Can compute scarcity limit the AI boom

Introduction

Frontier AI systems depend on a resource that is becoming as strategically important as data and algorithms: large-scale computing power. The most advanced models require enormous clusters of specialised chips, high-capacity data centres, reliable electricity and billions in infrastructure investment. This creates a potential constraint on the AI boom: if only a small number of companies, countries or cloud providers can obtain enough compute, the ability to build frontier systems may become concentrated among a few powerful actors.

Compute scarcity does not mean AI progress will stop. Hardware efficiency improvements, better algorithms and smaller specialised models can reduce the amount of computing needed for many tasks. But the economics of the frontier are increasingly shaped by access to scarce infrastructure. Training costs for the largest AI models have risen rapidly, with estimates suggesting the cost of the most compute-intensive training runs has increased roughly 2.4 times per year since 2016.[arXiv]arxiv.orgarXiv The rising costs of training frontier AI modelsThe rising costs of training frontier AI modelsMay 31, 2024…Published: May 31, 2024 The result is a tension at the heart of AI’s potential contribution to human flourishing: advanced AI could help accelerate science, medicine and abundance, but the infrastructure needed to create it may become concentrated in ways that limit who can participate.

Why frontier compute is difficult to access

Frontier AI development requires far more than simply buying faster computers. The challenge is assembling thousands or even hundreds of thousands of advanced processors into a tightly connected system that can train and operate large models efficiently.

Modern AI relies heavily on specialised accelerator chips, especially graphics processing units (GPUs), because they can perform the massive parallel calculations needed for neural networks. These chips are expensive, in high demand and difficult to deploy at scale. The supply chain itself is concentrated: the OECD notes that Nvidia has become the dominant supplier of AI-focused GPUs, with market estimates placing its share above 80% in AI GPU chips, while competition is limited by hardware development costs, software ecosystems and manufacturing complexity.[oecd.org]oecd.orgcomponent 5Overview of the AI supply chain: Competition in artificial intelligence infrastructure | OECDNovember 14, 2025…Published: November 14, 2025

The scarcity is not only about chips. A frontier AI cluster requires:

This creates a high entry barrier. A small research group or new company may have a good idea for a model but still be unable to compete because it cannot secure enough compute at the right time and price.

Research into AI resource access has found growing concerns about a divide between organisations with abundant computing resources and those without them. A survey of AI researchers published in the AAAI Conference on Artificial Intelligence examined compute access as a potential constraint on research progress and highlighted concerns about resource stratification between industry and academic researchers.[AAAI Open Access]ojs.aaai.orgOpen AccessResource Democratization: Is Compute the Binding Constraint on AI Research? | Proceedings of the AAAI Conference on Artificial Intelligen…

How infrastructure concentration affects competition

The risk is not simply that large companies will have better computers. The deeper concern is that control over compute could shape the entire structure of the AI ecosystem.

Frontier development may become a capital-intensive race

The largest AI models increasingly require budgets beyond the reach of most startups and universities. Stanford’s AI Index has documented rapid growth in training compute requirements, with training compute for notable AI models doubling approximately every five months in recent years.[Stanford HAI]hai.stanford.eduresearch and developmentStanford HAIResearch and Development | The 2025 AI Index Report | Stanford HAI…

This does not mean only the richest organisations can create valuable AI. Many important advances come from algorithmic improvements, open research and smaller models. Some research suggests that efficiency gains can reduce the amount of compute needed for breakthroughs.[arXiv]arxiv.orgarXiv Compute Requirements for Algorithmic Innovation in Frontier AI ModelsCompute Requirements for Algorithmic Innovation in Frontier AI ModelsJuly 13, 2025…Published: July 13, 2025

However, the frontier itself may become increasingly shaped by organisations capable of funding enormous infrastructure projects. This could narrow competition among companies attempting to create the most capable general-purpose systems.

A future in which only a handful of organisations can train frontier models could affect the direction of AI development. Companies with control over scarce compute may have greater influence over:

• which scientific problems receive AI investment;
• which business models dominate;
• how widely advanced capabilities are distributed;
• what safety approaches are prioritised.

For an AI bloom scenario — where advanced AI helps humanity overcome disease, scarcity and scientific bottlenecks — broad access matters because breakthroughs may depend on contributions from many researchers, institutions and regions.

Cloud providers can become essential gateways

Because buying and operating giant AI clusters is difficult, many AI developers rely on cloud providers. This creates a new dependency relationship: the companies building AI systems may depend on the companies controlling the infrastructure beneath them.

Cloud partnerships can accelerate progress by giving smaller organisations access to powerful computing without requiring them to build their own data centres. But dependence can also create strategic vulnerability. If a small number of cloud companies control much of the available capacity, they may gain influence over pricing, access terms and which AI projects can scale.

