Cloud & Infrastructure • 1 day ago • Neha Jamwal
AI is changing how cloud providers compete. As demand for GPUs, power and data-centre capacity surges, specialised cloud providers are emerging to serve AI workloads alongside the major hyperscalers. For enterprises, that is changing not only where they run AI, but also how they evaluate and buy cloud infrastructure..
For years, cloud competition was relatively easy to understand. Amazon Web Services, Microsoft Azure and Google Cloud built increasingly broad platforms and competed on price, scale, availability, services and developer ecosystems. AI is disrupting that model. The defining constraint of the next phase of cloud computing may not be software capability or even access to a particular AI model, but infrastructure capacity: the ability to secure GPUs and other accelerators, build data centres, obtain enough power, connect thousands of processors at high speed and deliver that capacity reliably to customers.
That is creating an opening for a new category of cloud provider. Known as neoclouds, these companies are building infrastructure specifically for AI and high-performance workloads rather than attempting to reproduce the broadest possible general-purpose cloud platform. Gartner estimates that neocloud providers could capture 20% of a $267 billion AI-cloud market by 2030, reflecting growing demand for AI-optimised infrastructure, flexible deployment and, increasingly, sovereign capabilities.
The emergence of these providers does not mean hyperscalers are losing their position. Quite the opposite: AWS, Microsoft, Google and Oracle are spending unprecedented amounts on AI infrastructure. But it does mean that the definition of a cloud provider is changing — and the competitive battleground is moving closer to the physical infrastructure underneath the cloud.
The cloud race is becoming a capacity race
The scale of recent infrastructure spending illustrates the shift. Oracle’s latest quarterly results provide one of the clearest examples. The company reported more than $30 billion in new AI cloud contracts in its first fiscal quarter, pushing its total remaining performance obligations to $664 billion. At the same time, Oracle spent $28.5 billion on capital expenditure during the quarter and maintained its fiscal 2027 capex forecast at $90–95 billion.
That is an extraordinary amount of infrastructure investment for a company historically better known for enterprise software and databases. Oracle is no longer simply selling software that runs on someone else’s infrastructure; it is increasingly competing to provide the physical infrastructure required to run AI. Its cloud infrastructure revenue grew 121% year over year in the latest quarter, highlighting just how quickly AI demand is translating into infrastructure consumption.
The same pattern is visible across the industry. Amazon has entered a long-term partnership with Qualcomm under which Amazon could purchase up to $60 billion of AI data-centre chips and related products, while the companies are also working on advanced optical connectivity for AI infrastructure. Google, meanwhile, announced a $15.1 billion investment in AI infrastructure in Finland over the next two years, including three data centres and a 22-year agreement to purchase up to half the output of a Finnish nuclear plant.
These developments point to the same conclusion: hyperscalers are increasingly designing, financing and operating infrastructure around specific AI workloads rather than relying entirely on a standardised pool of general-purpose cloud capacity. The cloud is becoming more specialised because AI workloads are becoming more demanding.
Why the neocloud emerged
Traditional cloud platforms were designed to serve an enormous range of workloads: databases, enterprise applications, websites, analytics, storage and developer environments. AI has introduced workloads with very different infrastructure requirements. Training and serving large models can require large numbers of accelerators working together, high-bandwidth networking, specialised storage and enormous amounts of electricity. Performance can depend on how efficiently thousands of components communicate with one another.
That creates an opportunity for providers willing to optimise almost everything around AI. Instead of building a cloud that does everything reasonably well, a neocloud can focus on doing a narrower set of things extremely well. The result is a different value proposition: access to scarce GPU capacity, AI-optimised infrastructure, specialised networking, flexible deployment and, increasingly, sovereign infrastructure.
Gartner describes neoclouds as providers built specifically for AI and high-performance workloads and says they are gaining traction because traditional cloud models can struggle to meet the requirements of GPU-intensive workloads. Its research also identifies sovereignty, performance and infrastructure specialisation as increasingly important factors in enterprise cloud decisions.
That distinction matters because scarcity changes the customer’s priorities. When computing capacity is readily available, enterprises can focus primarily on price and convenience. When capacity is constrained, simply securing the right infrastructure becomes a competitive advantage. A provider that can offer the right GPU capacity, in the right location, with the right networking and power infrastructure can therefore compete on something more fundamental than price.
GPUs are only part of the problem
It is tempting to describe the AI infrastructure race as a race for GPUs. That is incomplete. A data centre can have the latest accelerators and still fail to deliver useful AI capacity if it cannot secure enough electricity, cooling, networking equipment, physical space or connectivity.
This is why AI infrastructure is increasingly becoming an integrated systems problem. The Qualcomm-Amazon collaboration is revealing in this respect. The companies are working not only on customised AI chips but also on high-speed optical connectivity, addressing the growing requirements for compute, networking and bandwidth in large-scale AI data centres.
Power is another constraint. Google’s Finland investment illustrates how closely AI infrastructure and energy strategy are becoming linked. The company is pairing new data-centre capacity with long-term access to nuclear power, while Finland is already debating whether the rapid growth of data-centre demand could put pressure on electricity supply and transmission capacity.
That makes access to land, electricity, cooling, grid capacity and suitable connectivity increasingly strategic for cloud providers. The AI infrastructure race is therefore starting to look less like conventional software competition and more like an industrial buildout.
The financing problem behind the AI boom
There is another dimension to this transformation that enterprise technology buyers should pay attention to: who is financing all this capacity?
