Microsoft alone has committed more than $60 billion to neocloud partnerships. Meta has signed deals worth up to $62 billion combined across just two providers. This is no longer a scrappy alternative to the hyperscalers, it is where a meaningful share of frontier AI training now actually runs.
Microsoft alone has committed more than $60 billion to neocloud partnerships, including a reported $23 billion to the British startup Nscale for 200,000 next-generation GPUs. Meta has signed agreements worth up to $62 billion combined across CoreWeave and Nebius. Neocloud revenue is projected to reach roughly $20 billion in 2026 and approach $180 billion by 2030. This is no longer a scrappy alternative to the hyperscalers, it is where a meaningful share of frontier AI training now actually runs, because hyperscalers cannot build data centers fast enough to meet demand.
But the category is crowded, uneven, and full of brands with wildly different levels of substance behind them. Some own their fleets outright. Some are reselling capacity. Some are backed by a nonprofit endowment, others by NVIDIA itself, others by GPU-backed debt. This ranking cuts through the noise with a consistent methodology and profiles the ten providers that matter most heading into 2027.
Providers were evaluated and ordered on five weighted factors: fleet scale and GPU ownership, capital backing and financial durability, hyperscaler and frontier-lab partnerships as a proxy for real demand, pricing transparency, and genuine differentiation rather than brand repositioning. Rankings reflect public data available as of mid-2026 and will shift as the market consolidates.
Here's how they stack up, at a glance, before the full profiles below.
| # | Provider | Known For | Best Fit |
|---|---|---|---|
| 1 | CoreWeave | Largest fleet, IPO'd | Enterprise-scale reserved training |
| 2 | Nebius | Sovereign EU cloud | European data residency needs |
| 3 | Nscale | Fastest-growing valuation | Hyperscaler-scale European deals |
| 4 | Crusoe | Stranded/renewable energy | Sustainability-conscious buyers |
| 5 | Lambda Labs | Developer experience | Fast self-serve cluster launches |
| 6 | FluidStack | Custom AI data centers | Frontier labs needing bespoke builds |
| 7 | Voltage Park | Nonprofit-owned, transparent | On-demand bursts, budget teams |
| 8 | Together AI | Lowest on-demand pricing | Cost-sensitive inference workloads |
| 9 | Massed Compute | Owns fleet, supplies others | Buyers who want the carrier layer |
| 10 | Vultr | Broadest GPU SKU range | Small teams, simple self-serve |
CoreWeave remains the largest independent neocloud by nearly every measure. The company completed its IPO, has raised more than $7 billion in total funding with NVIDIA among its early investors, and is developing a $6 billion data center campus in Pennsylvania alongside a $3.4 billion expansion in the UK. Its Meta agreement, expanded to $21 billion as of April 2026, sits alongside major contracts with OpenAI and NVIDIA, underscoring a scale gap with the rest of the field, analysts tracking the sector place CoreWeave alone in the top tier.
Best for: enterprises and labs that need large, reserved GPU capacity with full InfiniBand fabric and are comfortable with a hyperscaler-style contract relationship rather than fast self-serve provisioning.
Amsterdam-based Nebius has built its position around European data sovereignty, partnering with Microsoft to deliver AI infrastructure that keeps data and compute inside the EU. Nebius holds Meta commitments reported as high as $27 billion and has scaled its GPU fleet aggressively, with a mid-term target around 240,000 GPUs. It sits in the same upper tier as major hyperscaler alternatives in independent provider assessments.
Best for: European enterprises and any organization with data residency or sovereignty requirements that rule out US-only infrastructure.
Nscale went from a crypto-mining spinout in 2024 to a $14.6 billion valuation by its March 2026 Series C, backed by NVIDIA, Dell, Citadel, and Jane Street among others, with Sheryl Sandberg, Nick Clegg, and Susan Decker joining its board. Microsoft has committed roughly $23 billion for 200,000 next-generation GPUs, and Nscale's Stargate Norway joint venture with Aker and OpenAI targets 100,000 GPUs and 230 megawatts of capacity by the end of 2026. The company has also taken on significant GPU-backed debt to fund its build-out, a financing pattern common across fast-scaling neoclouds and worth understanding before signing a multi-year contract.
Best for: hyperscaler-scale European deployments, particularly where green power sourcing (Nscale leans heavily on Nordic hydro) is a procurement priority. Less suited to a team wanting to rent a handful of GPUs for a weekend.
