The Power Play: Public Compute and the Future of AI

Compute: The New Currency of AI

In the rapidly evolving world of artificial intelligence, compute, or computational power, has become the lifeblood of progress. No longer just a technical component, it's the fundamental currency shaping who can build, access, and ultimately control the future of AI.

As AI models become increasingly sophisticated, particularly with the rise of large language models (LLMs), access to vast compute resources is no longer a luxury—it's a necessity. This has created a compute divide, separating the AI giants with their vast data centers from smaller companies, researchers, and nonprofits.

Public Compute: Leveling the Playing Field

Enter public compute: government-funded initiatives designed to bridge this divide. These initiatives range from providing direct access to state-of-the-art hardware and cloud services to fostering vibrant ecosystems of support and collaboration.

But the landscape of public compute is complex. From the multi-billion dollar Stargate Project in the US to the UK's AI Opportunities Action Plan, nations are grappling with how best to invest public funds for maximum impact.

Four Families of Public Compute

Our research reveals four distinct approaches to public compute provision: direct provision (generalist and AI-focused), decentralized provision, and market-based provision.

Direct provision, whether generalist or AI-focused, involves government ownership and operation of compute infrastructure. Generalist approaches, like the US Department of Energy's supercomputers, aim to support a wide range of research, while AI-focused initiatives, such as the UK's AI Research Resource, prioritize cutting-edge AI development.

Decentralized provision, exemplified by India's Open Cloud Compute, distributes smaller facilities across regions, potentially fostering local innovation. Market-based provision, like compute vouchers in India's IndiaAI mission, leverages existing commercial providers through subsidies or credits.

Navigating the Challenges

Each approach presents its own set of challenges. Direct provision faces risks of inefficient spending and legacy technology, while decentralized provision grapples with coordination and infrastructure limitations. Market-based solutions can inadvertently reinforce existing monopolies. Crucially, all models must grapple with value capture, ensuring public investment benefits the public, not just private interests.

Recommendations for Policymakers

To maximize the impact of public compute investments, policymakers must address these challenges head-on. Our research suggests several key recommendations:

• Insulate public compute strategies from undue industry influence.
• Develop holistic industrial strategies for compute procurement.
• Explore conditional access to ensure public benefit.
• Provide essential support resources beyond raw compute power.
• Implement complementary regulatory interventions to address market concentration.
• Prioritize strategic coherence and coordination between initiatives.
• Balance flexibility and longevity in planning and procurement.
• Integrate environmental sustainability goals into compute strategies.

The Future of Public Compute

Public compute stands at the forefront of AI policy. Done right, these initiatives can democratize access to this transformative technology, fuel public interest research, and ensure a more equitable future. By embracing a strategic, holistic approach, governments can harness the power of public compute to shape AI's trajectory for the benefit of all.