GPUaaS revenue will quadruple in the next 5 years, powering new data-centre investment opportunities

Long-term, multi-million-dollar leases between neocloud tenants and co-location providers are becoming increasingly common.

This is evidence that the maturing neocloud industry is entering a new phase: there are new opportunities for investors, but some real risks need to be assessed.

The neocloud leasing model is set to change in the coming year

‘Neocloud’ refers to newer cloud service models that offer a leaner service catalogue, mostly centred around graphics processing unit (GPU) compute services, and is often referred to as ‘GPU as a service’ (GPUaaS). These neoclouds need a vast number of racks housing GPUs hosted in advanced, liquid-cooled facilities.

Procuring liquid-cooling-capable facilities is a challenge. Most neocloud players turn to a co-location model, often built-to-suit, which gives them the safest, shortest, fastest path to market for deploying data-centre capacity.

2026 will see some important departures from the early neocloud model, which was characterised by small, opportunistic, short-term footprints. With large AI players (like hyperscalers) showing no sign of slowing down, neoclouds are now starting to seize their opportunity to grow into very large, long-term leased capacity blocks.

The neocloud model is becoming well entrenched in the US digital ecosystem, and Western European markets are starting to accelerate as well

In the USA, the trend is already apparent, with the example of Applied Digital’s two 15‑year leases with CoreWeave, worth about USD7 billion in total. In Western Europe, growth is coming from the expansion of US players, as seen with CoreWeave’s investments in Spain and Scotland, and some is from the development of smaller players, like Ori in the UK. While leases beyond 10 years have yet to be seen, longer-term partnerships are emerging, visible in Eclairion’s multi-year deal in France with Fluidstack.

The AI co-location market is maturing

The short‑tenor, flexible arrangements of early GPUaaS deployments reflected a variety of co-location market constraints, both on the demand and supply side. On the demand side, tensions in the supply chains (especially GPUs) made the deployment of very large footprints challenging. For players that were able to source equipment at scale, competitive pressure led to a race to make compute capacity available. On the supply side, the lack of standardisation in liquid cooling architecture brought risk – especially the high refurbishment costs for the data-centre host in the (likely) event of the neocloud churning at the end of its short-term lease. This translated into short-term contracts (typically 3 to 5 years) commanding a substantial rent premium.

In 2026, growing confidence in neocloud technology and business models will be allied with the strong need for greater security in revenue to create a ‘new normal’: larger, longer-term tenancies with clearer escalators and delivery schedules. Both pre‑leasing levels (that is, the amount of capacity that is sold before construction is complete) and scale are increasing, which lowers the effective rent for neocloud anchor tenants that can make firm commitments in terms of power and duration.

The data-centre infrastructure investment paradigm is changing

While leases are becoming longer, demand from neoclouds is changing the landscape of the co-location market.

There is a trend away from expensive, complex, core locations in the leading (FLAP-D¹) markets in Western Europe. As many neocloud workflows are less sensitive to latency, new markets (such as the Nordic region or Iberia, or in Tier II cities in more mature markets) become viable. Locations where capacity can be deployed more quickly, easily and cheaply become more attractive. In addition to cheaper leases and shorter times to market, these locations can also offer lower-cost and greener energy (for example, hydropower in the Nordics).

Another trend that will gather pace in 2026 is the move away from standard Tier III-type reliability architectures. Tier III equivalent infrastructures remain the gold standard for co-location, with the best combination of reliability and cost-effectiveness for cloud-like applications. But for many neocloud workflows, minor service interruptions can be tolerated, especially if the corollary is cheaper leases. Some neocloud players are considering using Tier I or II equivalent architectures, at least for some of their hosting requirements (typically for training purposes, but could extend to non-critical inference): the lower level of reliability balances the smaller amount of equipment to be deployed (for example, back-up generators). This translates into lower capex and potentially faster build time for the co-location operator, and ultimately cheaper leases for neoclouds.

The long-term viability of the business model remains uncertain

Booming demand for neocloud co-location is making it an increasingly interesting play for data-centre operators and investors, but there are risks.

The primary risk is the long-term sustainability of neoclouds. The current AI wars between historical giants and new emerging leaders mean demand for GPU compute is so high that neoclouds can easily find tenants, especially with NVIDIA ensuring these players get priority supply of the latest GPUs. However, the counterparty concentration raises questions about the long-term viability of some neoclouds, especially in the context of some hyperscalers building their own chips for AI applications.

The secondary risk is that of infrastructure obsolescence. Liquid cooling technology is not yet fully standardised, and a churning tenant might trigger a substantial and costly retrofit. However, much of the current hardware that may need to be retrofitted is in the ‘white space’, meaning that retrofit costs are often borne largely by the tenant, rather than the data-centre operator. Nonetheless, data-centre operators are increasingly adopting modular technology, which allows technical modules (for example, cooling units, power distribution units) to be switched relatively easily.

Investors must consider a few key factors in a neocloud hosting infrastructure deal

2026 will see the neocloud investment opportunity maturing and expanding, but success for data-centre operators and investors requires careful attention to the following.

• Tenor and ramp. Players must prioritise infrastructure projects with anchor tenants secured through 10- to 15‑year master service agreements (MSAs) with staged megawatt (MW) ramps and indexed escalators, in line with what can be expected from very large wholesale or hyperscale contracts.

• Liquid cooling readiness. Players must require warm/cold loops (for liquid and air-cooled racks) and ensure as much of the infrastructure as possible is handled by the tenants (CDUs, leak detection and so forth).

• Counterparty risk. Players must prioritise tenants with multiple clients/low client concentration and ensure, whenever possible, that the leases reflect contractual commitments by the neocloud’s own tenants.

¹ Frankfurt, London, Amsterdam, Paris, Dublin, the historical data-centre hubs.