Crusoe built its business converting stranded and wasted energy, flare gas and renewables, into data center power, and has carried that model into AI infrastructure with 4.5 gigawatts of natural gas contracts announced to fuel its GPU fleet. Its facilities use direct liquid-to-chip cooling and an API-driven platform aimed at production AI workloads rather than experimentation.
Best for: organizations with sustainability mandates or ESG reporting requirements who still need frontier-class GPU access without defaulting to a hyperscaler.
Lambda differentiates on simplicity: one-click multi-GPU cluster deployment, transparent hourly pricing well below AWS rates for comparable hardware, and NVIDIA as a direct investor, which helps ensure early access to new GPU architectures. It's less focused on hyperscaler-scale reserved contracts than on making it fast for a team to get GPUs working today.
Best for: startups, research teams, and any buyer who values fast provisioning and developer experience over the largest possible reserved fleet.
FluidStack made headlines with a reported $50 billion partnership with Anthropic to build custom AI data centers, a scale of commitment that puts it in the same conversation as the largest names in the category despite a lower public profile than CoreWeave or Nebius. Its model centers on building bespoke infrastructure for frontier labs rather than competing purely on self-serve rental.
Best for: frontier AI labs and large enterprises that need a dedicated, purpose-built infrastructure partner rather than a shared multi-tenant fleet.
Voltage Park is funded by the Navigation Fund, a nonprofit endowment created by Jed McCaleb, co-founder of Stellar and Ripple, with profits returning to that endowment rather than outside venture investors. It rents H100 and Blackwell clusters from $1.99 an hour on-demand or via dedicated reserve, with transparent hourly pricing similar in spirit to Lambda's model. A January 2026 merger with Lightning AI folded the two companies together, so buyers should factor Lightning's software roadmap into any long-term evaluation.
Best for: teams that want transparent on-demand pricing and are comfortable with a smaller fleet than the biggest neoclouds, in exchange for a mission-driven ownership structure.
Together AI consistently ranks among the lowest-cost options for on-demand GPU access per H100-hour, alongside spot marketplaces like Vast.ai. Its platform leans toward inference and fine-tuning workloads specifically, rather than positioning itself as a full hyperscaler-style reserved capacity provider.
Best for: cost-sensitive inference and fine-tuning workloads where on-demand pricing matters more than guaranteed large-scale reserved capacity.
Massed Compute occupies a distinct niche: it owns its GPU fleet outright and supplies capacity to other neoclouds that resell it under their own branding, a role founder and CTO Sunny Smith has described on the DataStorage.com Podcast as sitting underneath the rack-and-stack and layer-7 support tiers most buyers interact with directly. For buyers who care about the ownership question at the center of neocloud vetting, going closer to the carrier layer can mean fewer hops between contract and physical hardware.
Best for: buyers who already understand the importance of vetting neocloud ownership and specifically want to work closer to the source rather than through a reseller.
Vultr rounds out the list as the most budget-accessible and broadly self-serve of the ten, with per-second billing and one of the widest ranges of GPU SKUs available, including older and consumer-grade cards alongside data center hardware. It won't compete with CoreWeave or Nscale on frontier-scale reserved deployments, but for smaller teams and simpler workloads it remains one of the most approachable entry points into GPU cloud infrastructure.
Best for: small teams, students, and workloads that don't require the largest or newest GPU fleet, just fast, simple, affordable access.
Two things correlate most strongly with a provider's position here: whether it owns the hardware it rents out, and whether it has landed at least one deal with a hyperscaler or frontier lab large enough to validate its infrastructure at scale. CoreWeave, Nebius, and Nscale all check both boxes. Providers further down the list tend to win on a single sharp differentiator, price, sustainability, ownership model, developer experience, rather than competing on scale. This ties directly back to the ownership question raised in our coverage of CoreWeave's revenue miss and what it signaled for AI infrastructure.
That's not a knock on the smaller players. A buyer running production training at petabyte and gigawatt scale needs what CoreWeave or Nscale offer. A startup burning through a seed round needs what Lambda or Vultr offer. Matching the provider to the actual workload, rather than defaulting to whichever name is most visible, is the entire point of a ranking like this one.
The neocloud with the biggest funding round is not automatically the right neocloud for your workload. Match the provider to the job, not the headline.
Compare AWS, Google Cloud, Azure, and alternatives like Backblaze B2 Discover how much you could save in seconds
