The AI Race Beyond the US and China: Europe Sets Rules, Gulf Buys Compute, Japan and Korea Build Factories
The outcome of AI is no longer decided by frontier model scores alone. Chip export controls, EU compliance deadlines, and Japan/Korea's industrial AI push will determine where models can run, what hardware costs, and which standards win outside the US and China.
According to 2026 projections, global AI compute lands around 45 GW: the US holds 78 percent, China 11 percent, and the EU just 5 percent. That share gap is widening because the denominator grows roughly eightfold by 2031, so Europe's 27x expansion would only reach 15 percent share. The analysis works through what each region actually does given those odds.
Europe uses the AI Act as a time-buying instrument, pushing high-risk obligations to 2027-2028 and opening bids for up to seven AI gigafactories. The Gulf sells compute landlordship: Saudi HUMAIN committed ~$100B, UAE's Stargate UAE with G42, OpenAI, Oracle builds a 1 GW cluster. Japan and Korea skip frontier models and push physical AI into factories, with METI's 1.23 trillion yen and Korea's 550 trillion won for data centers. India runs a public compute pool at about one-third the global GPU price, Singapore builds governance, and the UK targets 20x public compute by 2030.
Share is a fraction with a growing denominator: Europe's 10x compute growth still leaves global share nearly flat, so Brussels frames its gigafactory spending around inference and data residency rather than training leadership.
Gulf compute is not sovereign: every chip purchase depends on US export approval, so the 'landlord' position buys capacity but not control over expansion.
Japan and Korea treat AI as an export multiplier for existing supply chains; models succeed on factory orders, not benchmark leaderboards.
India's public compute pool is smaller than one EU gigafactory's chip floor yet prices at one-third global rates, showing a strategy of developer habit and adoption over raw scale.