Who gets to own abundant intelligence?

Introduction

If advanced AI eventually becomes a form of everyday cognitive infrastructure, the question is not only how powerful the systems become. It is who owns the machinery that makes them possible.

Modern AI often appears weightless: a chatbot in a browser, an assistant on a phone, a research tool in a workplace. Yet behind that experience sits an unusually physical stack of infrastructure: specialised chips, vast data centres, electricity networks, cloud platforms, networking equipment and billions of pounds of capital. Access to useful AI may feel like access to software, but the supply of advanced AI depends on scarce industrial systems controlled by a relatively small number of companies and states.

This creates one of the central tensions in the broader AI bloom vision. AI could help make expertise, education, scientific assistance and problem-solving dramatically more available. At the same time, the infrastructure needed to provide that abundance may become one of the most concentrated forms of economic power in the world. Whether abundant intelligence becomes broadly available or tightly controlled may depend as much on compute governance as on model capability.[arXiv]arxiv.orgarXiv Computing Power and the Governance of Artificial IntelligenceComputing Power and the Governance of Artificial IntelligenceFebruary 13, 2024…Published: February 13, 2024

Why compute is the new access bottleneck

Many earlier digital technologies became cheaper because software could be copied at near-zero cost. Advanced AI is different. While models can be distributed digitally, creating and operating frontier systems requires enormous computing resources.

Training state-of-the-art models involves vast quantities of graphics processing units (GPUs), specialised processors originally designed for parallel computing. Running those models for millions of users also requires large-scale inference infrastructure: the servers and data centres that generate answers in real time. Google describes GPUs as essential because AI training depends on enormous numbers of simultaneous mathematical operations.[Google Cloud]cloud.google.comGoogle CloudWhat is a GPU & Its Importance for AIGPUs are used to train AI models by performing the complex mathematical operations that…

The result is that access to advanced AI increasingly depends on access to compute.

Researchers studying AI governance argue that compute has several unusual characteristics. Compared with data or algorithms, it is easier to measure, harder to hide, and produced through highly concentrated supply chains. That makes compute a strategic chokepoint. Whoever controls advanced compute can influence who builds powerful models, who gains access to them and under what conditions.[arXiv]arxiv.orgarXiv Computing Power and the Governance of Artificial IntelligenceComputing Power and the Governance of Artificial IntelligenceFebruary 13, 2024…Published: February 13, 2024

This is one reason governments have started treating AI infrastructure less like ordinary software and more like a strategic national asset, comparable to telecommunications networks, energy systems or semiconductor manufacturing capacity.[Digital Strategy]digital-strategy.ec.europa.euDigital StrategyEuropean approach to artificial intelligenceThe first European AI Strategy aimed at making the EU a world-class hub for A…

The stack is concentrated at almost every layer

The concentration problem is not limited to one company. It exists across multiple layers of the AI infrastructure stack.

Chips

The most obvious bottleneck is advanced AI hardware.

The Organisation for Economic Co-operation and Development (OECD) notes that the AI GPU market is highly concentrated, with Nvidia holding more than 80% of the market for AI-focused accelerators according to recent estimates.[OECD]oecd.orgOverview of the AI supply chain: Competition in artificial…14 Nov 2025 — As noted in Box 1, GPUs are currently the most used chips…

This matters because access to leading chips often determines who can train or serve advanced models economically. When demand surges, organisations without privileged relationships or deep capital reserves may struggle to obtain hardware at all.

The chip supply chain is even narrower than it first appears. Designing advanced processors, manufacturing them, packaging them and supplying the memory systems around them each depend on a small number of specialised firms. The practical consequence is that the world’s most advanced AI systems rely on infrastructure that cannot easily be reproduced by newcomers.[OECD]oecd.aipublic ai policies for democratic and sustainable ai infrastructuresPublic AI: Policies for democratic and sustainable AI…5 Dec 2025 — Public AI policies can support sustainable AI infrastructures throu…

Cloud providers

Owning chips is not enough. They must be housed, powered, cooled and connected.

This has strengthened the position of hyperscale cloud providers such as Amazon Web Services, Microsoft Azure and Google Cloud. These firms already controlled much of the world’s cloud infrastructure before the generative AI boom. AI demand has increased the importance of those platforms because they can deploy massive GPU clusters that few other organisations can afford.[AI Now Institute]ainowinstitute.orgcompute and aiAI Now InstituteComputational Power and AISep 27, 2023 — This concentration in compute also incentivizes cloud infrastructure providers t…

For many developers, startups, universities and governments, access to frontier AI effectively means renting capacity from a small group of cloud operators.

Data centres and power

The next bottleneck is increasingly physical rather than digital.

