The decade-long diversification of global tech investment is over. In 2020, U.S.-based startups accounted for roughly 40% of global venture funding; by 2025, that figure had surged past 75%, driven almost entirely by artificial intelligence. A sweeping analysis by Rest of World documents how the AI boom has reversed years of progress in distributing startup capital beyond Silicon Valley, concentrating funding, talent, and physical infrastructure in the United States at a scale that dwarfs anything the venture industry has seen before.
A golden age reversed
Between 2016 and 2021, something remarkable happened: non-U.S. startups grew their share of global venture funding from approximately 45% to nearly 60%, according to Crunchbase data. Venture-backed companies outside the U.S. raised more than $300 billion in 2021 alone, up from roughly $80 billion in 2016. That era produced breakout companies across the Global South — Indonesia’s Gojek, which grew from a motorbike ride-hailing app into a $10 billion decacorn, Brazil’s Nubank, which reached a $45 billion IPO valuation, and India’s Flipkart, which attracted a $16 billion investment from Walmart in 2018.
By 2024, the U.S. had recaptured the top position in startup funding, pulling in roughly $130 billion compared to $75 billion for the rest of the world. By 2025, the gap had widened to more than 3-to-1.
The numbers behind the concentration
The structural driver is AI. According to an OECD analysis, U.S. AI firms attracted 73% of global AI investment last year — approximately $120 billion — representing more than half of all global venture funding across every industry. The mega-rounds tell the story in miniature: in early 2026, Anthropic raised $30 billion at a $380 billion valuation, followed weeks later by OpenAI’s $40 billion round. Those two deals alone exceeded total venture funding raised across all of Africa, Southeast Asia, and Latin America in 2025 combined.
Since 2023, more than 5,000 venture-backed AI companies have been founded in the U.S. — roughly three times the number in the rest of the world combined, according to PitchBook data. The top 20 global AI investors directed an estimated 80% of their capital to U.S. AI companies last year. AI now accounts for more than 55% of all private venture funding globally, up from around 20% in 2021.
Physical infrastructure as a moat
The concentration reflects something deeper than investor preference. Foundational AI development demands data centres stacked with expensive, hard-to-source chips, massive amounts of water, and reliable electricity — resources most countries simply cannot marshal at the required scale. Africa hosts just 1.3% of the world’s data centre capacity, according to Cloudscene. The continent has seen fewer than 100 AI startups founded since 2023, raising a combined $400 million — less than a single Series B round for a mid-tier U.S. AI company. Meanwhile, the U.S. alone has committed more than $150 billion to new data centre construction through 2027, according to McKinsey, reinforcing the infrastructure advantage that makes training frontier models outside North America increasingly impractical.
The “sovereign AI” paradox
Even countries with significant advantages are struggling. India, with 1.5 million developers graduating annually and $1.2 billion in government AI investment pledged through 2027, has watched several promising AI startups collapse. Rest of World documents a pattern of once-hyped Indian AI firms — including speech-recognition company Uniphore’s domestic rivals and enterprise AI startup Niki.ai — closing due to funding shortages, while American competitors like OpenAI and Anthropic expand into the region with war chests 100 times the size of local players. Indian AI startups raised just $2.4 billion in 2025, compared to $120 billion for their U.S. counterparts.
Dipayan Ghosh, a fellow at the Harvard Kennedy School, has described the emerging dynamic as “structurally colonial,” with the growing movement among nations to build independent AI ecosystems — so-called “sovereign AI” — facing a hard reality: without access to frontier chips, training compute, and multi-billion-dollar funding pipelines, these efforts fail to close the gap with a small number of U.S. and Chinese firms that control the foundational model layer.
What this means
The pattern is worth examining alongside the broader question of how global innovation flows are shifting. Countries building AI applications on top of American and Chinese foundational models remain structurally dependent on those providers — exposed to shifting export controls, licensing terms, and geopolitical pressures. The technology often promoted as a great equaliser has, in practice, produced the most geographically concentrated investment cycle in modern venture capital history — one where two companies can raise $70 billion between them while entire continents fight over scraps.
Feature image by Tima Miroshnichenko on Pexels
