Whilst London Tech Week runs this week, last week across the Channel (in Cannes) the great and the good of the infrastructure standing behind the digital revolution met for the Datacloud Global Congress. A topic of discussion across both events is the rise of "neoclouds," specialist GPU cloud providers purpose-built for AI workloads, offered a sharp reminder of the infrastructure race underpinning artificial intelligence. A panel session at Datacloud set out a clear message: demand for AI compute is extraordinary, but whether supply can keep up depends on solving a web of physical, financial, and regulatory constraints that software innovation alone cannot solve. The neocloud market has exploded. ABI Research forecasts that GPU-as-a-Service revenue from neocloud providers will surpass US$250 billion by 2030, up from around US$42 billion in 2025.
CoreWeave, the largest listed neocloud, showed the scale of ambition: its revenue backlog now stands at nearly US$100 billion and its platform supports nine out of ten of the largest AI users globally. Since its IPO, when around two-thirds of its revenue came from a single customer, the company has diversified rapidly, with major commitments from OpenAI, Meta and others now spreading the risk. Because of this extreme demand, CoreWeave's Ben Richardson point echoed throughout the panel: "speed is the currency in this market."
What gives neoclouds their edge over the traditional hyperscalers, is agility. Petter Tommeraas of DC Consulting AS described how NScale secured a Norwegian site and began building within weeks, while a Google project in the same country started a decade ago remains incomplete.
The demand is underpinned by sovereignty requirements and the technology revolution. Andy Long of Verda made this point: "This is another Industrial Revolution. Everybody has to do it."
But among the excitement, there is a note of caution. Of the 250-plus neocloud operators in the market today, Tommeraas predicted only around ten will survive a coming wave of consolidation – and only those with genuine scale, bankable clients and access to capital will endure.
Financing all of this is no small matter. Panellists drew parallels with the colocation market of the late 1990s, when lease terms were under two years and the sector was considered too capital-intensive. Capital markets eventually pushed terms longer, and the same dynamic is playing out now. Neocloud lease terms on the cloud side are extending beyond five years, and a new class of GPU-backed financing has emerged, with investors split between those backing the technology layer and those underwriting the physical infrastructure. Nvidia, has played a key part in making GPUs a bankable asset class. Richardson stressed that the cost of capital is falling rapidly. CoreWeave itself has reduced its weighted average cost of debt by some 600 basis points since 2023, and against the enormous sums involved, even marginal reductions are material. A lot of education by the neocloud operators is required though, because the data points the funders need to consider are not the same as in other parts of the sector (e.g. high levels of retention rates are an important factor).
Geographically, 80 to 85 per cent of demand currently originates from the United States, but this is expected to fall to around 70 per cent by 2030. The driver is not a slowdown in the US but the pull of sovereignty, as alluded above. The European Commission's recently announced Cloud and AI Development Act, part of its wider Tech Sovereignty Package, calls for a tripling of EU data centre capacity within five to seven years, with a new four-tier framework for assessing cloud sovereignty in public procurement. Long called it "an accelerator," whilst noting that sovereignty is not binary but involves multiple layers of control. Physical constraints in the US, including local opposition to new data centre developments, are also pushing neoclouds outward into Europe and Asia-Pacific.
For the UK, the picture was felt to be broadly positive. Richardson praised the UK Government's engagement and commitment, noting that CoreWeave has scaled rapidly in Britain largely for that reason. Patrick Berghaeger of InferX looked at the UK's AI Growth Zones as a non-Brit, specifically designated sites with streamlined planning, prioritised grid connections and energy cost incentives, "with some envy," seeing the UK as a potential bridge between the US and Europe. The UK has now designated five AIGZs, although from conversations throughout the past two weeks, it seems there is a feeling of contemplations as to whether the benefits they are designed to confer will materialise.
Asked for predictions for June 2027, there was agreement: more agentic AI consumption of infrastructure, a consequent, welcome and necessary continuing of the diversification of offtakers (the rise of enterprise customers), greater focus on inference as the monetisation of AI accelerates, continued consolidation in the neocloud market, and a need for the industry to improve its public image. As Tommeraas urged, the sector must stop talking in technical jargon and start explaining in plain language why AI infrastructure matters, for the problems ordinary people actually care about, like medicine and climate change. As he said, “what better use of energy is there than producing AI to assist in resolving these challenges”.
For lawyers advising in this space, the takeaways are practical: deal structures are evolving rapidly, with novel GPU financing, still relatively short but lengthening lease terms, and complex multi-party project arrangements. Regulatory frameworks, from EU sovereignty rules to UK planning reforms, are moving fast and creating both opportunity and risk. The window is narrow, and legal advisers have a real role to play in helping clients participate in the speed-to-market imperative – whether by accelerating deal execution, navigating new regulatory regimes, or structuring the increasingly complex financing arrangements that underpin these projects. A clear takeaway from Datacloud and further discussions throughout London Tech week is that those who can execute quickly will win; those who cannot, will be left behind.
