The state of the data-centre market, October 2026: power is the new land

The data-centre market in late 2026 is defined by a simple imbalance. Demand for AI capacity is contracted years ahead, capital is plentiful, and the buildings themselves are well understood. What is scarce is deliverable power on a date an occupier will sign against. For investors, that changes where risk sits and where value is created.

This note summarises the market as of October 2026, drawing on Data Center Dynamics reporting cross-checked against JLL, CBRE, Colliers, NESO and company disclosures.

Demand: contracted, concentrated and early

The largest US hyperscalers are spending at a pace without precedent, with combined capital expenditure of roughly $700bn expected for 2026. Much of that is committed through long-term leases and build-to-suit agreements signed before construction starts.

The second wave of demand comes from neoclouds and AI labs. CoreWeave alone reports around 1.5 GW of capacity live and about 4.2 GW contracted. Enterprise and sovereign programmes add a third layer, particularly in the Gulf.

The practical consequence is that new capacity is largely pre-let. In the US, vacancy across primary markets is around 1%. In London, CBRE's mid-year review expects vacancy to reach an all-time low of 5.5% by the end of 2026, with new supply in Essex and Hertfordshire not expected before 2028 or 2029, depending on grid availability.

Supply: plenty of capital, not enough power

Roughly 31.7 GW of data-centre capacity is under construction globally. The pipeline behind it is much larger, and most of it is waiting on electricity rather than on funding.

Two responses have emerged. The first is behind-the-meter generation: gas engines, turbines and fuel cells built on site to bridge the years before a grid connection, or to replace it entirely. The second is a scramble for sites that already hold a connection, which is why connected land near London and in the major US hubs now prices like infrastructure rather than real estate.

Equipment: the hidden critical path

Once a site has a power strategy, the next constraint is equipment. Large heavy-duty gas turbines are effectively sold out beyond 2029; GE Vernova reported a gas-power backlog of about 83 GW at the end of 2025. Aeroderivative turbines are quoted anywhere from 12 to 40 months. Reciprocating gas engines, at 12 to 24 months, have become the 2026 workaround, which is why so many recent "turbine" headlines are in fact engine orders.

Costs have moved with scarcity. BNEF puts new combined-cycle plant at about $2,157 per kW, 66% higher than in 2023. Rabobank estimates reciprocating engine plants at $1,700 to $2,000 per kW installed. Electricity from on-site generation typically costs $100 to $165 per MWh, against $90 to $95 for grid supply in the same markets.

For an investor, that premium is often worth paying. The value of a powered campus that energises two or three years early dwarfs the extra cost of the electrons.

The Gulf: open, but more selective

The Gulf remains one of the most active growth regions, led by sovereign-backed platforms in the UAE and Saudi Arabia. Two developments in 2026 shape the investment case. In March, drone strikes affected cloud facilities in the UAE and Bahrain, and JLL estimates around 2.6 GW of Middle East projects were subsequently deferred. In July, the US moved the UAE into its unrestricted tier for advanced AI chip exports, removing a major compute-supply risk for UAE projects.

The result is a market that still rewards strong sponsors with secured power and compute, and is less forgiving of speculative capacity.

Where value is forming

Across markets, the same pattern holds. Value accrues to whoever controls the scarce input at the right moment:

• Powered land and connection rights, especially where a connection date is contractually firm.
• On-site generation and equipment slots, which convert a five-year grid wait into an 18 to 24-month build.
• Operator-led development positions, before a formal capital raise, where entry pricing reflects development risk that a secured power plan has already reduced.
• Contracted compute capacity, where multi-year GPU agreements now underpin the asset.

The corollary is that the largest risk in a 2026 data-centre investment is rarely the building or the tenant. It is the date on which power and equipment arrive.

What this means for investors

• Underwrite the power plan first: connection offer, date, firmness, and the bridge strategy if the grid date slips.
• Ask which long-lead equipment is secured, with slot dates, and which is still only quoted.
• Treat pre-let quality and power certainty as linked: an occupier will only sign against a credible energisation date.
• In the Gulf, test sponsor strength, compute-supply access and the security and resilience plan.
• Price the gap between grid and on-site power costs, but weigh it against the value of energising years earlier.
• Nodus works with institutional investors and capital allocators on qualified opportunities screened for power, site and capital readiness.