Building frontier AI infrastructure requires land, electricity, cooling systems, fibre connections and large capital expenditures. Investors and infrastructure firms increasingly treat data-centre capacity itself as a strategic asset. Reuters reported major new acquisitions and investments aimed specifically at expanding AI-serving infrastructure.[Reuters]reuters.comI Squared bets on AI inference with $225 million data center buy from CogentThis strategic move includes an additional $1 billion commitment for upgrades, expansion, and further acquisitions. The data centers, loc…

Recent research suggests that AI data centres are becoming geographically concentrated as well. One study projects that North America, Western Europe and parts of the Asia-Pacific region could account for more than 90% of future compute capacity, creating regional dependencies and pressure on local power systems.[arXiv]arxiv.orgarXiv Computing Power and the Governance of Artificial IntelligenceComputing Power and the Governance of Artificial IntelligenceFebruary 13, 2024…Published: February 13, 2024

This means the future distribution of intelligence may partly depend on the distribution of electricity, grid capacity and industrial infrastructure.

How concentrated infrastructure shapes prices and permissions

The most visible effect of concentration is price.

A model may appear publicly available while remaining expensive enough to exclude large parts of the world. Universities, researchers, schools, charities and smaller firms often face compute costs that large technology companies can absorb far more easily.

But concentration affects more than pricing.

The operators of cloud infrastructure can determine which models are hosted, what usage policies apply, which regions receive access first and what activities are permitted. AI services therefore do not merely involve technical capability. They involve governance decisions embedded inside infrastructure.

This is one reason AI researchers increasingly discuss “compute governance” rather than only model governance. The practical power often sits below the model layer, in the systems that allocate processing resources.[arXiv]arxiv.orgarXiv Computing Power and the Governance of Artificial IntelligenceComputing Power and the Governance of Artificial IntelligenceFebruary 13, 2024…Published: February 13, 2024

A useful comparison is electricity. Households generally do not own power stations. Yet societies developed rules, utilities, regulators and public obligations to ensure electricity became broadly available rather than remaining a luxury service for a small elite.

AI may face a similar transition. If advanced cognitive assistance becomes economically important, societies may increasingly ask whether access should depend solely on commercial pricing decisions.

Why frontier AI may naturally favour giant organisations

The economics of modern AI create powerful advantages for large actors.

Training frontier models can require investments measured in hundreds of millions or even billions of pounds. The companies competing at the frontier increasingly raise enormous sums specifically to finance infrastructure. Recent funding rounds have highlighted how closely model development and compute acquisition are now linked.[Reuters]reuters.comThis development intensifies the competition between the two AI giants for market dominance. Anthropic’s valuation has more than doubled…

The result is a feedback loop:

This dynamic can create winner-takes-most tendencies even if no single company achieves a formal monopoly.

Infrastructure ownership also encourages vertical integration. Firms increasingly seek control over chips, cloud services, model development and deployment simultaneously. Microsoft’s Maia programme, Google’s Tensor Processing Units (TPUs), Amazon’s Trainium chips and similar efforts reflect attempts to reduce dependence on external suppliers while strengthening control over the entire AI stack.[IT Pro]itpro.comIT Pro What is Microsoft Maia?Unveiled in November 2023, the Maia 100 is engineered to handle large-scale AI workloads, including training and inference for generative…Published: November 2023

From a bloom perspective, the concern is not merely corporate success. It is whether access to advanced cognitive tools becomes structurally dependent on a handful of gatekeepers.

The risk is not only monopoly but dependency

Discussion of AI concentration sometimes focuses on antitrust concerns. Those matter, but dependency may be equally important.

A country, university system, healthcare network or scientific community can become dependent on infrastructure it does not meaningfully control.

That dependency can create several vulnerabilities:

These concerns become more significant if AI systems eventually play major roles in education, scientific research, medicine, administration or economic coordination.

A future in which most people rely on AI assistance while only a small number of organisations control the underlying infrastructure would resemble a highly centralised cognitive utility. That arrangement might still generate large benefits, but it raises questions about resilience, accountability and democratic oversight.

Open models do not automatically solve the problem

Open-source and open-weight AI projects are often presented as an answer to concentration.

They can help. Open models make it easier for researchers, startups and public institutions to inspect, modify and adapt systems. They reduce dependence on a single provider and can spread technical knowledge more broadly.

Yet open models face a major limitation: they still require compute.

Researchers have increasingly argued that open-source AI alone cannot guarantee broad access if the infrastructure needed to run advanced systems remains concentrated. A model can be publicly available while remaining practically inaccessible because operating it requires expensive hardware and cloud resources.[arXiv]arxiv.orgarXiv Computing Power and the Governance of Artificial IntelligenceComputing Power and the Governance of Artificial IntelligenceFebruary 13, 2024…Published: February 13, 2024

This creates a distinction between software openness and infrastructure openness.

A society may possess open models on paper while still depending on a narrow set of companies for the actual delivery of intelligence.

