Skip to content
OGERIA — Observatory of Global Evidence on Risks in AISynthesis report · 2026 ed.
Updated 2 Oct 2026

Chapter 01 · Risk map

The ability to train and serve frontier models runs through a handful of factories and contracts, and whoever controls that funnel decides who competes.

Severity
Catastrophic
Horizon
1–3 years
Evidence
Projected
Consensus
Medium

Almost all AI regulation points at the models. Power, however, sits upstream, in a physical funnel that can be counted. Epoch AI measured it: TSMC is the sole producer of advanced CoWoS packaging for leading-edge chips, and four designers —Nvidia, AMD, Google, Amazon— were consuming 80-85% of total supply at the end of 2025; HBM memory has three suppliers in the world, all three sold out through 2026, and Nvidia’s share of its global consumption went from ~45% to more than 75% between 2024 and 2025 [375]AI Chip Supply Chain Bottlenecks and Capacity (AI Chip Components Explorer)Somala, Venkat · 2026 · reportView source ↗Accessed on 9 September 2026.

Capital follows the same shape: private AI investment grew 127.5% in 2025 and is already 60% of the total, and the United States committed 23 times more than China [34]The AI Index 2026 Annual Report — Chapter 4: EconomyStanford Institute for Human-Centered AI · 2026 · reportView source ↗Accessed on 9 September 2026. And so does adoption: more than 50% of the working-age population uses AI in the Emirates and Singapore against less than 10% in many low-income economies [538]International AI Safety Report 2026Bengio, Yoshua · 2026 · reportView source ↗Accessed on 9 September 2026 [71]Anthropic Economic Index report: Uneven geographic and enterprise AI adoptionAppel, Ruth; McCrory, Peter; Tamkin, Alex et al. · 2025 · reportView source ↗Accessed on 9 September 2026.

What this does not demonstrate. That the input is concentrated does not prove that the rents concentrate: that step is projection. The most cited forecast in that direction comes from a skeptic of large macro effects —Acemoglu estimates no more than a 0.66% increase in total factor productivity over ten years and still predicts that the capital-labour gap will widen— and it is a model, not a measurement [15]The Simple Macroeconomics of AIAcemoglu, Daron · 2025 · paperView source ↗Accessed on 9 September 2026. On the other side, labour’s share of US GDP has been stable at around 60% for more than a century, and that is the regularity one would have to see break. The funnel, moreover, cuts both ways: in June 2026 it was the state that used it, suspending worldwide access to two frontier modelsFrontier AIThe most advanced AI systems in existence at a given time, the ones pushing the limit of what the technology can do. A handful of companies with enormous resources build them.For exampleLike Formula 1 cars: there are few of them, they are extremely expensive and only a few teams build them, but what gets tested there ends up in everyone's car. for eighteen days [85]Statement on the US government directive to suspend access to Fable 5 and Mythos 5Anthropic · 2026 · institutional blogView source ↗Accessed on 9 September 2026.

Chain of materialisation

  1. PreconditionObserved

    The physical bottleneck is narrow and measured

    TSMC is the exclusive producer of advanced CoWoS packaging for leading AI chips, and four designers consumed roughly 80-85% of total supply at the end of 2025. HBM memory has three suppliers worldwide, all effectively sold out for 2026, and Nvidia's share of global HBM consumption went from ~45% to over 75% between 2024 and 2025.

  2. TriggerObserved

    Capital concentrates at the same pace as the input

    Private AI investment grew 127.5% in 2025 and already accounts for 60% of the total; the United States committed 23 times more than China. Component spending by the four main designers went from 22 billion dollars in 2024 to 52 billion in 2025.

    Observed and demonstrated evidence ends here. What follows is projection.

  3. CascadeProjected

    Rent shifts from labour to capital

    Acemoglu projects modest macro effects -no more than a 0.66% increase in total factor productivity over ten years- and still concludes that AI is predicted to widen the gap between capital and labour income. It is a model, not a measurement: the same author who says the pie will grow little says it will be split worse.

  4. ImpactSpeculative

    A lever over who can compete that no model regulation touches

    The June 2026 episode shows the lever working in a state direction: an export control directive suspended access to two frontier models for every customer worldwide. What distribution of power that funnel produces over a decade is modelled by nobody.

    Precedents: An export control directive suspends access to Fable 5 and Mythos 5

See on the map →Report a mistake in this entry →

Sources

  1. [375] AI Chip Supply Chain Bottlenecks and Capacity (AI Chip Components Explorer) · Epoch AI 2026
  2. [34] The AI Index 2026 Annual Report — Chapter 4: Economy · Stanford HAI 2026
  3. [15] The Simple Macroeconomics of AI · MIT / NBER 2025
  4. [538] International AI Safety Report 2026 · International AI Safety Report (panel con representantes nominados por más de 30 países) 2026
  5. [550] Gen-AI: Artificial Intelligence and the Future of Work · Fondo Monetario Internacional 2024 archived copy only
  6. [71] Anthropic Economic Index report: Uneven geographic and enterprise AI adoption · Anthropic 2025
  7. [336] The Intelligence Curse · Drago, Luke 2025
  8. [85] Statement on the US government directive to suspend access to Fable 5 and Mythos 5 · Anthropic 2026
  9. [367] What did US export controls mean for China's AI capabilities? · Epoch AI 2024
  10. [145] The AI supply chain (BIS Papers No 154) · BIS 2025

Ask OGERIA

It answers only with what the observatory publishes and can be wrong: check the entries it cites. Your questions are sent to an AI model, so don't write personal data. More in the privacy policy.

Up to 500 characters.

Support OGERIA on Ko-fi

The payment is processed by Ko-fi, not by this site. Open on ko-fi.com