// Compute
GOV BIS EXPORT CONTROLS // H20 chip class added restricted list — Q2 2026 effective MKT NVIDIA market cap elevated // compute demand pricing above Q3 2025 baseline GOV EU AI ACT // Enforcement timeline for high-risk systems active — April 2026 SYS COMPUTE // Global H100/H200 allocation constrained — hyperscaler backlog extends to Q4 2026 MKT AI CAPITAL // Frontier lab fundraising rounds total $40B+ — Q1 2026 GOV US EXECUTIVE ORDER // AI export control framework revision — draft circulating SYS ENERGY // US data center power demand projected +40GW by 2028 — grid planning active
Global Realist  /  Compute

Compute

Compute · Models · Policy · Capital · Infrastructure · Open Source
Elevated

The AI strategic domain is the fastest-moving axis of great-power competition in the current international order. Control of the AI stack — from semiconductor fabrication through training compute to deployed frontier models — is now treated as a national security variable by the US, China, and the EU simultaneously. The first-mover advantage in AI capability may be decisive across military, economic, and informational dimensions before governance frameworks can establish stabilizing norms. This is not a technology story. It is a power story.

Updated: 7 Apr 2026
← Home
// Domain Segments
Select a segment to load the focal panel
Segment 01
Compute
Hardware layer: GPU clusters, H100/H200 allocation, TSMC fab capacity
Critical
Segment 02
Models
Frontier capability race: GPT, Claude, Gemini, Qwen — gap narrowing
Elevated
Segment 03
Policy
Regulatory fragmentation: US export controls, EU AI Act, China AI governance
Elevated
Segment 04
Capital
$40B+ invested Q1 2026 — concentration at frontier, drying up at mid-tier
Elevated
Segment 05
Infrastructure
Data center buildout, energy demand, grid stress, water consumption
Elevated
Segment 06
Open Source
Llama, Qwen, Mistral — open weights accelerating diffusion including to adversaries
Watch
// Current Focal Segment
Compute
Infrastructure Layer Critical Watch
COMPUTE
Strategic Assessment

AI capability is now a direct function of compute access — and compute is concentrated, geography-bound, and increasingly militarized. The US, China, and EU are running parallel industrial strategies to control the hardware layer of the next power order. NVIDIA H100/H200 GPU clusters are the productive capacity of the current AI era — export-controlled by the US Commerce Department and allocated by hyperscalers. The entity that controls compute access controls the intelligence ceiling. TSMC controls 90%+ of leading-edge chip fabrication on a 36,000km² island 180km from mainland China.

Pressure Indicators
H100/H200 Allocation Constrained
TSMC Fab Capacity Critical Single Point
Export Control Coverage Elevated
Chinese Domestic GPU Progress Watch / Improving
US Domestic Fab Capacity Watch / Growing
Recent Developments
  • BIS added H20 chip class to export control list — Q2 2026 effective
  • NVIDIA market cap sustained above $2T on compute demand
  • Intel Gaudi 3 and AMD MI300X expanding competitive supply
  • China's Huawei Ascend 910C progressing — partial capability replacement
  • TSMC Arizona fab Phase 1 operational — limited capacity
// GR Interpretation
Compute is the oil of the AI era — and the US has weaponized the supply chain. BIS export controls on H100/H200/H20 chips represent the first explicit use of semiconductor access as a geopolitical instrument since COCOM. The strategic logic is sound: constrain Chinese AI capability at the hardware layer to maintain the capability gap at the model layer. The structural risk: China accelerates domestic alternatives while the control regime leaks through third-country routing, and the window to establish a decisive advantage closes.
Related Segments
Source Architecture

Intelligence Layers

Grouped by tier. What is happening · What others are saying · What Global Realist assesses — these categories must remain distinct.

// Tier 1 — Official
Anthropic
Frontier model developer — system cards, Responsible Scaling Policy, alignment research
Key reports
OpenAI
Frontier model developer — Preparedness Framework, GPT / o-series disclosures
Key reports
Google DeepMind
Frontier model developer — Frontier Safety Framework, Gemini disclosures
Key reports
NIST AI Safety Institute
US government AI risk framework and frontier-model evaluation body
Key reports
UK AI Safety Institute
Pre-deployment frontier-model evaluation — most credible state evaluator
EU AI Office
AI Act enforcement — GPAI obligations, high-risk system compliance
BIS / US Commerce
Export Administration Regulations — AI chip and compute controls
White House OSTP
Executive-branch AI policy guidance and interagency coordination
xAI
Grok release notes and frontier capability disclosures
Meta AI
Llama releases — the reference point for open-weight frontier models
// Tier 2 — Reporting
// Tier 3 — Think Tank
Georgetown CSET
Compute governance, export-control analysis, and US–China AI competition
Key reports
Epoch AI
Compute, training-run, and capability-scaling trend tracking
Key reports
Stanford HAI
Annual AI Index, capability benchmarks, policy research
Key reports
Apollo Research
Independent evaluations of frontier-model deception and scheming behavior
Key reports
RAND AI
Military AI, biosecurity misuse, and open-weight risk analysis
Brookings AI
AI governance, fragmentation, and democratic-implications analysis
AI Now Institute
Concentration-of-power critique — labor, surveillance, regulatory capture
MIRI
Foundational alignment research — long-horizon existential framing
// Tier 4 — Global Realist
// Tier 5 — Strategic Data
Domain Telemetry

AI Domain Pressure Indicators

Context-sensitive — AI stack stress telemetry across compute, capital, policy, and capability dimensions. Not a tech dashboard. System state only.

Compute Concentration
Critical
TSMC/NVIDIA dependency: 90%+ of leading-edge AI chips on a single island. Single point of failure for the global AI capability build.
// TSMC/NVIDIA dependency — structural
Export Control Effectiveness
Elevated
BIS H100/H200/H20 controls active. Third-country routing leakage ongoing. China domestic alternatives advancing — control window narrowing.
// BIS — active enforcement
Frontier Capability Gap — US vs China
Elevated / Narrowing
US maintains frontier model lead. DeepSeek demonstrated near-frontier reasoning at fraction of compute cost. Benchmark gap closing Q1 2026.
// RAND, CSET assessment — Q1 2026
AI Energy Demand
Elevated / +40GW
US data center power demand projected +40GW by 2028. Grid operators in Virginia and Texas issuing AI load warnings. Nuclear revival driven partly by AI demand.
// IEA, NERC — 2026 projections
Capital Concentration at Frontier
Elevated
$40B+ raised by frontier labs Q1 2026. Mid-tier AI funding compressing. Stargate $500B infrastructure commitment announced. Capital intensity creating entry barrier.
// PitchBook, lab disclosures — Q1 2026
Open Source Diffusion
Elevated
Meta Llama 3 weights downloaded 350M+ times globally. DeepSeek R1 open-weight. Open model layer porous in ways hardware layer is not — export controls cannot contain.
// Meta AI, Hugging Face — download metrics
Policy Fragmentation
Elevated
Three incompatible governance regimes: US innovation model, EU rights-based regulation, China state-aligned controls. No international AI arms control framework in negotiation.
// EU AI Act active — April 2026
Military AI Integration
Watch / Accelerating
US DoD AI adoption expanding. Chinese PLA AI integration opaque but assessed as accelerating. No international framework governing autonomous weapons or AI-enabled targeting.
// SIPRI, RAND — military AI assessments