# Compute Forecast — Europe 2031 > Compute forecast: how AI compute capacity in Europe, the US, China and the rest of the world could evolve through 2031. Canonical URL: https://europe2031.ai/compute-forecast Language: en Last updated: 2026-06-11 --- A central factor in our scenario is compute: the AI data centres and chips required to train and run frontier AI systems. We believe that compute is rapidly becoming one of the world’s most important strategic resources, as the amount of compute a country or continent controls increasingly determines its ability to build advanced AI systems, deploy them at scale, capture the resulting economic gains, and maintain geopolitical leverage. If a continent already owns a meaningful share of global AI compute, foreign companies or governments will think twice about restricting its access to AI models or related infrastructure. Below, we estimate (i) the growth of global AI compute until 2031 and (ii) the relative shares of different world regions over time. Our estimate is based on publicly available information about historical growth rates, supply chain bottlenecks and announced data center projects. See our full quantitative forecast [here](https://docs.google.com/spreadsheets/d/17spngWSIDffy1SucsOP8KpWZ3GQmPTnvHNFdH5t1_gw/edit?gid=580811579#gid=580811579). ## 1. Global AI compute Our projections draw heavily on a previous [compute forecast](https://docs.google.com/spreadsheets/d/1Ko-olwjDy6h8rXLZBFpP-e2GibGqRhfV28S4v8xtzrM/edit?gid=1866551567#gid=1866551567) by the [AI Futures Project](https://www.aifutures.org/), published in April 2025 as part of their AI 2027 scenario. The authors describe their assumptions in detail [here](https://ai-2027.com/research/compute-forecast). Importantly, the compute forecast does not presuppose their other predictions about future AI progress. In 2025, the AI Futures Project estimated that global AI compute will grow to [435 GW](https://docs.google.com/spreadsheets/d/1Ko-olwjDy6h8rXLZBFpP-e2GibGqRhfV28S4v8xtzrM/edit?gid=1866551567#gid=1866551567) by 2030. We assume, in line with their own forthcoming work, that actual growth will be somewhat slower. This is because supply chain bottlenecks, mainly around EUV lithography, are likely to decelerate AI compute growth – continuing the already observable [slowdown](https://epoch.ai/data/ai-chip-sales?view=graph&tab=power) from 3.4x annual growth historically to around 2.2x today to an (expected) 1.25x by 2031. Taking into account this slowdown, we estimate that the global supply of AI compute will reach around 370 GW by 2031 – still more than a tenfold increase of the approximately [30 GW](https://epoch.ai/data-insights/ai-datacenter-power) of global AI data centre capacity at the beginning of this year. *[Chart 1 — interactive figure; see the canonical HTML page.]* This is well within the upper bound of what the world’s EUV lithography machines – the most complex tool required to produce cutting-edge chips – would be physically capable of producing by 2031. Building on a recent estimate by chip researcher [Dylan Patel](https://www.dwarkesh.com/p/dylan-patel), we assume that by 2031, there will be around [700 EUV machines](https://docs.google.com/spreadsheets/d/17spngWSIDffy1SucsOP8KpWZ3GQmPTnvHNFdH5t1_gw/edit?gid=40304550#gid=40304550), in theory capable of producing 780-870 GW of AI compute from 2026 onwards. In practice, the effective output will be much less, as there are bottlenecks to chip production other than EUV (e.g. high-bandwidth memory or advanced packaging) and some EUV machines are used to produce non-AI chips. If just 50% of all EUV capacity went into AI chip production, these machines could in theory produce 390-430 GW of AI compute between 2026 and 2031. Our estimate of 373 GW is within this conservative upper bound. ## 2. Relative shares in the default scenario *[Chart 2 — interactive figure; see the canonical HTML page.]* We assume Europe’s current share of global AI compute to be roughly 5%, down from the 6-7% share that [analysts](https://epoch.ai/publications/trends-in-ai-supercomputers) previously reported for early 2025. In the default scenario, we estimate that Europe’s share will briefly increase to 8% in 2028 before decreasing again to slightly above 5% in 2031. This is based on a bottom-up estimate of European compute buildout over the next 5 years, drawing on public information about existing and announced AI data centres. Our [database](https://docs.google.com/spreadsheets/d/13QaT2TAQLjHDscwfJBBigtgoLWphyyVKf0pXFLGEV7Y/edit?usp=sharing) of publicly known or announced AI data centres (≥10 MW) in