As AI faces power shortages and public backlash within the United States, the next destinations for computing infrastructure are seeing accelerated migration southward and northward. Anthropic is betting 15 billion US dollars on Australia, while Meta is investing 10 billion US dollars to expand into Canada. Will this massive infrastructure migration involving tens of billions of dollars reshape the global computing power landscape?
Amid mounting difficulties, AI infrastructure is shifting its footprint away from the United States.
After the Digital Gateway data center project in Virginia, once billed as the world’s largest, was scrapped following years of lawsuits and protests from local residents, a growing number of tech giants are turning their sights to overseas markets.
As recently revealed by the Australian Financial Review (AFR), leading AI firm Anthropic is looking to secure a minimum of 1.4 gigawatts of data center computing power in Australia with a massive $15 billion land and facility investment, aiming to bring 1 gigawatt of capacity online by the end of next year.
The scale of this investment is staggering. After all, the total installed capacity of currently operational data centers in Australia is approximately 1.4 gigawatts.
If Anthropic’s investment is ultimately implemented, the installed capacity from its single procurement will be roughly equivalent to Australia’s total data center installed capacity.
Not far behind, Meta recently announced a roughly $10 billion investment to build a data center in Alberta, Canada, with an installed capacity of 1 gigawatt, equal to the power consumption of 750,000 households.
Once completed, this data center will become Meta’s largest overseas data center, supported by an additional investment of approximately 60 million Canadian dollars in local infrastructure including roads and water utilities.
Beyond vast, resource-rich countries like Canada and Australia, the Middle East, Africa and Europe have also become hotspots for heavy computing investment by major U.S. AI companies.
The successive multi-billion-dollar overseas infrastructure layouts by tech giants send a clear signal: the marginal growth of AI infrastructure is spilling beyond domestic borders, and the computing infrastructure competition among large model companies has officially gone global.
This is no ordinary data center construction plan, but a high-stakes bet on the locations for the training, management, operation and commercialization of next-generation cutting-edge artificial intelligence.
01
Escaping the Domestic Bottlenecks of the United States
Behind the overseas expansion of tech giants lies the increasingly severe practical bottlenecks in the construction of data centers within the United States.
The power shortage constitutes the primary bottleneck. With the vigorous development of data center construction, the power demand in the United States has grown explosively. However, the growth of U.S. power demand has remained almost stagnant for more than a decade, with an annual growth rate of less than 1%.
In its latest report released on July 8, Bank of America stated that the United States may face a power shortage of 100 gigawatts between 2026 and 2030, driven by the booming production and surging demand for chips as well as the inability of U.S. utility companies to meet current needs.
Bank of America analysts predict that the demand for power capacity will reach 230 gigawatts or more from 2026 to 2030. Nevertheless, the bank estimates that the power supply from utility companies will only amount to 93 gigawatts.
Data from the Electric Power Research Institute also shows that in Northern Virginia, a global hub for data centers, data centers already consume approximately 25% of the United States' total electricity, and this proportion is projected to soar to 57% by 2030.
The expansion of onshore data centers in the United States has hit a snag, dealt a crippling blow by the NIMBY effect.
Soaring electricity bills driven by AI are turning data centers into the "least desirable neighbors".
A recent U.S. poll indicates that 70 percent of Americans oppose the construction of AI data centers near their residences, with 48 percent strongly opposed and merely 7 percent in favor of building such facilities in their own neighborhoods.
According to the latest figures from AI research firm Data Center Observatory, the number of active opposition groups has more than doubled in the first three months of this year, surging from 396 at the end of 2025 to 833 across 49 U.S. states.
These community groups have successfully blocked or postponed no fewer than 75 relevant projects with a total value of approximately 130 billion US dollars (equivalent to around 880 billion Chinese yuan).
More importantly, the grassroots wave of opposition has rapidly evolved into legislative games at the regulatory level.
Virginia has passed the nation's first electricity consumption tax for data centers, and New York State has imposed a one-year moratorium on approvals. In just the first six weeks of 2026, more than 30 states across the United States have proposed over 300 relevant bills.
Tech giants are more acutely aware than ever that building a diversified computing power supply chain is crucial in the AI competition.
02
Moving South and Moving North
The overseas layouts of Anthropic and Meta reflect two major directions of the migration of U.S. data center infrastructure: moving south and moving north.
As generative AI training and inference workloads become increasingly tolerant to latency and gain greater flexibility in site selection, Australia, with its unique advantages, is emerging as a new hub for computing power.
Geographically located between the United States, Asia and the Pacific region, Australia has the potential to become a hub for computing power centers in the Asia-Pacific region.
According to the 2025 annual report released by Knight Frank, Australia ranks second worldwide (only behind the United States) in the global ranking of data center investment destinations.
Anthropic’s consideration of expanding its presence in Australia in the Southern Hemisphere stems from the country’s abundant land resources and rich renewable energy endowments.
