The Chart: Who Is Spending What
AI investment by bloc, 2024–2026
Private AI investment vs. announced public/sovereign commitments, USD billions. Public figures are multi-year pledges, not annual budgets — compare categories within a series, not across. Partly substantiated
Read the gap between the bars as the story: the US lead is private and corporate; everyone else's play is state-led. Whether state capital can substitute for an ecosystem is the open question of the decade.
The Players
$725 billion in one year
Microsoft, Alphabet, Amazon, and Meta plan roughly $725B of capital expenditure in 2026, nearly double 2025, almost all for AI infrastructure. Add Stargate (the $500B OpenAI–SoftBank–Oracle project) and the US effort is the largest peacetime industrial mobilisation since the interstate highway system. Company guidance · CNBC
The policy posture is acceleration plus containment: export controls on chips, investment screening outbound, and a federal push to pre-empt state AI laws. The risk is the one this whole site is about — the spend is corporate, debt-assisted, and predicated on demand that has not yet been proven at scale.
The DeepSeek shock
On 20 January 2025, DeepSeek released R1, an open-weights reasoning model that matched US frontier performance at a claimed fraction of the training cost — built on constrained hardware under US export controls. NVIDIA lost roughly $600B of market cap in a single session, the largest one-day wipeout in US corporate history. Market data, 27 Jan 2025
China's strategy is forced autarky: Huawei's Ascend line on SMIC's ~7nm process, state-directed compute pools, and a talent base that still produces more AI PhDs than any country. The export-control regime narrowed China's ceiling — DeepSeek proved it did not eliminate the floor.
The AI Act
The world's first comprehensive AI law phased in through 2025–2026: banned practices (February 2025), general-purpose model obligations (August 2025), high-risk system rules (August 2026 for embedded products). Brussels traded speed for sovereignty-by-regulation, betting that rule-setting power compounds like the GDPR effect. European Commission
The counter-argument is visible in the numbers: Europe's share of private AI investment remains in single digits, and its largest AI champions are increasingly American-owned or American-dependent.
IndiaAI Mission
The ₹10,000-crore (~$1.2B) IndiaAI Mission funds compute access, datasets, and a sovereign model ecosystem, layered on the world's largest English-speaking engineering workforce. India's play is not frontier research — it is becoming the deployment and data layer of global AI, the way it became the back office of global IT. MeitY · IndiaAI
Petrodollars to petaflops
The UAE's G42 and MGX, and Saudi Arabia's Public Investment Fund and HUMAIN, are converting oil wealth into compute, stakes, and campuses — including a announced 5GW AI data-centre corridor. The Gulf is the one bloc where state capital and energy abundance coincide: the two inputs the rest of the race is short on. Company announcements · FT
The complication is alignment: US controls have already restricted G42's access to American chips over China-linked ties, making Gulf AI a live theatre of the US–China contest.
From Bletchley to the AISI
The UK hosted the 2023 Bletchley Park AI Safety Summit and built the AI Safety Institute (AISI) into the reference model for frontier-model evaluation — since rebranded and folded into a broader AI Security Institute. With DeepMind and ARM on its soil and no hyperscaler capex of its own, Britain's strategy is convening power and evaluation standards. DSIT · AISI
Manufacturing the race
Japan is underwriting TSMC's Kumamoto fabs (JASM) and Rapidus's 2nm bid in Hokkaido — a ~¥1 trillion state commitment to repatriate leading-edge logic. Korea anchors the other chokepoint: SK Hynix and Samsung make the high-bandwidth memory without which no AI accelerator runs, and HBM has become Korea's single most strategic export. METI · company filings
The Perception Inversion
In Pew Research Center's 2025 survey of 24 countries, for the first time in the survey's history, more respondents viewed China favorably than the United States — a reversal driven largely by collapsing US favorability rather than rising Chinese favorability. Pew Research Center, 2025
AI leadership is not only compute; it is whose stack the world adopts. If the Global South's default technology partner tilts, export controls lose enforcement partners, Chinese open-weights models gain deployment surface, and the "AI Washington consensus" — the assumption that American platforms are the default — quietly expires. The DeepSeek moment was technical; the Pew moment is political. They are the same trend.
Confidence Assessment
| Claim | Status | Confidence |
|---|---|---|
| US hyperscaler AI capex reaches ~$725B in 2026 | Company guidance; subject to revision each quarter | Partly substantiated |
| DeepSeek R1 matched frontier performance on constrained chips | Benchmark evidence published; training-cost claims disputed but direction confirmed | Substantiated |
| Export controls have failed to constrain China | Frontier capability demonstrated; capacity ceiling still lower than US access | Unresolved |
| The EU AI Act will shape global standards | GDPR precedent supports it; enforcement record so far is thin | Unresolved |
| Gulf sovereign funds can buy AI sovereignty | Capital committed; talent and energy advantages real, alignment risk unresolved | Unresolved |
| Global opinion has tilted from US toward China | Pew 2025: first favorable-China majority in survey history | Substantiated |