Europe shows that the European compute base will likely reach around 2 GW by the end of 2026 and almost 21 GW by 2031.1 While European AI compute is mostly concentrated in the Nordics today, our data indicates a shift by 2031, with the three largest compute owners by then being France (37%), Norway (9%) and Germany (9%). The largest AI clusters by 2031 include GW-scale projects in France and Portugal, and >500 MW facilities in the Netherlands, Romania, Norway, Germany, Italy and the UK. Our default estimate assumes that privately and publicly announced projects will mostly be realized and that European compute policy retains its current level of ambition. We assume that, by default, the compute share of the rest of the world (ex-US/China/Europe) will exceed Europe’s share and reach around 11% by the end of the decade. GW-scale announcements in several countries – UAE ([5 GW](https://openai.com/index/introducing-stargate-uae/)), Saudi-Arabia ([1.5 GW](https://www.datacenterdynamics.com/en/news/datavolt-plans-15gw-data-center-campus-in-neoms-oxagon/), [1 GW](https://newsroom.cisco.com/c/r/newsroom/en/us/a/y2025/m11/amd-cisco-and-humain-to-form-joint-venture-to-deliver-world-leading-ai-infrastructure.html), [1 GW](https://www.datacenterdynamics.com/en/news/saudi-telecom-company-signs-mou-with-humain-to-develop-1gw-of-data-center-capacity/)), India ([3 GW](https://techcrunch.com/2026/02/19/reliance-unveils-110b-ai-investment-plan-as-india-ramps-up-tech-ambitions/), [1 GW](https://www.adani.com/newsroom/media-releases/adani-and-google-partner-to-build-indias-largest-data-centre-campus-in-visakhapatnam), [1 GW](https://www.datacenterdynamics.com/en/news/digital-connexion-signs-mou-to-build-1gw-campus-in-andhra-pradesh-india/)), South Korea ([3 GW](https://www.tomshardware.com/tech-industry/artificial-intelligence/worlds-largest-data-center-gets-go-ahead-from-korean-govt-facility-to-require-3-gw-of-power), [1 GW](https://www.datacenterdynamics.com/en/news/south-korea-plans-1gw-data-center-campus-in-gangwon-province/)), Canada ([2.4 GW](https://thelogic.co/news/the-big-read/wonder-valley-data-centre-alberta-kevin-oleary/), [1.2 GW](https://www.datacenterdynamics.com/en/news/details-emerge-around-beacon-ais-planned-400mw-alberta-data-center-campuses/)) and Brazil ([1.5 GW](https://www.prnewswire.com/news-releases/elea-announces-rio-ai-city-a-landmark-brazilian-data-center-project-with-capacity-up-to-3-2-gw-of-renewable-energy-supporting-ai-growth-302449289.html), [1 GW](https://www.prnewswire.com/news-releases/scala-ai-city-brazils-ministry-of-mines-and-energy-approves-5-gw-power-connection-for-data-center-city-in-rio-grande-do-sul-302455610.html)) – alone make up a total of almost 24 GW, exceeding Europe’s projected share of 21 GW by 2031. Even if multi-GW projects face higher execution risk than in the US or Europe, the rest of the world’s share is still very likely to be greater than in Europe, especially when taking into account that several data centres in the multiple-hundred MW range are also planned in countries such as Australia, Malaysia or Japan. According to Epoch AI’s [chip owners](https://epoch.ai/data/ai-chip-owners?view=graph&tab=power) database, China’s current AI compute share is around 11% (when measured in GW).2 In line with forthcoming work by the AI Futures Project, we assume that China’s share will increase to around 15% by 2031. If China owns around 15% of global compute, Europe around 5% and the rest of the world (ex-US) around 11%, this leaves a bit less than 70% for the US in the default scenario. ## 3. Relative shares according to the epilogue *[Chart 3 — interactive figure; see the canonical HTML page.]* In the story’s epilogue we hint at a possible different future in which Europe successfully changes course on AI, including by pursuing a much more ambitious compute policy. In such a world, we assume that Europe would aim to reach a 15% share of global AI compute by 2031, tripling its current 5% share. In the medium term, it would put Europe on track to reach a compute share roughly proportional to its global GDP share ([~25%](https://en.wikipedia.org/wiki/List_of_continents_by_GDP#Continents_by_GDP_%28PPP%29) currently). Since building large AI data centres has multi-year lead times even in optimistic scenarios, most of such additional capacity would come online in 2030/2031, assuming Europe decided in 2026 to onshore a greater compute share. Building enough data centres would be a European effort, making use of suitable sites and grid connections wherever they are available, though compute would plausibly be concentrated