What public-interest compute could change

If compute concentration becomes a major constraint, one response is public-interest compute.

The idea is not necessarily that governments build complete alternatives to every commercial AI platform. Rather, it involves ensuring that researchers, universities, public institutions and smaller innovators have meaningful access to advanced computing resources.

Several emerging approaches include:

The United States’ National Artificial Intelligence Research Resource (NAIRR) is one example of an effort to create shared access to AI computing, models and data for researchers and educators rather than restricting cutting-edge resources to large private laboratories.[NSF - U.S. National Science Foundation]nsf.govNational Science FoundationNational Artificial Intelligence Research ResourceThe National Artificial Intelligence Research Resource (NAIR…

European policymakers have similarly framed AI infrastructure partly through the lens of technological sovereignty and public capacity rather than pure market competition.[Digital Strategy]digital-strategy.ec.europa.euDigital StrategyEuropean approach to artificial intelligenceThe first European AI Strategy aimed at making the EU a world-class hub for A…

Advocates of public AI argue that public compute can help narrow what they call the “compute divide”: the growing gap between organisations that can afford frontier infrastructure and those that cannot.[OECD.AI]oecd.aipublic ai policies for democratic and sustainable ai infrastructuresPublic AI: Policies for democratic and sustainable AI…5 Dec 2025 — Public AI policies can support sustainable AI infrastructures throu…

The strongest argument for concentration

There is also a serious counterargument.

Building and operating frontier AI infrastructure is genuinely difficult. Concentration may not simply be the result of market failure. It may reflect economies of scale.

Large firms can:

Some degree of concentration may therefore be unavoidable if the goal is pushing capability frontiers quickly.

Supporters of this view argue that the fastest path to scientific acceleration, advanced medical discovery and powerful AI tools may require precisely the kind of large-scale infrastructure investments now being made by major technology firms. The challenge is then not preventing concentration entirely but ensuring that the resulting capabilities diffuse outward.[KKR]kkr.comBeyond the Bubble: Why AI Infrastructure Will Compound…Will long-term AI infrastructure demand justify current activity? Explore th…

The key question becomes whether concentrated production can coexist with broad access.

The deeper bloom question

The long-term significance of compute governance is not mainly about today’s AI subscriptions. It is about the future distribution of intelligence itself.

If advanced AI becomes a general-purpose amplifier of learning, creativity, research and problem-solving, then access to AI may increasingly resemble access to education, electricity or the internet: a foundational condition for participation in modern society.

The optimistic AI bloom vision depends on more than creating powerful systems. It depends on making cognitive abundance genuinely available. That means the institutions controlling chips, cloud platforms, data centres and compute allocation may end up shaping the social consequences of AI as much as the model developers themselves.

A future of broad human flourishing is easier to imagine when intelligence becomes widely accessible infrastructure. It is harder to imagine if access to that infrastructure remains permanently scarce, expensive or dependent on a small number of gatekeepers. The struggle over compute is therefore not a side issue to abundant intelligence. It is one of the places where the future distribution of abundance may be decided.

Amazon book picks

Further Reading

Books and field guides related to Who gets to own abundant intelligence?. Use these as the next step if you want deeper reading beyond the article.

Book

Chip War

By Chris Miller

Explains the chip and compute bottlenecks behind advanced AI access.

See on Amazon

Book

The Coming Wave

By Mustafa Suleyman

Discusses concentration and containment of powerful AI infrastructure.

See on Amazon

Book

The Master Switch

By Tim Wu

Useful analogy for cloud and compute gatekeeping.

See on Amazon

Book

Platform Revolution

By Geoffrey G. Parker, Marshall W. Van Alstyne et al.

Explains ecosystem control and access rules in platform markets.

See on Amazon

Browse more on Amazon:Chip WarThe Coming WaveThe Master Switch

As an Amazon Associate I earn from qualifying purchases.

eBay marketplace picks

Marketplace Samples

Example marketplace items related to this page. Use the search link to explore similar finds on eBay.

Shop location

UsingUSA USA USAUKCanadaAustraliaIreland

Example eBay listing

Mos Technologies 8568 Commodore C128 Vdc Chip Video Display Controller Dip-48

SearcheBay.co.uk: computer chip display

Browse similar oneBay.co.uk

Example eBay listing

MOS Technologies 8568 Commodore C128 VDC Chip Video Display Controller DIP-48

SearcheBay.co.uk: computer chip display

Browse similar oneBay.co.uk

Example eBay listing

MOS Technologies 8566 R3X VIC IIe Chip Video Display Controller DIP-48 C128 IC

SearcheBay.co.uk: computer chip display

Browse similar oneBay.co.uk

Example eBay listing

LP156WF6(SP)(B1) 30 Pin LVDS LG Display 15.6" Slim Screen Small Chip 0C3MWM

SearcheBay.co.uk: computer chip display

Browse similar oneBay.co.uk

Browse more oneBay.co.uk

Example items shown for inspiration; availability and pricing can change. Branchoria may earn a commission if you purchase through outbound eBay links.