The OECD has identified competition concerns around AI infrastructure, including the possibility that concentrated markets in chips, cloud services and data-centre capacity could create barriers for new entrants.[oecd.org]oecd.orgfull reportCompetition in artificial intelligence infrastructure | OECDNovember 14, 2025…Published: November 14, 2025

This does not mean cloud providers will necessarily abuse their position. Large infrastructure companies also have incentives to expand access because more AI developers create more demand for their services. The challenge is maintaining a competitive environment where new AI builders have realistic alternatives.

The global competition problem

Compute concentration is also becoming a geopolitical issue. Advanced AI depends on global supply chains involving chip design, semiconductor manufacturing, cloud infrastructure and energy resources.

The most advanced AI chips are difficult to produce and rely on specialised manufacturing capabilities. This creates strategic dependence on a small number of companies and locations. Stanford’s AI Index has highlighted the concentration of AI hardware supply chains, including the importance of leading semiconductor manufacturing capacity and the growing scale of global AI infrastructure investment.[Stanford HAI]hai.stanford.eduresearch and developmentStanford HAIResearch and Development | The 2026 AI Index Report | Stanford HAI…

Governments increasingly view access to AI compute as part of national competitiveness. The European Union, for example, has launched initiatives to expand AI computing infrastructure through AI factories and proposed larger AI gigafactories designed to provide more access to advanced computing for startups, researchers and industry.[European Commission]commission.europa.euEuropean Commission AI continentEuropean CommissionAI continent - European CommissionApril 9, 2025…Published: April 9, 2025

Export controls add another layer of complexity. Restrictions on advanced chips are intended to protect strategic advantages and manage security risks, but they can also reshape who has access to the tools needed for AI research. Recent reporting on Chinese AI development has highlighted how difficult it can be to separate AI progress from the global availability of advanced chips and computing resources.[tomshardware.com]tomshardware.comDue to the popularity and resource demands of Kimi K3, compute resources have been strained, placing new subscriptions on a waiting list…

Possible routes toward broader AI access

Compute scarcity does not have only one possible outcome. Several developments could reduce concentration and widen participation.

More efficient AI could reduce dependence on massive clusters

One route is technological improvement. Better algorithms, more efficient training methods and specialised models can deliver useful capabilities with less computing power.

Recent years have shown that progress is not driven only by making models larger. Techniques such as improved architectures, better data strategies and efficient fine-tuning methods have allowed smaller systems to achieve capabilities that previously required much larger models.

This matters because a future where useful AI requires less compute could make the benefits of AI more widely available. Scientific researchers, universities, public institutions and smaller companies would have more opportunity to contribute.

Alternative hardware could weaken dependence on dominant suppliers

Another possibility is greater competition in AI hardware. Companies are developing alternative accelerators designed to challenge Nvidia’s position. Research comparing alternative AI processors has suggested that competing hardware can achieve competitive results in some workloads, although software ecosystems and developer adoption remain major obstacles.[arXiv]arxiv.orgarXiv Debunking the CUDA Myth Towards GPU-based AI SystemsarXiv Debunking the CUDA Myth Towards GPU-based AI Systems

The challenge is that hardware competition is not only about producing chips. Nvidia’s advantage also comes from its mature software ecosystem, developer tools and widespread adoption. Breaking this advantage requires alternatives that are not merely technically capable but easy for researchers and companies to use.

Public and shared compute could widen participation

Governments and research institutions are exploring shared AI infrastructure as a way to prevent frontier computing from becoming available only to the largest corporations.

Public compute resources could allow universities, startups and researchers to experiment with advanced models without needing billions in private investment. The European Union’s AI infrastructure plans explicitly aim to provide access for startups, researchers and industry rather than limiting frontier computing to major technology companies.[European Commission]commission.europa.euEuropean Commission Artificial IntelligenceEuropean Commission Artificial Intelligence

Shared infrastructure will not eliminate competition concerns, but it could create a broader base of participants able to contribute to AI progress.

Why compute access matters for the AI bloom vision

If advanced AI helps humanity enter a period of abundance, the infrastructure question becomes central. Scientific acceleration, medical breakthroughs, climate solutions and new forms of education may depend on increasingly capable AI systems. But those capabilities will have wider impact only if access to the underlying technology is not permanently restricted to a narrow group.

The risk is not simply economic inequality between companies. Concentrated compute could influence which societies, researchers and communities are able to participate in shaping the technology that may become one of civilisation’s most important tools.

At the same time, avoiding concentration does not mean eliminating large-scale infrastructure. Frontier AI may genuinely require enormous resources, just as major scientific projects such as particle accelerators and space programmes do. The key question is whether these resources become broadly accessible foundations for human progress or strategic bottlenecks controlled by a small number of organisations.

