Topic 8.6

AI Infrastructure & Compute Economics

Understand how compute, cloud, chip supply chains, and data form the production core of the AI-driven digital economy. Analyse concentration risks, public compute initiatives, and valuation of data as intangible capital.

⏱️Approx. 25 min
🖥️Infrastructure Economics
🧭Stage 6 · Compute

AI Infrastructure Stack

Three layers underpin AI capability. Use the architecture view as a checklist for research or policy analysis.

Hardware & Energy

  • Advanced semiconductors (NVIDIA H100, AMD MI300, TPU v5)
  • Specialised interconnects (NVLink, InfiniBand)
  • Power & cooling infrastructure (liquid immersion, waste heat reuse)

Cloud & Platform Layer

  • Hyperscaler regions/availability zones
  • Managed AI services (SageMaker, Vertex, Azure ML)
  • Data engineering pipelines, vector databases, orchestration

Data & Governance

  • Training data supply chains, copyright/licensing
  • Data residency, sovereignty, trust architectures
  • Responsible AI governance, risk monitoring, evaluation

Compute Concentration & Economics

Market Share Snapshot (Q4 2024)

  • AWS: 32% global cloud market; $60B+ annual capex
  • Microsoft Azure: 23%; strategic OpenAI partnership
  • Google Cloud: 10%; TPU-centric AI offerings
  • Alibaba/Tencent: 9% combined; regional dominance in APAC

Cost & Access

  • Frontier model training budgets exceed $100M (hardware + energy).
  • <5% of academic labs can access GPT-3-scale compute.
  • NVIDIA H100 GPU unit price $25k–40k; lead times 6–9 months.
  • TSMC produces 90%+ of sub-7nm chips (single point of failure).

Concentration poses innovation and geopolitical risks. Research directions include market structure analysis, antitrust implications, and alternative compute governance (public compute, cooperative models).

Public Compute & Open Access

Global Initiatives

  • UK AI Research Resource: £900M to build national compute clusters with academic access quotas.
  • EU EuroHPC JU: €7B programme delivering exascale compute and AI Factories across member states.
  • US NSF ACCESS: Expanding cloud credits and NSF-funded GPU clusters for researchers.
  • Singapore NCCS: National Supercomputing Centre focusing on climate + AI workloads with heat reuse pilots.

Key challenges: scale (public capacity often 1–10% of commercial hyperscalers), usability, sustained funding, interoperability.

Data as Capital & Intangible Assets

The 2025 update to the System of National Accounts proposes capitalising data, databases, and AI models. Pilot studies estimate data capital contributes 0.5–2% of GDP in advanced economies, though valuation methods remain experimental.

Valuation Approaches

  • Cost-based (collection, cleaning, storage, compute)
  • Income-based (marginal revenue/productivity gains)
  • Market-based (data trades, licensing, synthetic data)

Research Directions

  • Designing data trusts/cooperatives for shared value capture
  • Integrating compute + data into productivity accounting
  • Measuring data localisation effects on innovation and inclusion

🎯 Key Takeaways

  • AI infrastructure spans chips, cloud/data centres, and data governance—each with distinct policy levers and concentration risks.
  • Compute access is highly unequal; public initiatives seek to democratise resources but face scale and usability gaps.
  • Data capitalisation is reshaping national accounts and corporate valuation, though consensus on valuation methods is nascent.
  • Geopolitical tensions (export controls, onshoring) and supply chain chokepoints (TSMC, ASML) influence research agendas and policy priorities.
  • Integrating sustainability (energy, water) with compute economics is critical for responsible AI scaling.