Over the last few months, the artificial intelligence industry has been flooded with eye-catching infrastructure deals — and almost all of them revolve around a new class of vendors known as neocloud providers. These are up-and-coming cloud companies created specifically to service the heavy computational demands of generative AI. Instead of offering the sprawling suite of enterprise services that define Amazon Web Services, Google Cloud and Azure, neoclouds focus on delivering high-performance GPU clusters, ultra-fast networking and data centers engineered solely to train and run large AI models.
What is a Neocloud Provider?
Neocloud providers are cloud startups built to handle AI workloads. They run and train large models on specialized infrastructure, widening access to high-performance GPU computing at lower cost than traditional hyperscalers.
This once-niche category is now at the center of some of the industry’s biggest deals. Anthropic announced a $50 billion partnership with the neocloud firm Fluidstack to build custom, AI-focused data centers in Texas and New York explicitly designed to support future iterations of Claude. And Microsoft’s neocloud spending has reportedly surpassed $60 billion. Meta also signed a $3 billion deal with Amsterdam-based Nebius, a company that reported massive year-over-year revenue growth thanks to a soaring demand for GPU-dense compute. This is in addition to the $14.2 billion agreement Meta already signed with CoreWeave, another major neocloud player that’s scaling at a breakneck pace.
Meanwhile, all of this is unfolding amid an unprecedented infrastructure arms race. In 2025 alone, tech giants like Amazon, Google, Microsoft and Meta are projected to spend up to $320 billion building new AI data centers — a level of construction that’s already straining America’s power and water systems. Meta alone plans to drop $600 billion on new facilities over the next three years, while the SoftBank-OpenAI-Microsoft-Oracle “Stargate” initiative is aiming to build a vast network of AI-first supercomputing sites throughout the country that could reach 10 gigawatts of capacity by decade’s end.
Taken together, all of these deals signal a broader shift in where companies believe they can reliably source the infrastructure needed for frontier AI development. After years of relying almost entirely on major cloud providers like AWS and Azure as the default, the industry’s top players are looking elsewhere, turning to neocloud providers that can deliver capacity faster, offer more flexible terms and keep costs down. Training cutting-edge models has become absurdly expensive, and traditional platforms can’t always keep up. Neoclouds are stepping in to fill that gap, quietly reshaping the map of AI infrastructure and the rules of the race.
What Is a Neocloud Provider?
Neocloud providers are cloud companies built almost exclusively for AI and high-performance computing workloads. Unlike legacy cloud providers, which juggle thousands of general-purpose software services, neoclouds are optimized for modern AI accelerators — the engines that turn vast data sets into intelligent models. Their data centers are packed with dense GPU clusters, designed for raw speed, memory bandwidth and efficient cooling systems rather than a broad mix of products and services.
The term “neocloud” has only recently emerged, as early GPU shortages amid the generative AI boom exposed the limits of traditional hyperscalers. Many got their start in crypto mining, then transitioned over to GPU-as-a-service, haphazardly creating an entirely new type of AI-focused cloud computing altogether. And their business models tend to focus on simpler service layers, more flexible procurement and a singular focus on AI compute, allowing them to offer more immediate access to cutting-edge hardware in ways hyperscalers can’t.
Today, there are more than 100 neocloud providers worldwide, but only about ten are gaining serious traction in the United States. Their presence is also expanding across Europe, the Middle East and Asia, with established brands like CoreWeave and Lambda edging out front, trailed by newcomers like Nebius, Fluidstack and TensorWave.
How Do Neocloud Providers Work?
Neocloud providers don’t try to do everything. They focus on what AI companies need most: fast, reliable access to large-scale compute. To do this, they build specialized data centers, source hardware strategically — like repurposing old crypto mining rigs or hunting down early-access chips — and build high-bandwidth, low-latency networks that keep large training jobs running efficiently. That narrow focus lets them move faster than big cloud players and support more customized AI model setups.
To secure capacity, many neoclouds rely on long-term contracts with AI developers and enterprises, giving them steady revenue while guaranteeing customers’ continual access to the tens of thousands of GPUs they might need months or years in advance. For companies running these huge models, that stability makes planning and scaling a lot less stressful.
