From AI Strategy to Implementation: Insights from the AIP2 Lab in Washington DC

On 12 January 2026, Access Partnership convened the first of the AI Policy to Practice (AIP2) Labs in Washington DC, bringing together leaders from technology companies, government agencies, financial institutions, and civil society. Organised in partnership with the Atlantic Council, the session focused on what it takes to move from AI strategy to implementation.

The conversation centred on four core areas: scaling compute and digital infrastructure, data governance and cross-border data flows, the energy and power demands of AI systems, and how governments can translate industrial AI strategies into tangible deployment outcomes.

Democratic systems and practical partnerships

A central theme emerged early: democratic systems in democratic nations should shape AI deployment and investment. This principle extends beyond regulatory frameworks to encompass how practical tools and partnerships can translate policy into action. The India Impact Summit initiatives offer one example of how convening diverse stakeholders can bridge the gap between strategy and implementation.

Powering AI: Energy constraints and opportunities

The challenges of powering AI, particularly hyperscale data centres presents one of the most immediate constraints to scaling deployment. Rising power demands are forcing a rethink of energy infrastructure, creating both obstacles and opportunities. Interest is growing in diverse energy sources such as nuclear, small modular reactors, and geothermal, alongside greater interconnection among clustered data centres to optimise energy use.

Several strategies could help address these challenges. Community engagement around data centre development can build local support and identify shared priorities. Industry commitments on sustainability and grid impact provide accountability and clarity for policymakers. Leveraging federal assets, including federal land for data centres, offers one pathway to support future development and grid modernisation without competing directly with other land uses.

Policy frameworks and deployment barriers

Infrastructure alone won’t solve the deployment challenge. Compute and energy constraints intersect with policy decisions around tax incentives, procurement rules, and regulatory frameworks. The EU’s restrictive procurement rules and differences in the US’s talent environment illustrate how policy choices shape where and how companies can build and deploy AI products at scale.

Navigating AI sovereignty

Nations increasingly view AI as a critical national asset, with significant implications for how data governance and infrastructure decisions are made. Fragmented data transfer rules create barriers that complicate cross-border AI development, even as global supply chains and user bases demand it.

A practical tension surfaced throughout the discussion: data centres cannot be built in every country, yet latency-related user drop-off remains a significant challenge. Proximate compute matters in certain regions, particularly where connectivity gaps and security vulnerabilities already limit access for underserved populations.

Ensuring these populations can access and benefit from AI meaningfully requires more than infrastructure. Digitalising local history, language, and cultural context as input for AI models can make the technology relevant and useful, rather than a tool designed primarily for dominant markets and languages.

Edge devices and the AI stack

The conversation turned to how edge devices fit within the broader AI ecosystem and why distinguishing between data centre and edge device ‘AI stacks’ matters. Edge devices will not replace the cloud or centralised data centres, but advances in on-device AI can complement centralised compute and expand access, particularly in regions with connectivity constraints.

The popularity of Chinese AI models in some regions came up as a reminder that technological competition is already global. Strengthening the US AI stack and tech diplomacy efforts will be essential to maintaining competitiveness in markets where alternatives are readily available.

Building resilient innovation environments

The discussion concluded with a focus on what enables sustainable AI deployment over the long term. Flexible regulation that adapts to technological change, resilience against misinformation, and stronger environments for R&D and local innovation all emerged as critical enablers.

Three key outcomes emerged from the session:

1. Enable project-financing for data centre development. Innovative financing mechanisms are needed to support the scale of infrastructure investment required for AI deployment.
2. Support the development of regional compute hubs to address latency constraints. Strategic investment in regional infrastructure can balance the economic realities of data centre development with the user experience requirements of low-latency applications.
3. Develop differentiated “AI stack” frameworks for edge devices and data centres. Recognising the distinct technical and policy considerations for edge computing versus centralised infrastructure can enable more targeted and effective approaches to AI deployment.