The future of AI competition may therefore depend on finding a balance: enough scale to build powerful systems, but enough openness and diversity that the benefits of advanced intelligence can contribute to a wider human flourishing.

Amazon book picks

Further Reading

Books and field guides related to Can compute scarcity limit the AI boom. Use these as the next step if you want deeper reading beyond the article.

Book

The Coming Wave

By Mustafa Suleyman

"We are approaching a critical threshold in the history of our species. Everything is about to change. Soon you will live surrounded by A...

See on Amazon

Book

AI Superpowers

By Kai-Fu Lee

THE NEW YORK TIMES, USA TODAY, AND WALL STREET JOURNAL BESTSELLER "Kai-Fu Lee believes China will be the next tech-innovation superpower...

See on Amazon

Book

Chip War

By Chris Miller

***Winner of the 2022 Financial Times Business Book of the Year Award*** ***Selected as one of Barack Obama's Favourite Books of 2023***...

See on Amazon

Book

The Nvidia Way

By Tae Kim

"The Nvidia Way is a riveting history of Nvidia's unexpected ascent to the top of the tech sector--and a compelling case for why cofounde...

See on Amazon

Browse more on Amazon:The Coming WaveAI SuperpowersChip War

As an Amazon Associate I earn from qualifying purchases.

eBay marketplace picks

Marketplace Samples

Live-tested eBay searches with available results related to this page.

Shop location

UsingUSA USA USAUKCanadaAustraliaIreland

Selected fromdata center poster oneBay.co.uk.

Current eBay listing

Data Center Garden Work Framed Wall Art Poster Canvas Print Picture

GBP 12.99 | Shipping GBP 4.99NewSeller: jjestoreltd (99.6% positive)Ships from: GB

View listing on eBay

Current eBay listing

An Advanced Futuristic Data Center Framed Wall Art Poster Canvas Print Picture

GBP 29.90 | Shipping GBP 4.95NewSeller: mahome7 (99.5% positive)Ships from: GB

View listing on eBay

Current eBay listing

Future Proof Data Center Employing Framed Wall Art Poster Canvas Print Picture

GBP 29.90 | Shipping GBP 4.95NewSeller: mahome7 (99.5% positive)Ships from: GB

View listing on eBay

Current eBay listing

Data Center Interior Bristling With Framed Wall Art Poster Canvas Print Picture

GBP 8.99 | Shipping GBP 4.99NewSeller: ruppalsingh (99.6% positive)Ships from: GB

View listing on eBay

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 The rising costs of training frontier AI models
Link:https://arxiv.org/abs/2405.21015

Published: May 31, 2024

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

Published: November 14, 2025

3. Source: ojs.aaai.org
Title: Open Access
Link:https://ojs.aaai.org/index.php/AAAI/article/view/29959

4. Source: hai.stanford.edu
Title: research and development
Link:https://hai.stanford.edu/ai-index/2025-ai-index-report/research-and-development

5. Source: arxiv.org
Title: arXiv Compute Requirements for Algorithmic Innovation in Frontier AI Models
Link:https://arxiv.org/abs/2507.10618

Published: July 13, 2025

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

Published: November 14, 2025

7. Source: hai.stanford.edu
Title: research and development
Link:https://hai.stanford.edu/ai-index/2026-ai-index-report/research-and-development

8. Source: tomshardware.com
Link:https://www.tomshardware.com/tech-industry/artificial-intelligence/chinas-moonshot-ai-reportedly-used-nvidia-blackwell-chips-for-training-kimi-k3-company-circumvented-both-u-s-export-and-chinese-import-controls-to-acquire-compute

9. Source: arxiv.org
Title: arXiv Debunking the CUDA Myth Towards GPU-based AI Systems
Link:https://arxiv.org/abs/2501.00210

10. Source: nvidianews.nvidia.com
Title: europe unveils a record 35 new nvidia ai supercomputers
Link:https://nvidianews.nvidia.com/news/europe-unveils-a-record-35-new-nvidia-ai-supercomputers

11. Source: nvidianews.nvidia.com
Title: vera rubin full production agentic ai factory
Link:https://nvidianews.nvidia.com/news/vera-rubin-full-production-agentic-ai-factory

12. Source: investor.nvidia.com
Link:https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Vera-Rubin-Ramps-Into-Full-Production-to-Power-Agentic-AI-Factories-Worldwide/default.aspx

13. 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

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

15. Source: hai.stanford.edu
Title: 2025 ai index report
Link:https://hai.stanford.edu/ai-index/2025-ai-index-report?sf223786129=1

16. Source: hai.stanford.edu
Title: research and development
Link:https://hai.stanford.edu/ai-index/2024-ai-index-report/research-and-development