Benefits of Neocloud Providers
AI companies are gravitating toward neocloud providers for a number of reasons:
Cost Efficiency
Without the overhead of large enterprise toolsets, neocloud vendors can offer GPU compute at more affordable prices than major hyperscalers. Some analyses suggest companies can save as much as two thirds by sourcing from neoclouds versus traditional clouds. This makes them especially appealing to startups and smaller AI developers who need flexible, cost-effective access in order to stay competitive. For example, one report shows how CoreWeave’s pricing strategy lets it undercut hyperscalers through 30 to 50 percent discounts tied to multi-year contracts, a tactic also used by rival Lambda.
Faster GPU Access
A tight hardware pipeline and AI-focused data center design mean developers can actually get chips when they need them. Teams can launch big multi-GPU jobs within minutes, instead of watching them sit in a queue for hours or days like they would on a typical cloud.
Flexibility
Neoclouds make it easier for companies to avoid getting boxed into a single cloud provider, something many teams tied to Azure or Google Cloud have struggled with. And that freedom pays off. Hugging Face, for instance, now spreads its experiments across different, third-party clouds rather than committing to one.
Specialized Infrastructure
Neoclouds are built for AI from the ground up, with fast, low-latency networks that keep data moving quickly between GPUs. That means heavy-duty tasks like training models or running simulations can get done faster and more smoothly than on a traditional cloud.
Challenges of Neocloud Providers
That being said, the neocloud trend comes with its share of uncertainty and risk.
Bubble Concerns
The sheer quantity of money going into these startups — based on yet-to-be-determined demand — has raised comparisons to the dot-com era’s overspending. These kinds of speculative bets can inflate valuations far beyond reality, making the market feel like it’s riding a bubble ready to pop.
Efficiency Could Cut Demand
As AI research increasingly focuses on smaller, more efficient models and techniques that reduce training compute, the long-term need for enormous GPU clusters could decline over time. Companies that over-invest in hardware now may risk having underused capacity further down the line.
Industry Fragmentation
With more than 100 neocloud providers, the market is already overcrowded. Many will likely consolidate or fail as competition intensifies and capital constraints tighten, potentially leaving some companies in the lurch.
Environmental and Infrastructure Strain
Neocloud’s AI workloads rely on massive data centers, which are already pushing the grid to its limits. In 2024, U.S. data centers consumed 183 terawatt-hours of electricity — about 4 percent of the nation’s total — and usage is projected to more than double by 2030. Cooling these facilities alone used an estimated 17 billion gallons of water last year.
How Neoclouds Fit into the Future of AI
The AI boom is driving an insatiable demand for computational power, sparking a fierce race to expand cloud infrastructure. Companies all over the world are scrambling to secure reliable capacity, and that competition is shaping an entirely new type of cloud ecosystem.
Neoclouds are answering that call alongside traditional hyperscalers, creating a hybrid approach where general enterprise workloads run on conventional clouds while GPU-heavy training and inference tap neocloud capacity. As frontier AI models require ever-larger compute budgets, this setup is quickly evolving from its current status as a convenience to an essential part of the AI development pipeline.
This shift, however, reflects broader market pressures. Expanding AI capacity is first and foremost a race to secure power guarantees. Running colossal clusters of energy-sucking GPUs requires an exorbitant amount of networking, storage and cooling, which has fueled an environment where rapid expansion often relies on significant debt and venture financing. Billions in funding and loans are being cycled between the same group of players to purchase chips and construct data centers, enabling fast growth while also concentrating financial risk.
Hyperscalers are responding to the rise of neoclouds by diversifying their suppliers and exploring alternative hardware options, while chipmakers and networking companies court neoclouds with new solutions. This push and pull between specialized, agile neoclouds and large, established hyperscalers is defining not just where AI runs, but how quickly models can be developed and deployed.
For now, the key takeaway is that neoclouds don’t replace traditional clouds — they complement them. They offer speed, reliability and infrastructure that hyperscalers can’t always match. Organizations that thoughtfully integrate neoclouds into their broader cloud strategy can gain a competitive edge in a fast-moving, ever-shifting landscape.
Frequently Asked Questions
Who are the top neocloud providers?
The two frontrunners in the neocloud market at the moment are CoreWeave and Lambda Labs, followed by smaller startups Nebius, Fluidstack and TensorWave.
