Ongoing · Technology / Energy
These are algorithmically-created hypotheses — not forecasts.
The central question is how long the compute and power bottleneck constrains AI buildout. The branches suggest a sustained supply crunch — extended GPU lead times and power-grid lag — is the most plausible path, with stricter regulatory constraint as a secondary brake and an efficiency breakthrough that relieves the bottleneck as the lower-probability scenario. Resolution likely depends on fab and grid capacity additions and on whether a step-change in model efficiency materialises.
Authored 2026-05-21 · OpenWatch editorial
Two consecutive Nvidia datacenter-segment earnings prints below consensus by more than 15%, paired with public hyperscaler capex guidance cuts of 10%+ — would refute the "sustained supply crunch" framing and signal the AI-infrastructure cycle has peaked.
Each branch below shows the most likely ways this plays out — with its own winners, losers, and supporting signals.
View possible paths ↓AI-generated hypothesis. Not investment advice. Always verify independently with a qualified financial advisor.
Public prediction markets matched by AI to this scenario — agree or disagree, the bet is yours. OpenWatch does not recommend any position.
AI industry downturn triggered by capex overcapacity and GPU overbuild; resolves when three of five conditions (including NVIDIA revenue decline, AI stock drops, funding collapse) occur within 90 days.
Frontier model capability milestone directly measures AI model performance advancement on standardized benchmark, core indicator of frontier model race progress.
Frontier model competition via Epoch Capabilities Index, measuring advancement of leading AI models in the race.
Timing of AI sector correction triggered by capex cycle overshoot and subsequent crash. Core trigger for branch resolution.
EU AI Act enforcement action against frontier AI labs directly instantiates regulatory constraint on AI infrastructure development and deployment in 2026.
AI bubble pop timing. Capex overshoot and crash cascade branches on this bubble-burst signal in AI infrastructure sector.
Market prices are raw values. Political contracts may exhibit favourite-longshot bias.
If this scenario occurs — possible paths
Signal counts measure media attention over the last 7 days — not the likelihood of an outcome.
Branch % = conditional on this scenario occurring · Path % = joint probability of this exact path from today
Trade lens —NVDA and ASML sustain pricing power; data-center and nuclear-PPA names bid on baseload scarcity; INTC AI-chip share compresses. · meaningful · slow
Policy lens —FERC opens an emergency rulemaking on data-centre power interconnection priority; the Commerce Department publishes updated AI Export Administration Regulations restricting advanced-node GPU exports; the EU AI Office activates its first systemic-risk designation under the AI Act.
Trade lens —AI-compliance names (PANW) and incumbents (MSFT) capture moat; NVDA capex pace decelerates at the margin; European industrial-AI adoption slows. · meaningful · slow
Policy lens —The EU AI Office activates systemic-risk obligations for GPAI model providers under the AI Act; the White House issues an updated Executive Order on AI safety incorporating mandatory model-evaluation thresholds; NIST publishes binding AI Risk Management Framework standards for critical-sector deployment.
Trade lens —NVDA and ASML repriced lower on demand-shock; application-layer names (MSFT) capture value; Taiwan/Netherlands semis-capex multiple compresses. · structural · slow
Policy lens —FERC cancels priority interconnection orders for seven large datacentre projects citing changed demand assumptions; the Commerce Department revisits GPU export-control thresholds as the compute-per-dollar bottleneck softens; Congressional appropriators request a revised national AI-compute infrastructure assessment from OSTP.
Editorial framing — events outside our X→Y→Z partition. Authored as paired 'what if positive' / 'what if negative' to capture asymmetric tail outcomes. No probability is assigned; the lean indicator is directional only.
An open-weight model with 10-100x lower inference cost matches frontier capability; the compute-supply premium collapses, AI capex shifts from "more GPUs" to "more applications", and AI-deployment economics broaden dramatically.
A high-profile AI-system failure (autonomous vehicle pile-up, medical-AI mass-misdose, or AI-orchestrated cyber incident with casualties) prompts emergency regulation that halts frontier deployment for 6-12 months across multiple jurisdictions.
Low-probability outcomes that do not belong to the conditional partition above. Surfaced alongside, never ranked, never given a probability. See the card for the trigger mechanism and the names that move if it materializes.
Mechanism: Cluster-scale training pauses or sharply slows across the three jurisdictions simultaneously, collapsing the leading-edge GPU pricing power and shifting capital toward inference / specialized accelerators and on-prem deployments.
Within a 6-month window, the US, EU, and China each independently impose binding restrictions on frontier-model training or deployment — different motivations (US security review, EU AI Act enforcement, China data-sovereignty), same effective constraint on hyperscaler training-cluster economics. Outside the modeled partition because the existing branches assume asymmetric regulation; a synchronized one would invalidate the GPU-bottleneck thesis itself.
Contingency note — Watch for any reciprocal-precedent language in regulatory proposals across the three blocs in the same quarter. Synchronization risk is usually telegraphed in advance via trade-policy committee minutes.
Mechanism: Emergency executive action across multiple jurisdictions imposes deployment restrictions on frontier models in 30-90 days, accelerates licensing requirements, and shifts capital sharply toward safety / monitoring / specialist providers rather than scale.
A high-confidence, attributed AI-driven attack on critical infrastructure (power grid, financial-market microstructure, election integrity, or hospital systems) creates a forcing function — different from the synchronized regulatory crackdown branch because this is reactive, faster, and produces emergency executive orders, not multi-year rule-making. The partition assumes orderly buildout; this assumes a Three-Mile-Island moment for AI deployment.
Contingency note — Watch for any confirmed-attribution AI-attack disclosure from CISA, NCSC, or ENISA. Emergency-rule timelines telescope when an incident is publicly attributed.
Fewer than 5 historical episodes — tilts are indicative only. Use with extra caution.
Based on 4 distinct escalation episodes 2018–2019 and 2025 (Section 301 tariffs, Huawei entity list, chip export controls, 2025 tariff spiral); sector returns measured over the 3-month window following each escalation announcement.
Countries and companies most at risk or with most upside across this scenario overall
Information cutoff: 2026-05-21 · Authored: AI-generated, council-reviewed · Live signal counts updated hourly