Seven Mental Models to Understand the AI Compute Era

Key Takeaways

• AI compute capacity grew 8.5× from Q1 2024 to Q4 2025.
• Growth measured in H100‑equivalent units, not just chip counts.
• Control of power infrastructure determines long‑term AI leadership.
• Seven mental models map infrastructure, supply chain, and geopolitical risks.
• Strategic bets on compute sovereignty shape relevance through 2029.

Pulse Analysis

The AI arms race is being reframed from a focus on model size and benchmark scores to the physical substrate that powers those models. Between early 2024 and late 2025, the industry added roughly 19 million H100‑equivalent units, a surge that dwarfs traditional chip‑count narratives. This expansion is driven by massive data‑center builds, power‑grid negotiations, and strategic partnerships that lock in the electricity and cooling capacity needed for petaflop‑scale workloads. Understanding this shift is essential for any stakeholder who wants to gauge true competitive momentum in artificial intelligence.

To make sense of the infrastructure surge, the author proposes seven mental models: compute sovereignty (ownership of power assets), flywheel coupling (how compute and data reinforce each other), export control (regulatory barriers on high‑end hardware), infrastructure arbitrage (locating compute where costs are lowest), consolidation pressure (mergers among cloud and chip providers), standardization (common APIs and hardware interfaces), and geopolitical moats (national policies that protect domestic AI ecosystems). Each model offers a lens to evaluate risks and opportunities that are invisible in headline‑grabbing model releases but critical for long‑term strategic planning.

For investors and corporate strategists, these models translate into actionable insights. Companies that secure renewable‑energy contracts, diversify supplier bases, or invest in proprietary cooling technologies gain a durable edge. Policymakers can shape the competitive field by adjusting export restrictions or incentivizing domestic chip fabs. As compute capacity continues to multiply, firms that align their roadmaps with these mental models will be better positioned to capture market share, command higher valuations, and influence the next wave of AI breakthroughs.