A few US AI giants, notably NVIDIA and Microsoft, hold decisive structural leverage over Norway’s Stargate data‑centre ambitions—creating technological and financial dependencies that undercut Oslo’s claims of AI sovereignty. Structural power is not universal across MNAICs but concentrated, meaning host-state outcomes are shaped more by dominant firms than by national policy rhetoric.

Main Finding

Structural power among Multi-National AI Corporations (MNAICs) is dispersed and multidimensional rather than monolithically concentrated. A quantitative neo‑Strangean typology of 79 MNAICs and a qualitative technographic analysis of Stargate Norway show that Microsoft and NVIDIA occupy analytically central positions. Their combined strengths—particularly in production and knowledge structures—enable mechanisms (vertical stack integration, sovereign‑cloud abstraction, and financial lock‑in) that substantially shape state‑market relations in non‑US data‑center projects and limit host‑state ambitions for data/AI sovereignty.

Key Points

• Theoretical framing: A neo‑Strangean synthesis revives Susan Strange’s four dimensions of structural power (security, production, finance, knowledge) and pairs them with Culpepper’s conception of structural power to link underlying positional advantages to observable corporate mechanisms.
• Mixed‑methods design: Sequential approach—(1) large‑N quantitative typology to map structural positions across 79 MNAICs; (2) qualitative technographic document analysis of Stargate Norway (press releases, government/public reports, corporate blogs) to trace mechanisms in situ.
• Typology result: No single firm dominates all four Strangean dimensions. Instead, a small set of firms concentrate complementary strengths. NVIDIA (production/knowledge) and Microsoft (production/finance/knowledge) emerge as pivotal.
• Mechanisms identified:
  • Vertical stack integration: firms lock in customers and regions by controlling hardware, software, and cloud services.
  • Sovereign‑cloud abstraction: cloud/platform offerings abstract control away from host states, limiting regulatory or operational leverage.
  • Financial lock‑in: large-scale investments and public–private de‑risking create durable dependencies by host states and local partners.
• Case findings (Stargate Norway): Despite Norway’s discursive emphasis on data/AI sovereignty and state/local involvement (e.g., hydropower access), project outcomes and operational realities remain primarily shaped by US MNAICs operating largely outside Norwegian jurisdiction.
• Empirical novelty: The thesis produces the first broad comparative, quantitatively grounded neo‑Strangean baseline for AI sector structural power and demonstrates a heterogeneous “corporate patchwork” rather than unitary Big Tech dominance.
• Epistemic stance & limits: Uses critical realism—structures are treated as real causal conditions though strict causal proof is acknowledged as methodologically difficult.

Data & Methods

• Quantitative phase:
  • Sample: 79 Multi‑National AI Corporations.
  • Data: publicly available indicators mapped to the four Strangean dimensions (production, knowledge, finance, security). (The thesis operationalizes these dimensions into index components and performs typological clustering; formulas and statistical steps were provided ahead of analysis.)
  • Outcome: a six‑class empirical typology of MNAICs and descriptive distributional statistics (including unexpected peripheral position of OpenAI).
• Qualitative phase:
  • Case: Stargate Norway (520 MW, ~100k GPUs; partners include OpenAI, NVIDIA, Microsoft; municipal/state cooperation in Narvik).
  • Data sources: press releases, corporate blogs, public/governmental reports, regulatory documents.
  • Method: technographic document analysis tracing how typologically indicated structural positions translate into concrete mechanisms across legal, energy, technological, and partnership domains.
• Reproducibility & AI use: Quantitative merging and index calculations were assisted by Claude Code under human‑provided formulas; all AI suggestions were checked, and AI‑detection/proofing tools were used for text work.
• Methodological limitations: Structural claims aim to approximate causal influence (critical realist approach) but do not purport to offer tight causal identification.

Implications for AI Economics

• Market structure and competition:
  • Vertical stack integration and complementary positionings (hardware + cloud + knowledge) can raise barriers to entry, entrench dominant suppliers, and produce durable market concentration even without full‑spectrum dominance by a single firm.
• Sovereignty and policy design:
  • Host states’ investments (e.g., in energy or land) do not automatically translate into operational sovereignty if upstream technology and platform layers remain controlled by foreign MNAICs.
  • Policy responses should consider conditional investment terms, supplier diversification, requirements for data/localization controls, and negotiated service‑level or governance clauses that address sovereign‑cloud abstraction risks.
• Finance and public‑private partnerships:
  • Large-scale de‑risking and financing arrangements are channels for long‑term lock‑in; economic planners must weigh short‑term gains (jobs, capex) against persistent dependency and limited regulatory reach.
• Energy and local economies:
  • Hyperscale AI projects reshape local energy demand (e.g., hydropower allocation); economic implications include altered electricity markets, rent capture questions, and potential distributional effects for local communities.
• Research and policy measurement:
  • The neo‑Strangean typology provides a replicable quantitative tool for economists and policymakers to map corporate structural power and forecast likely leverage points in cross‑border infrastructure projects.
• Future research directions:
  • Extend the typology with dynamic time‑series indicators (investment flows, changing patent/paper shares) to monitor evolving structural power.
  • Evaluate policy experiments (conditional investments, procurement rules) empirically to test mitigation of financial lock‑in and sovereign‑cloud abstraction.

If you want, I can (a) extract the exact index components and clustering algorithm used in the typology (from the thesis appendices) into a short technical note, or (b) draft a one‑page policy brief for Norwegian (or similar) host‑state policymakers summarizing actionable measures. Which would be more useful?