Endnotes

1. Source: arxiv.org
Title: arXiv Computing Power and the Governance of Artificial Intelligence
Link:https://arxiv.org/abs/2402.08797

Published: February 13, 2024

2. Source: cloud.google.com
Link:https://cloud.google.com/discover/gpu-for-ai

3. Source: oecd.org
Link:https://www.oecd.org/en/publications/competition-in-artificial-intelligence-infrastructure_623d1874-en/full-report/component-5.html

4. Source: reuters.com
Title: I Squared bets on AI inference with $225 million data center buy from Cogent
Link:https://www.reuters.com/business/media-telecom/i-squared-bets-ai-inference-with-225-million-data-center-buy-cogent-2026-05-26/

5. Source: arxiv.org
Link:https://arxiv.org/abs/2604.06198

Published: March 13, 2026

6. Source: reuters.com
Link:https://www.reuters.com/business/anthropic-raises-65-billion-now-valued-965-billion-2026-05-28/

7. Source: arxiv.org
Title: arXiv If open source is to win, it must go public
Link:https://arxiv.org/abs/2507.09296

Published: July 12, 2025

8. Source: oecd.ai
Title: public ai policies for democratic and sustainable ai infrastructures
Link:https://oecd.ai/en/wonk/public-ai-policies-for-democratic-and-sustainable-ai-infrastructures

9. Source: nsf.gov
Link:https://www.nsf.gov/focus-areas/ai/nairr

10. Source: kkr.com
Link:https://www.kkr.com/insights/ai-infrastructure

11. Source: reuters.com
Title: marvell technology forecasts quarterly revenue above estimates 2026 05 27
Link:https://www.reuters.com/technology/marvell-technology-forecasts-quarterly-revenue-above-estimates-2026-05-27/

12. Source: oecd.org
Title: component 6
Link:https://www.oecd.org/en/publications/competition-in-artificial-intelligence-infrastructure_623d1874-en/full-report/component-6.html

13. Source: oecd.org
Link:https://www.oecd.org/en/topics/sub-issues/ai-principles.html

14. Source: arxiv.org
Link:https://arxiv.org/pdf/2412.05761

15. Source: arxiv.org
Link:https://arxiv.org/html/2504.16026v1

16. Source: ainowinstitute.org
Title: compute and ai
Link:https://ainowinstitute.org/publications/compute-and-ai

17. Source: digital-strategy.ec.europa.eu
Link:https://digital-strategy.ec.europa.eu/en/policies/european-approach-artificial-intelligence

18. Source: itpro.com
Title: IT Pro What is Microsoft Maia?
Link:https://www.itpro.com/infrastructure/what-is-microsoft-maia

Published: November 2023

19. Source: openfuture.eu
Link:https://openfuture.eu/our-work/ai-and-the-commons/

20. Source: digital-strategy.ec.europa.eu
Link:https://digital-strategy.ec.europa.eu/en/policies/expert-group-ai

Additional References

21. Source: marketsandmarkets.com
Link:https://www.marketsandmarkets.com/Market-Reports/ai-infrastructure-market-38254348.html

22. Source: medium.com
Link:https://medium.com/%40gpuanalist/gpu-cloud-computing-an-honest-market-analysis-for-ai-applications-in-2026-6e150a69a7c0

23. Source: onu.delegfrance.org
Link:https://onu.delegfrance.org/statement-on-inclusive-and-sustainable-artificial-intelligence-for-people-and

24. Source: opensource.org
Link:https://opensource.org/ai/webinars/a-data-pathway-to-building-public-ai

25. Source: industry.gov.au
Title: expectations data centres and ai infrastructure developers
Link:https://www.industry.gov.au/publications/expectations-data-centres-and-ai-infrastructure-developers

26. Source: visualcapitalist.com
Title: charted the battle for ai data center revenue 2021 2025
Link:https://www.visualcapitalist.com/charted-the-battle-for-ai-data-center-revenue-2021-2025/

27. Source: un.org
Link:https://www.un.org/sites/un2.un.org/files/governing_ai_for_humanity_final_report_en.pdf

28. Source: GOV.UK
Title: artificial intelligence sector study 2024
Link:https://www.gov.uk/government/publications/artificial-intelligence-sector-study-2024/artificial-intelligence-sector-study-2024

29. Source: precedenceresearch.com
Title: ai chips for data center and cloud market
Link:https://www.precedenceresearch.com/ai-chips-for-data-center-and-cloud-market

30. Source: linkedin.com
Link:https://www.linkedin.com/posts/kumar-priyadarshi-b0a2a7a2_4-major-companies-monopolising-ai-chip-supply-activity-7433354306311065600-Dea4