Steel, Silicon and Sovereign Compute: Why Countries Are Building AI From the Ground Up
Artificial intelligence is no longer just a software story. It is becoming a hard-infrastructure story — about where the data centers sit, who controls advanced chips, how much electricity is available, and whether a country has domestic access to the compute needed to train and run modern AI systems. The OECD says compute is now a key enabler — and a potential bottleneck — for AI, while the IEA notes that there is no AI without electricity for data centers. The OECD also finds the physical geography of AI compute remains highly uneven: in one pilot measure, the United States and China together hosted 49 public-cloud AI regions, almost as many as the rest of the world combined at 52. (OECD)
That imbalance helps explain why governments are moving from AI strategy papers to concrete spending plans. In Europe, the European Commission says its InvestAI initiative will mobilise €200 billion for AI investment, including €20 billion for up to five AI gigafactories, while the broader AI Continent plan lists 19 AI factories to support startups, industry, and research. Europe is also reinforcing the semiconductor layer: in February 2025, the Commission approved €920 million in German state aid for Infineon’s new semiconductor manufacturing facility in Dresden. (European Commission)
Canada has made the sovereignty argument explicit. Ottawa’s Canadian Sovereign AI Compute Strategy is backed by $2 billion over five years, with up to $705 million earmarked for a Canadian-owned, Canadian-located high-performance supercomputing system and up to $300 million through an AI Compute Access Fund designed to reduce the cost of compute and address limited domestic capacity. This is not just about supporting research labs; it is about making sure local firms do not have to rely entirely on foreign infrastructure for a technology that is becoming economically strategic. (ISED Canada)
The United Kingdom is following a similar path. Its Compute Roadmap says the government will invest up to £2 billion in a national compute ecosystem, including over £1 billion to expand the AI Research Resource twenty-fold by 2030 and up to £750 million for a new national supercomputer in Edinburgh. Alongside that, the UK says its Sovereign AI Unit is backed by up to £500 million, with compute designated as a priority area because control over infrastructure is now seen as part of long-term strategic advantage. (GOV.UK)
India’s approach shows that sovereign compute is not only about frontier labs; it is also about broadening access. The government says the IndiaAI Mission, launched in March 2024 with an outlay of ₹10,372 crore, has already onboarded more than 38,000 GPUs for a common compute facility. A February 2026 government note added that 1,050 TPUs had also been onboarded and that GPU access was being offered at ₹65 per hour, which it described as roughly one-third of the global average cost. In practice, that means national AI policy is being translated into subsidised, shared infrastructure for startups and academia. (Press Information Bureau)
Elsewhere, the buildout is increasingly tied to industrial policy and sovereign capital. The White House said in May 2025 that Saudi-backed DataVolt planned to invest $20 billion in AI data centers and energy infrastructure in the United States. Reuters later reported that the UAE-backed “Stargate UAE” campus in Abu Dhabi was designed for 5 gigawatts of AI data-center capacity, and that South Korea on February 27, 2026 signed a deal with Hyundai Motor Group involving about 9 trillion won, including 5.8 trillion won for an AI data center with 50,000 GPUs. Japan, meanwhile, continues to subsidise semiconductor capacity: Reuters reported that Tokyo’s support for TSMC in Japan could push taxpayer-funded subsidies beyond 1 trillion yen, while Rapidus sits at the center of a broader $65 billiongovernment plan to revive the country’s chip and AI industries. (The White House)
The big fact is this: AI policy is becoming infrastructure policy. Countries are no longer competing only on models, apps, or talent. They are competing on land, power, semiconductors, supercomputers, and domestic cloud capacity. The OECD’s recent work makes clear why: compute is concentrated, access is uneven, and location matters for economic development, supply-chain security, governance, and resilience. The race to build AI is increasingly a race to own part of its physical backbone. (OECD)
