AI is driving record valuations, but the real constraint isn’t chips, it’s infrastructure. This is according to AI Infrastructure: The Physical Bottleneck of AI report by Heligan Group that calls to attention the fastest-emerging constraint in the AI economy: power, land, and cooling capacity, not GPUs.
“As hyperscalers push past historic investment levels and model sizes continue to accelerate, the physical layer of AI infrastructure has become the limiting factor to global innovation. A single NVIDIA H100 draws up to 1,000 watts; racks exceed 100 kW. This is no longer a race for silicon – it’s a battle for grid access, land and engineering capability” said Andrew Dickinson, Head of Infrastructure Services at Heligan Group.
“AWS’ recent commitment to invest up to $50 billion to AI and supercomputing infrastructure for the U.S. government is a perfect example: the global AI arms race is shifting from chips to the physical systems that power, cool and house them. For investors, the opportunity lies in platforms that can deliver power, cooling and scalability at speed.”
Heligan’s analysis shows that global data centre M&A is expected to exceed $80 billion in 2025, while more hyperscale’s are projected to invest $400 billion this year to secure power-ready capacity.
“This is consolidation with strategic intent and deals over $10 billion are now routine. AI workloads are redefining what infrastructure must deliver. Platforms are being valued not just for installed capacity but on expandable megawatts, retrofit potential, and proximity to reliable power.
“The engineering layer of data-centre design is also becoming a decisive competitive advantage. Where legacy data centres were engineered for predictable loads, AI demands thermally intense, rapidly scaling environments. Cooling is shifting from traditional air to liquid loops, immersion tanks, and submerged modules. Some operators are even experimenting with offshore locations and deep-sea water for thermal efficiency.”
Dickinson adds, “Power architecture is evolving alongside this. Vertical substations, on-site generation, and dynamic load balancing are replacing legacy cabling and redundancy systems. The most valuable platforms are built to retrofit and reconfigure for changing chip designs, sustainability mandates, and regulation.”
At the same time, grid strain is already triggering moratoria in megahubs such as Northern Virginia, Dublin and Frankfurt, with construction lead times now exceeding five years in many regions.
“Energy access, permitting and build viability now outrank tenant demand. With cooling accounting for up to 40% of total energy use, investors are targeting platforms with waste heat recovery, district heating, and renewable baseloads. The ability to demonstrate low carbon delivery is emerging as a pricing premium.”
Despite global pressure, the report highlights the UK as an emerging leader in AI-ready infrastructure.
“The UK is highlighted as a standout market, with a data centre pipeline now exceeding £36 billion and power capacity forecast to reach 6GW by 2030 under the government’s Compute Roadmap.
“UK engineering and construction firms are rapidly expanding to meet hyperscale demand, delivering modular builds, advanced cooling systems and grid-adjacent campuses. his positions the UK as Europe’s most competitive market for AI infrastructure, with hyperscale campuses anchoring long-term growth.”
Dickinson concluded, “AI’s expansion is no longer defined by chips; it’s constrained by infrastructure. Control of physical systems is now driving competitive edge, and the winners will be those who can build, power and cool at speed.”
“The UK has the momentum, engineering depth and policy alignment to anchor the next decade of global AI capacity and we expect investment activity to secure megawatts and own the platforms that power intelligence. The most valuable assets will be those built to reconfigure for changing chip designs, sustainability mandates and regulatory pressure.”