in regions with stable, fossil-free energy sources: the Nordics (hydropower), France (nuclear) or Iberia (solar). Countries with decommissioned industrial areas, such as Germany or Poland, will also play a major role, as some of these sites have existing, GW-scale grid connections that could be repurposed for large AI data centres. Hosting more AI compute in Europe likely won’t affect the world’s total compute capacity: in light of [rapidly increasing](https://epoch.ai/gradient-updates/is-a-compute-crunch-coming) demand, the supply chain for AI compute is already producing chips at close to maximum capacity, irrespective of where they are shipped next. The main difference to the default scenario thus concerns Europe’s share relative to other world regions.3 We assume that an ambitious European compute strategy would mostly reduce the US share (by around 9pp), as Europe makes it more attractive for US AI companies to invest and leverages its supply chain position to obtain more US-made chips for sovereign compute projects in Europe. China’s share would also likely decrease, as more US chip exports go to Europe rather than China. However, the effect would be much smaller (around 1.5pp), as China is actively decoupling its compute stack from the West. Finally, the rest of the world (ex-US/China) would also have a slightly lower share, as chip exports that would have otherwise gone to countries like the UAE or Saudi-Arabia would go to Europe instead.
  1. Two limitations of our database are worth flagging: On the one hand, it may overestimate European AI compute, as merely announced data centers face execution risk and may not actually be built. On the other hand, it may underestimate European AI compute, as some plans for data centers may not be public yet. We might also have missed some existing or planned data centers, since a complete and publicly accessible list of European AI data centers doesn’t exist yet. Generic hyperscale and co-location data centers, not specifically designated as AI data centers and thus not tracked in our database, might also host some amount of AI capacity, though we expect this share to be small in comparison. Despite these limitations, we think our database provides the most comprehensive and up-to-date public overview of present and future European AI compute: (1) we’ve drawn from a wide variety of sources over several iterations (e.g. company press releases, industry publications, national and local newspapers, existing datasets such as Epoch AI’s [GPU clusters](https://epoch.ai/data/gpu-clusters?view=table&tab=point) database), (2) the biases above point into different directions and partly balance each other out and (3) our results broadly align with informal estimates from other experts and [previously observed](https://epoch.ai/publications/trends-in-ai-supercomputers) trends. ↩
  2. Note that measuring compute in GW rather than H100-equivalents overstates China’s compute share, as Chinese hardware is [less power-efficient](https://www.chinatalk.media/p/how-much-ai-does-1-get-you-in-china) than US alternatives. For example, when measured in H100-equivalents, China’s current share of global compute is just around [5%](https://epoch.ai/blog/introducing-the-ai-chip-owners-explorer#chinese-customers-own-just-5-of-global-ai-compute), compared to 11.5% when measured in GW. We still use GW as a metric for two reasons. First, it’s the more relevant metric for policymakers, who need to think about AI compute buildout in the broader context of energy policy and grid planning. Second, we’re mostly interested in how the relative shares of the US and Europe shift in the two scenarios. Here, GW is perfectly fine as a metric, since the US and Europe rely on the same, equally power-efficient hardware. ↩
  3. An ambitious European compute policy could also lead to data centres coming online *faster*, if European sites offered a faster ‘time to power’ than in other Western regions – though the exact magnitude and even the direction of this effect are unclear. (For example, some AI companies might prefer to build data centres in Europe even if time to power was a bit longer, e.g. for political reasons.) ↩
--- This Markdown is a machine-readable rendering of the canonical HTML page at https://europe2031.ai/compute-forecast, provided for AI agents and crawlers. The HTML page is the canonical, user-facing version; this file omits navigation, scripts, and styling. Citation: Europe 2031, by Daan Juijn, Stan van Baarsen, Judith Dada, Lily Stelling, Philip Fox, Alex Petropoulos, Michiel Bakker, Tom Chivers. https://europe2031.ai