Compute
Compute Geopolitics
Chips, power, and model access are strategy.
TSMC holds ~72% of global foundry market share and produces ~90% of the world's most advanced chips. U.S. export controls restrict China's access to advanced AI semiconductors. China's domestic chip share reached 41% in 2025, but it remains years behind at leading-edge nodes. Compute access is now a geopolitical variable, not a procurement detail.
Pressure index by operating layer
Signal concentration
Capitalized attention split
Problem to company flow
What changed
The semiconductor supply chain has become a primary instrument of geopolitical competition. U.S. export controls restrict advanced chip sales to China, while TSMC's Arizona, Japan, and Germany fabs are being built to reduce single-point-of-failure risk in Taiwan. China's domestic semiconductor market share hit 41% in 2025, but leading-edge node capability remains years behind. The proposed U.S. MATCH Act would further restrict equipment sales, extending jurisdictional reach over allied nations' chip industries. AI companies now face compute access as a strategic planning variable alongside talent, capital, and market access.
What leaders should do
Map your compute dependency chain: which cloud regions use which chip architectures, manufactured where, with what export control exposure? Score model providers by geographic risk. Evaluate multi-cloud strategies not just for cost and latency but for geopolitical resilience. Build contingency plans for compute disruption scenarios — a Taiwan Strait crisis, expanded export controls, or foundry delays could constrain AI workloads within quarters, not years.
What ZOAK wants to build
A compute risk mapping product for AI-dependent enterprises: it scores model providers, cloud regions, chip sourcing, energy access, and policy exposure in a single dashboard. The output is a quarterly compute risk report and a set of automated alerts when policy changes affect the enterprise's AI infrastructure stack.
Operating analysis
TSMC's dominance (~72% foundry share, ~90% at advanced nodes) makes it a single point of failure for the global AI supply chain. The company's expansion outside Taiwan — fabs in Arizona, Kumamoto (Japan), and Dresden (Germany) — is a hedge, but new facilities take 3–5 years to reach full production. In the meantime, any disruption to Taiwan-based manufacturing would cascade through every major AI lab, hyperscaler, and device manufacturer globally.
For enterprises, the implication is that compute access is no longer a vendor negotiation — it's a strategic planning exercise. Which models run on which chips, manufactured where, deployed in which cloud regions, under what export control jurisdiction? The company that builds this mapping as a product category has a clear market.
| Signal | Why it matters | Action |
|---|---|---|
| TSMC concentration | ~72% foundry share, ~90% at advanced nodes — extreme single-point risk. | Map model providers by chip sourcing and geographic risk. |
| Export control expansion | Proposed MATCH Act extends U.S. jurisdictional reach over allied chip industries. | Monitor policy pipeline and score enterprise exposure quarterly. |
| China's domestic push | 41% domestic chip share in 2025, but trailing at leading-edge nodes. | Evaluate implications for dual-market enterprises with China exposure. |
What would we build first?
A compute dependency audit tool: input your cloud providers, model APIs, and chip-intensive workloads; output a risk score by geographic concentration, export control exposure, and supply chain depth. Start with the top 20 AI model providers and 5 major cloud regions.
Who needs this?
Any enterprise running AI workloads that can't answer: "If TSMC's Taiwan fabs went offline for 6 months, which of our AI systems would be affected, and what's our fallback?" That's most Fortune 500 companies.
How would we measure success?
Time from geopolitical event (export control change, supply disruption) to enterprise impact assessment should drop from weeks to hours. Unplanned compute migration costs should decrease by 40%+ for organizations using the mapping tool.
ZOAK_BUILD_THESIS = {
category: "Compute geopolitics",
first_principle: "compute access is a strategic variable, not a procurement detail",
target_lift: "+48% geopolitical risk-response speed",
next_move: "prototype compute dependency audit for top 20 AI model providers"
}Sources: CSIS — Semiconductor Export Controls, R Street Institute — MATCH Act Analysis, The New Global Order — TSMC Analysis
Related engagement
Need to understand your compute risk exposure?
Walk us through your AI infrastructure stack — we'll map the geopolitical dependencies.
Start a conversation