Neoclouds – data centre evolution or revolution?

Do Neoclouds represent an evolution of cloud computing or is a whole new class of data centre design and scale going to shake up the market. 

There is no doubt that the Neoclouds have landed. The discussion is whether the services they provide are simply a natural evolution of cloud that fills an important gap in providing organizations with fast access to AI-native, GPU-powered, and scalable computing platforms or whether those classed as ‘Neos’ signify a new revolution in hyperscale data centres.  

A pure definition of NEO Cloud has yet to emerge. Some NEO cloud data centre operators present as hybrid GPU as a service providers combined with traditional cloud data centre offerings such as bare metal, IAAS, and cloud Kubernetes. (See box out below.)

For other players NEO Cloud marks the beginning of a radical shake up of the sector in terms of size of addressable market, GW scale of new builds for pure high performance compute, high density design and low carbon power. 

It is fast moving space. Neocloud data centre providers and GPU-as-a-service specialists are attracting significant investor interest and even as traditional players pivot NVIDIA itself has launched  into the GPU Cloud market. (See bow out below - A selection of Neo Cloud player investment rounds, M&A activity and data centre strategies to date)

The GPU-as-a-Service market is projected to grow significantly in the next five years – industry reports suggest a USD13bn market by 2030 in the US, Europe and Asia.

The investment valuations of individual NEO cloud players have already reached $10bn scale with CoreWeave raising $10bn+ in equity and debt and Oxford University founded Fluidstack having signed a $10bn MOU with the French Government to develop a 1GW AI supercomputer campus. 

The impact of Neocoud has been rapid says Penny Madsen, VP, Strategy Research, EMEA, STACK Infrastructure. 

“The rise of GPU-as-a-Service and Neoclouds has been quite fast. From AI training to inferencing, we expect the AI-focussed data centre landscape to shift over the coming years as enterprises work out how they will deploy AI. In the meantime, data centre providers need to remain focused on being responsible custodians, not only of data and infrastructure, but land, power, and other resources too. Energy efficiency becomes even more important – from server use down to data centre cooling and supply chains,” she says.

Speaking with Platform Insights,  César Maklary, co-founder of Fluidstack, an AI platform trusted by the world’s leading labs which as founded in 2017 at Oxford University covered the firm’s unique approach to scaling the AI data center market.

In February 2025, Fluidstack signed an MOU with the French Government for a 1GW Decarbonized Energy-Powered Data Center through an initial €10Bn investment. In April, it announced a strategic financing deal with Macquarie to supply GPUs to premier AI research labs across Europe.

‘Because we were profitable from day one, we have had the luxury of choosing our investors. Macquarie are one of many good finance partners,’ says César.

‘In terms of data centre strategy, right now we believe that the current bottleneck to providing AI is the access to compute, but most importantly the access to data centres and soon the access to power.’  

Maklary says the key to solving the problem of compute relies on the ability to build data centres extremely quickly. The company has plans for GW size build outs in 2026, with multiple GW size build outs in 2027 and beyond. 

‘That's what we're really bracing for. France is a logical location for us because they have very stable, sustainable power with a surplus of 10 GW of power every year. This is set to grow further as more nuclear powerplants come on stream. Coupled with low, stable energy prices, it presents a positive scenario for AI data centre builds at GW scale.’

While some Neo Clouds rely on a mixture of existing colocation and build outs or acquisition, Fluidstack takes a different approach to serving its key markets. The company has focused on the compute needs of large AI labs, ‘really the leading frontier labs such as Meta, Black Forest Labs, and many others.’

‘Our software sits on top of 150 existing data centers where we have readily-available capacity. We don't build speculatively. When we build out compute server and data centre infrastructure, we build at a very large scale on a project by project basis for contracted customers. Our business is not to build first, sell capacity second. But we do look to secure power or have options to purchase powered land in advance because we know this will be the bottleneck.’

Most data centers being built today have a typical build time of 18 months. César says that for Fluidstack, the crux of success is being able to build and deliver a complete data centre in a third of that timeline. This requires them to handle significant operational workloads upfront, as well as secure a very strong supply chain for specific components, especially cooling and the MEP Infrastructure. The rest is  about execution, he says.

‘I think there's a lot of legacy work still being done. We are working on completely new designs which are specific to high density. These are completely different to traditional cloud data centres. I think some people have just settled or accepted that 18 to 24 months for a data centre build was normal. If you review everything from first principles, you realise that you can build things much, much faster.’

Conclusion

In a note in February 2025, the Uptime Institute described Neo Clouds as ‘a cost-effective AI infrastructure alternative.’  And in December 24 Forbes reported that investors had to that  point poured $20bn into Neo Clouds ‘driving the arms race.’ 

How the GPU as a service market builds momentum will direct the evolution of this new ecosystem. At the same time building NEO Cloud AI data centres at GW scale sounds like the heralding of a revolution in investment, design, build and operation.