, and renewable energy potential.

The best recent signal is Mexico. CloudHQ announced a $4.8 billion campus in Querétaro with six facilities and 900 MW of electrical capacity. That is not a small regional colo project. That is the kind of project that changes how investors look at a market.

Latin America’s story is still mostly regional localization. It needs data centers because users, banks, governments, streamers, fintechs, and cloud customers need lower latency and local infrastructure. The region is not yet competing with the U.S. or China in global AI capacity, but it is no longer a footnote.

Is Australia building more data centers too?

Yes. Australia is already a serious data center market, and the recent growth is sharp.

Sydney and Melbourne are the core. Sydney has close to 900 MW of built-out colocation capacity, while Melbourne is around half that size but growing quickly. A separate market report found that Australian data center occupancy expanded from 37 MW in 2005 to about 1.3 GW in 2025, with much of that growth coming after 2020.

The recent signal is political as much as commercial. Data centers already use a meaningful share of Australia’s electricity, and policymakers are now talking about rules that would make operators bring new renewable energy, pay for grid infrastructure, and coordinate better with energy planners.

So where are data centers today, and where will they be built next?

Today, the data center world still has a clear core: the United States first, then Europe, China, Japan, Singapore, and a few large regional hubs. If we are talking about existing capacity and hyperscale depth, the U.S. is still the obvious center.

But if we ask where the map is heating up now, the answer changes. The new action is in secondary U.S. markets, India, Johor in Malaysia, Saudi Arabia and the UAE, Canada, the Nordics, Australia, and selected Latin American hubs like Querétaro, São Paulo, and Santiago.

So, at the end, it looks like existing data centers are where the internet economy was already dense. New AI-era data centers are going where power can still be found.

Country or region	Are there many data centers?	What is really happening?
United States	Yes	Still the global leader. New builds are spreading beyond Northern Virginia and the classic hubs because power and land are easier in frontier markets.
Europe	Yes	Very dense in FLAP-D, but power and permitting are pushing growth toward secondary markets.
China	Yes	Scaling at national-infrastructure speed, with policy pushing compute toward energy-rich regions.
Japan	Yes	Mature and high-quality. Tokyo and Osaka dominate, while constraints force more creative site choices.
India	Yes, fast-growing	Mumbai leads, but Hyderabad, Chennai, Delhi NCR, Pune, and Bengaluru are becoming more important.
Singapore	Yes, but selective	Still a premium hub, but capacity is filtered through efficiency, power, and land constraints.
Malaysia	Yes, very fast-growing	Johor is becoming Southeast Asia’s physical expansion layer for the Singapore-centered ecosystem.
Saudi Arabia / UAE	Yes, emerging fast	The Gulf is building compute as strategic infrastructure, not just as a normal cloud market.
Russia	Not really globally	The domestic market exists, but sanctions and geopolitical risk keep it outside the global hyperscale core.
Canada	Yes, selectively	Strong for AI and high-density workloads when power, land, climate, and North American proximity matter.
Nordics	Yes, selectively	Strong for renewable-powered compute and AI training, less ideal for ultra-low-latency city workloads.
Brazil	Yes, regionally	São Paulo is Latin America’s main cloud and enterprise hub.
Mexico	Yes, growing fast	Querétaro is becoming a serious data center cluster, with very large new projects.
Chile	Yes, emerging	Santiago benefits from stability, connectivity, and renewable energy potential.
Australia	Yes	Sydney and Melbourne are established, and AI demand is making grid planning a national issue.

If you want more recent data on this point, please see our latest data center market report.

Who is the author of this content?

NEW MARKET PITCH TEAM

We track new markets so founders and investors can move faster

We build living "market pitch" documents for emerging markets: AI, synthetic biology, new proteins, and more. Instead of outdated PDFs or hallucinated LLM answers, our clients get a clean, visual, always-updated view of what's really happening: key players, deals, regulations, and signals that matter. Learn more about us.