AI infrastructure requires enormous upfront investment, while the economic returns from that investment may take years to materialise. Oracle’s latest results illustrate both sides of the equation. Its AI business is generating significant contracted demand, but the company is also spending tens of billions of dollars on infrastructure. Its $28.5 billion quarterly capex included significant investment in data centres and hardware, while $11.36 billion was offset by customer prepayments.
That customer-financing element is particularly important. The traditional cloud model largely asks providers to build infrastructure and customers to consume it. The AI infrastructure market is increasingly creating arrangements in which customers commit to capacity, prepay for infrastructure or provide hardware themselves. Oracle has explicitly discussed models including customer prepayments and “bring your own hardware” arrangements as ways of spreading the capital burden.
For enterprises, a long-term AI infrastructure commitment is therefore no longer simply a procurement decision. It can become a capacity, financial and counterparty-risk decision. The question is not only whether a provider can offer enough compute today, but whether it will have the financial strength, hardware supply and physical infrastructure to deliver that capacity several years from now.
Hyperscalers are fighting back with custom silicon
The rise of neoclouds does not mean the hyperscalers will simply surrender specialised AI workloads. They are responding by going deeper into the infrastructure stack themselves. Amazon has developed its own accelerator technology, while Microsoft and Google have also invested heavily in custom AI silicon. Qualcomm’s new partnership with Amazon is another sign that the market for alternative AI accelerators is expanding.
There is a practical reason for this. Owning more of the infrastructure stack can give cloud providers greater control over performance, cost and supply while reducing dependence on a single accelerator supplier. The result is a cloud market in which differentiation is increasingly happening below the software layer.
Silicon, networking, power efficiency and data-centre design are becoming cloud features.
Neoclouds are not necessarily trying to replace AWS or Azure
The most interesting outcome may not be a battle in which neoclouds displace hyperscalers. It may instead be a more fragmented cloud market in which different providers serve different infrastructure requirements.
Gartner expects neoclouds to become a significant part of the AI-cloud market precisely because enterprises have increasingly specialised requirements. Its research points to AI-optimised infrastructure, flexible deployment, performance and data sovereignty as factors helping these providers gain traction.
That suggests a future in which an enterprise might use a hyperscaler for its core cloud environment, a neocloud for a particular AI workload and another provider for sovereign or regulated infrastructure. AI could therefore accelerate the move from cloud consolidation toward workload-specific infrastructure choices.
That has implications for cloud architecture, procurement and FinOps. The question for CIOs may increasingly become not “Which cloud should we choose?” but “Which infrastructure is economically and technically appropriate for each workload?”
What enterprise buyers should watch
For enterprises evaluating AI infrastructure providers, the obvious questions — GPU type, performance and hourly price — are no longer enough. Buyers will increasingly need to examine capacity availability, power and data-centre access, hardware supply, networking, financial strength, contract structure, data sovereignty, portability and operational resilience.
A provider promising future GPU capacity, for example, should be evaluated differently from one offering immediately available compute. Enterprises should also understand whether a provider owns its facilities, leases capacity or relies on third parties, how it finances expansion and what happens if hardware availability or construction schedules change.
These questions become particularly important as enterprises move from AI experimentation into production. A cheap GPU hour is not necessarily cheap if capacity disappears when the business needs to scale, or if moving the workload to another provider later proves prohibitively difficult.
The bigger shift: cloud is becoming infrastructure procurement
The rise of neoclouds ultimately points to something bigger than the emergence of another category of cloud vendor. It suggests that AI is pulling cloud computing closer to the physical infrastructure underneath it.
For years, cloud abstracted away the physical data centre. Customers could largely think in terms of virtual machines, storage, APIs and managed services. AI is reversing some of that abstraction. Enterprises increasingly care about the accelerator underneath the workload, the networking fabric connecting it, the power available to the data centre and the location and jurisdiction in which the infrastructure operates.
The physical layer is becoming strategically important again. That does not make the cloud model obsolete; it makes the infrastructure underneath the cloud more consequential.
And that is why the neocloud story deserves attention. The winners of the next phase of cloud computing may not simply be the companies with the biggest software platforms. They may be the companies that can secure power, build capacity, finance infrastructure and deliver specialized computing capacity reliably at the scale AI demands.
The cloud race is becoming a capacity race. For enterprises, the next cloud decision may increasingly look like an infrastructure investment decision.
Key Takeaways
- AI is turning cloud infrastructure into a capacity race. Access to GPUs, power, networking and data-centre capacity is becoming as important as software capabilities and pricing.
- Neoclouds are emerging to fill specialised AI infrastructure needs. Rather than competing with hyperscalers across the entire cloud stack, they are focusing on AI and high-performance workloads where dedicated infrastructure can offer an advantage.
- Hyperscalers are responding by moving deeper into the infrastructure stack. Custom AI chips, specialised networking and massive data-centre investments are becoming central to cloud strategy.
- Power and physical infrastructure are becoming strategic cloud assets. AI data centres require enormous amounts of electricity, making energy availability, grid access, cooling and location increasingly important to infrastructure planning.
- AI infrastructure is creating new financial risks. Massive capital requirements and long-term capacity commitments mean enterprises need to evaluate a provider’s financial strength and ability to deliver capacity over time.
- Enterprise cloud procurement is changing. Buyers will increasingly need to evaluate compute availability, infrastructure resilience, portability, sovereignty and counterparty risk — not just GPU pricing.
- The future is likely to be more workload-specific. Enterprises may combine hyperscalers, neoclouds and sovereign infrastructure providers depending on the technical, economic and regulatory requirements of each AI workload.