· The Geopolitical Lens
Where the gigawatts actually sit.
Frontier AI clusters announced or under construction by May 2026. Marker size scales with planned 2027 capacity in megawatts. Geography is not flat.
Compute lead, energy gated.
42 GW announced. Permitting cycle averages 38 months. Texas grid interconnect queue measured in years.
Sovereign stack, fast build.
38 GW under construction. State led permitting averages 9 months. Compute parity targeted by Q4 2027.
Regulation clarity, capacity gap.
9 GW announced. AI Act in force. The only mature accountability regime. Capacity follows policy by 24 months.
What each region actually requires.
Three different regimes, three different operating realities. A workforce across all three must plan in months, quarters, and years simultaneously.
| Dimension | United States | China | European Union |
|---|---|---|---|
| Primary regime | Sector led Executive orders, NIST guidance, state laws (CA, CO, NY). |
State led Generative AI rules, deep synthesis law, content moderation. |
Horizontal AI Act (Aug 2026 high risk). GDPR. Data Act. DSA. |
| Time to deploy | 4 to 6 months | 3 to 5 months (with license) | 9 to 14 months (high risk) |
| Workforce disclosure | Limited. State patchwork emerging. | Mandatory for state affecting use. | Mandatory for high risk. Worker impact assessments required. |
| Talent visa friction | High and rising. H-1B caps unchanged. | Low. State sponsored returnee schemes. | Moderate. EU Blue Card reform pending. |
| Energy permitting | 38 months average | 9 months average | 52 months average |
| Implication for HR | Move fast, lobby state by state. | Localise stack, hire returnees. | Document everything. Audit ready by default. |
Capex by region, 2026 versus 2027.
Total AI infrastructure committed or contracted. The gap is widening, not converging. By 2027 the US runs at roughly 5x Europe and 2x China by absolute dollar.
2026 committed
2027 forecast
Energy is the binding constraint.
Building a power plant is only half the job. The plant must then be wired into the grid: the network of cables and substations that carries its electricity to the data centres. That hook-up needs its own permits and construction, and the waiting list for it runs to years.
Which region, on which constraint?
On record dates from frontier lab leadership and policy makers. The dispersion is wide. The strategic question is no longer "when" but "which region, on which axis".
Where the engineers actually go.
Net AI talent migration in 2025 to 2026. The arrows reveal where the strategic bench is forming and where it is leaking.
- Bay Area still draws 31% of frontier hires globally
- UAE based labs now compete head on with US comp packages
- Average frontier comp: $880K total
- 1.8M open roles, 1.2M active candidates
- State sponsored returnee schemes accelerated post 2024
- Returnees lead 40% of frontier labs in Beijing, Shanghai
- Average frontier comp: $420K equivalent
- 2.4M engineer pipeline (largest by absolute count)
- Most leave for US frontier labs within 24 months of PhD
- 440K open AI roles vs only 230K active candidates
- EU Blue Card reform delayed to 2027
- Average frontier comp: 165K euros
Reading the three regions together.
The evidence points to three different operating realities, not one race with a single leader. Each region is gated by a different binding constraint, and that sets its pace of change.
One race, three constraints, three tempos.
United States
Leads on compute and capital but is gated by energy. With grid interconnect queues measured in years, the binding constraint is megawatts, not dollars. The source material frames this as the defining bottleneck on US pace through 2027.
China
Trails on absolute capital but compresses time. State led permitting and rapid energy build let projects move from plan to power in months. On the source's account this is a speed advantage, with chip access the open question.
European Union
Lags on compute and capacity but leads on rules. The AI Act gives it the most defined accountability regime, while capacity tends to follow policy on a multi year lag. The trade off is clarity against pace.
read across three regimes.
Speed is gated by power.
Compute and capital lead, but energy permitting sets the ceiling. Capacity decisions track the grid queue more than the budget.
The stack is localised.
Returnee flows and a sovereign stack define the talent supply, and fast permitting compresses the build timeline. Chip access remains the open variable.
Rules arrive before capacity.
The AI Act gives the clearest accountability regime, with high risk provisions in force from August 2026. Capacity follows policy on a multi year lag.
One of five briefings.
Distilled from over 250 episodes of Moonshots with Peter Diamandis and the Moonshot Mates. Forecast dates are the source's, and are marked as forecasts.