17. Source: hai.stanford.edu
Title: 2024 ai index report
Link:https://hai.stanford.edu/ai-index/2024-ai-index-report?hl=en-US

18. Source: hai.stanford.edu
Title: 2025 ai index report
Link:https://hai.stanford.edu/ai-index/2025-ai-index-report?b=wsj-events&source-key=wsj.com

19. Source: hai.stanford.edu
Title: 2025 ai index report
Link:https://hai.stanford.edu/ai-index/2025-ai-index-report?authuser=0

20. Source: commission.europa.eu
Title: European Commission AI continent
Link:https://commission.europa.eu/topics/competitiveness/ai-continent_en

Published: April 9, 2025

21. Source: commission.europa.eu
Title: European Commission AI Gigafactories
Link:https://commission.europa.eu/topics/competitiveness/competitiveness-coordination-tool-projects/ai-gigafactories_en

22. Source: commission.europa.eu
Title: European Commission Artificial Intelligence
Link:https://commission.europa.eu/topics/artificial-intelligence_en

23. Source: commission.europa.eu
Title: eu Strengthening Europe’s tech sovereignty
Link:https://commission.europa.eu/news-and-media/news/strengthening-europes-tech-sovereignty-2026-06-03_en

24. Source: digital-strategy.ec.europa.eu
Title: eu A I Factories | Shaping Europe’s digital future
Link:https://digital-strategy.ec.europa.eu/en/policies/ai-factories

25. Source: eur-lex.europa.eu
Title: eu Regulation
Link:https://eur-lex.europa.eu/legal-content/EN/TXT/?qid=1780778296020&uri=CELEX%3A32026R0150

26. Source: thekb.eu
Title: stanford hai ai index report 2025 trends 2025 04 07
Link:https://www.thekb.eu/en/fiches/stanford-hai-ai-index-report-2025-trends-2025-04-07/

27. Source: cor.europa.eu
Link:https://www.cor.europa.eu/en/our-work/opinions/cdr

28. Source: digital-strategy.ec.europa.eu
Title: eu A I Innovation Package | Shaping Europe’s digital future
Link:https://digital-strategy.ec.europa.eu/en/factpages/ai-innovation-package

29. Source: doi.org
Link:https://doi.org/10.48550/arXiv.2405.21015

30. Source: commission.europa.eu
Title: eu More support for artificial intelligence start-ups to boost innovation
Link:https://commission.europa.eu/news-and-media/news/more-support-artificial-intelligence-start-ups-boost-innovation-2024-01-24_en

31. Source: digital-strategy.ec.europa.eu
Link:https://digital-strategy.ec.europa.eu/en/news/commission-launches-ai-innovation-package-support-artificial-intelligence-startups-and-smes

32. Source: digital-strategy.ec.europa.eu
Link:https://digital-strategy.ec.europa.eu/en/library/communication-boosting-startups-and-innovation-trustworthy-artificial-intelligence

33. Source: interoperable-europe.ec.europa.eu
Link:https://interoperable-europe.ec.europa.eu/eu-digital-funding-opportunities

Additional References

34. Source: youtube.com
Link:https://www.youtube.com/watch?v=s6Ygo4_bpWU

35. Source: youtube.com
Title: Ep. 6 Is Nvidia Getting Too Big for Its Britches? | AI Insights and Innovation
Link:https://www.youtube.com/watch?v=vdiPD964TCM

36. Source: nature.com
Title: A I’s computing gap: academics lack access to powerful chips needed for research
Link:https://www.nature.com/articles/d41586-024-03792-6

37. Source: youtube.com
Title: The Dark Side of NVIDIA: The GPU Giant
Link:https://www.youtube.com/watch?v=fY9T8iYHwuQ

38. Source: frontier-economics.com
Link:https://www.frontier-economics.com/uk/en/news-and-insights/articles/article-i21406-ai-beyond-the-cloud-navigating-competition-innovation-and-regulation/

39. Source: cambridge.org
Link:https://www.cambridge.org/core/journals/cambridge-forum-on-ai-law-and-governance/article/architectures-of-ai-tech-power-broking-war/90CEC9E0D9E88CCD48301228A33F2335

40. Source: trendforce.com
Link:https://www.trendforce.com/presscenter/news/20260408-13003.html

41. Source: carnegieendowment.org
Link:https://carnegieendowment.org/research/2026/06/the-compute-coalition-how-to-build-the-future-of-ai-in-the-free-world

42. Source: siliconforward.com
Link:https://siliconforward.com/frontier-labs/frontier-ai-labs-compute-oligopoly-global-[distribution

43. Source: insight.goover.ai
Link:https://insight.goover.ai/report/202605/go-public-report-en-9d66ac1b-d653-4a32-ba59-4d7df63c53d7-0-0.html