A computing power scale of 1.4GW is equivalent to the output of several nuclear power units. Australia’s relatively stable energy supply and favorable climate serve as inherent natural advantages.
More importantly, relevant policies have paved the way. In March this year, Anthropic signed a memorandum of understanding with the Australian government to cooperate on AI safety research and the national AI plan, removing obstacles for the large-scale rollout of infrastructure.
Meta also made a well-thought-out site selection in Sturgeon County, Alberta.
The province’s key advantages include low-cost natural gas, a relatively cool climate, and permission to build on-site power supplies, which allows tech companies to bypass capacity constraints of the public power grid.
According to Reuters, Meta's initial project scale in Canada is 1GW, with the capacity to expand to 1.8GW. The power supply mainly relies on natural gas power generation.
Meta will fund the construction of new power generation facilities to connect to the power grid and has signed long-term energy supply agreements with multiple energy enterprises.
This underscores Alberta's unique appeal: the province is transforming AI data centers into a new export avenue for its natural gas industry.
Alberta boasts abundant low-cost natural gas, mature energy engineering capabilities, cooling advantages brought by its cold climate, and a relatively business-friendly and tax-friendly environment.
The Prime Minister of Canada has personally endorsed the initiative, vowing to build Canada into "the world's best location for data center construction".
Regulatory certainty and accessible energy have jointly created a "lowland effect" for computing power investment in the region.
03
The Next Hotspot for Computing Power?
In fact, even before tech giants set their sights on Australia and Canada, the Middle East, Europe and even Africa have once emerged as popular destinations for computing power investment.
Leveraging abundant capital and energy advantages, the United Arab Emirates and Saudi Arabia are striving to become the "new oilfields" of global AI computing power.
Amazon announced in 2024 a new data center project worth over 10 billion US dollars in Saudi Arabia; Microsoft stated in 2025 that it would invest more than 15.2 billion US dollars in the United Arab Emirates by 2029; OpenAI also announced in 2025 that it would build the 1-gigawatt "Stargate UAE" data center in Abu Dhabi.
Nevertheless, the region is now facing severe challenges. In March 2026, three Amazon data centers in the United Arab Emirates and Bahrain suffered drone attacks amid regional conflicts, resulting in service outages.
Facilities of Google, Microsoft, NVIDIA and other companies have also been listed as potential targets. This incident may prompt US tech giants to adopt a more cautious approach to future investments in the region.
The construction of AI data centers in Europe is mainly driven by the European Union's "InvestAI" initiative, which aims to triple computing power within the next 5 to 7 years compared with the current level.
The largest single project to date is the 1 billion-euro partnership between NVIDIA and Deutsche Telekom. However, Jensen Huang has pointed out that the EU still lags behind China and the United States in AI investment, highlighting the urgency for Europe to accelerate infrastructure development.
The African market is widely regarded as the next growth hotspot, yet its current data center capacity accounts for less than 1% of the global total. Major US tech giants are entering the market through partnerships with local enterprises or independent construction, though their projects remain small in scale and fraught with implementation challenges.
Microsoft once planned to build a 100-megawatt data center in Kenya. Nevertheless, the project is currently under review due to its enormous power demand — the first phase alone would occupy approximately 3% of the country’s total installed power capacity — and unresolved government guarantee issues.
Anthropic’s massive investment in Australia marks the country’s emergence as the next computing power hub in the global AI competition.
However, great opportunities do not come without costs.
Research cited by Australia’s Climate Council indicates that if the growth of data centers is not matched by new renewable energy capacity, the average wholesale electricity price on Australia’s main grid could rise by more than 20% by 2035, with greater pressure in regions such as New South Wales and Victoria. Water resources and community acceptance will also serve as constraints.
More importantly, if Australia’s role is merely to host data centers, supply electricity, provide labor and bear environmental burdens while the majority of the value flows overseas, it may not be able to build strong influence in the AI competition.
Canada faces similar challenges.
Alberta’s core advantage lies in inexpensive natural gas, which also creates a key contradiction: the "clean computing power" touted by AI giants does not always align with the actual marginal power sources of the projects.
Meta claims its electricity consumption will be fully matched by 100% clean and renewable energy. However, Reuters has pointed out that the emission intensity of Alberta’s power grid is significantly higher than Canada’s national average.
Meanwhile, a June report from the Canadian Broadcasting Corporation (CBC) also noted that large-scale data centers exert environmental impacts on surrounding communities in terms of carbon emissions, water consumption and noise pollution, with related controversies still ongoing.
It is undeniable that in the second half of the global computing power competition, the contest is no longer about who can purchase the most advanced chips first, but about who can make computing power take root and settle down with lower institutional costs and higher system efficiency.
Anthropic and Meta are merely the starting point of this "great exodus".
¡Su mensaje debe tener entre 20 y 3.000 caracteres!
