Artificial intelligence (AI) compute has become a strategic national asset, with venture capital for AI infrastructure more than doubling year-on-year to $123.9bn in 2025. Meanwhile, the UK’s commitment to spend up to £2bn on its Compute Roadmap is one of the leading bets placed by governments around the world.
That’s according to the Organisation for Economic Co-operation and Development (OECD), which published its Digital economy and society outlook 2026 today.
In the report, the OECD says AI compute now sits alongside data, algorithms and skills as a determinant of who can build and deploy advanced AI systems. It says venture capital investment in AI firms in IT infrastructure and hosting rose from $54.5bn in 2024 to $123.9bn in 2025, which is the largest increase since 2012.
The report namechecks the UK Compute Roadmap, launched in 2025 with up to £2bn pledged to “deliver a holistic and user-centred compute ecosystem” and £1bn to expand the AI Research Resource, plus a new national supercomputing service. It also notes the emergence of neocloud providers of infrastructure purpose-built for AI workloads, such as the UK’s Nscale, the US’s CoreWeave and Netherlands-headquartered Nebius.
The OECD is an intergovernmental body whose member governments use it to agree non-binding standards for digital policy. Its reports do not set law, but governments fold their recommendations into regulation – such as the EU AI Act – so the organisation is an early indicator of auditability and data governance requirements that can later become compliance obligations.
The OECD warns in the report that sizeable capital investment in infrastructure, networking and energy access gives traditional hyperscalers a considerable market advantage, which is compounded by developer ecosystems and hardware partnerships that build dependencies. Newer AI-specific providers may expand, it says, but buyers will likely continue to rely on existing hyperscaler capacity – a familiar dynamic for UK organisations balancing cloud lock-in against AI ambition.
The report flags two bottlenecks that may limit compute performance. The first is that accelerator performance has tripled every two years while memory (DRAM) has scaled at just 1.6 times over the same period – a gap it calls “the memory wall”, which can leave processors idle and waiting for data. Moving data between chips forms a second constraint – “the interconnect wall”, which slows distributed training and inference across racks and nodes.
For IT leaders, it says, compute also collides with energy and planning. The OECD notes that hyperscale datacentre buildouts can take up to three years, while securing multi-gigawatt power can take more than four – a mismatch that already shapes grid connection queues and the UK’s push towards 6GW of AI-capable capacity.
“Alongside data, algorithms and skills, AI compute has emerged as a strategic asset with profound implications for innovation and competitiveness,” the report says.
Other government investments in AI include Canada, which committed C$2bn (approx. £1bn) to a Sovereign AI Compute Strategy in 2024; the EU, which will channel €7bn (approx. £5.5bn) through its EuroHPC joint undertaking for 2021-27; and the US, which backs the Stargate datacentre venture.
The OECD report finds that 60% of member countries now assign national digital strategy to a dedicated digital ministry, up from 24% in 2016, and that cyber security has risen to the second-highest policy priority.
The scale of the figures demands caution, the report says. “Figures often reflect announcements and commitments rather than implementation,” it says, with delivery dependent on supply chain resilience, energy and water availability, and shifting national priorities.
On that reading, the investment surge can be viewed as a statement of intent rather than evidence of built capacity. Turning the UK’s £2bn into operational infrastructure will depend on grid and planning constraints that decide whether promised datacentres arrive on time.
