Neoclouds Explained: How CoreWeave, Nebius, and Nscale Are Challenging the Hyperscalers

Direct answer: "Neoclouds" are GPU-specialized cloud providers — CoreWeave, Nebius, Lambda, Crusoe, and Groq are the most prominent — that build data centers purpose-built for AI training and inference rather than offering the full general-purpose cloud catalog of a traditional hyperscaler. They posted explosive growth in Q2 2026, with Nebius revenue up 454% year-over-year to $582.3 million and CoreWeave up 112% to $2.575 billion, while racing to convert contracted gigawatts of power into active, billable capacity. This matters now because neoclouds have become a real, fast-growing alternative to AWS, Azure, and Google Cloud for AI compute specifically, and because their financing structures, pricing, and customer concentration are exactly the kind of details a business evaluating AI infrastructure options needs to actually understand rather than treat as interchangeable.

What a Neocloud Actually Is (and Isn't)

The term "neocloud" describes a category of cloud infrastructure provider that emerged specifically to serve the AI compute boom, and it's worth being precise about what separates it from a traditional hyperscaler, because the distinction explains almost everything else in this piece. AWS, Microsoft Azure, and Google Cloud built their businesses over roughly two decades around a broad, general-purpose catalog: compute, storage, databases, networking, managed services, and a long tail of enterprise software integrations, serving every kind of workload from a small business website to a global bank's core systems. GPU capacity for AI is one product line among hundreds inside that catalog.

Neoclouds inverted that model. Rather than building a broad platform and adding GPU capacity as one more service, companies like CoreWeave, Nebius, Lambda, Crusoe, and Groq built their entire business around one thing: securing power, land, and Nvidia (or in some cases AMD or custom) chips as fast as possible, and renting that compute out to AI labs and enterprises that need it. That narrow focus lets a neocloud move faster on the specific bottlenecks that matter most in the current market — signing power purchase agreements, standing up data centers, and getting the latest GPU generation racked and billable — without the organizational overhead of maintaining a sprawling general-purpose service catalog alongside it.

This specialization comes with real tradeoffs in both directions. A neocloud typically can't match a hyperscaler's breadth of managed services, global compliance certifications, or decades-deep enterprise sales relationships — if a customer needs a GPU cluster plus a mature managed database service plus enterprise support in a dozen regulatory jurisdictions, a hyperscaler remains the more complete answer. But for a customer whose primary need is large-scale GPU capacity, provisioned quickly, at a competitive price, a neocloud is frequently faster to deploy with and, on a pure compute-hour basis, can be meaningfully cheaper — which is precisely the value proposition that's fueled their growth.

It's also worth naming plainly what a neocloud is not: it is not simply a reseller or a thin layer on top of a hyperscaler's infrastructure. The five companies covered in this piece own or lease their own data center facilities, negotiate their own power agreements, and in most cases purchase GPUs directly in bulk from Nvidia or AMD — they are infrastructure operators in their own right, not intermediaries, which is also why their capital structures, debt loads, and financing arrangements are complex and consequential enough to draw serious analyst and investor scrutiny, discussed later in this piece.

How an Entire New Category of Cloud Provider Emerged This Fast

It's worth pausing on how unusual it is for an entirely new category of infrastructure company to emerge, scale to multi-billion-dollar revenue, and in two cases reach public markets, within just a few years. Cloud computing itself took the better part of a decade to move from a novel idea to a trusted enterprise default. Neoclouds compressed a comparable rise into a fraction of that time, and the reason traces directly back to the specific, acute bottleneck that created the opportunity in the first place: demand for GPU capacity to train and run AI models outpaced what the existing hyperscalers could provision, at a scale and speed that opened room for entirely new entrants to compete for a meaningful share of a rapidly growing market rather than having to displace incumbents from an already-saturated one.

Traditional hyperscalers, for all their scale, weren't originally built around the assumption that a single category of specialized hardware would suddenly represent the majority of incremental infrastructure demand. Reallocating that much of a general-purpose cloud business toward one hardware category, that fast, meant competing internally against every other product line's capital budget, existing data center design standards, and long procurement cycles. Neoclouds didn't carry any of that institutional weight — a new company built from day one around GPU capacity could make capital allocation, facility design, and power procurement decisions entirely in service of that one goal, without having to justify the tradeoff against a legacy business.

That structural advantage is also precisely why neoclouds have been able to move so aggressively on the specific constraints that matter most right now: securing power agreements fast enough to keep pace with GPU delivery schedules, and negotiating direct relationships with chip suppliers rather than working through years of established hyperscaler procurement relationships. It's a genuinely different starting position than a hyperscaler adding a new product line, and it's the structural reason this category could grow as fast as the revenue figures described below actually show.

None of this makes neoclouds inherently a better long-term bet than hyperscalers — it explains why they exist and why they've grown quickly, not whether that growth is durable. That durability question is exactly what the rest of this piece, and the broader investor attention described later on, is actually about.

The Growth Numbers Behind the Hype

The headline growth figures from Q2 2026 explain why neoclouds have gone from a niche industry term to a mainstream investment story. Nebius reported revenue of $582.3 million for the quarter, up 454% year-over-year — an extraordinary growth rate by the standards of almost any established industry, let alone one already operating at hundreds of millions of dollars in quarterly revenue. Nebius's AI cloud annual recurring revenue reached $3.0 billion, giving a sense of the run-rate scale the business has reached. CoreWeave, already larger in absolute terms, grew 112% year-over-year to $2.575 billion in quarterly revenue — a growth rate that would be remarkable on its own, and looks even more so layered on top of an already much larger base than Nebius's.

Behind those revenue figures sits an equally important, less flattering number: both companies continue to report growing losses even as revenue accelerates, because they're spending aggressively to build out data center capacity ahead of confirmed, billable demand. CoreWeave's 2026 capital expenditure guidance sits at $35-39 billion, funded in part by $31 billion in recourse debt — a debt load that means CoreWeave itself, not just a project-level financing vehicle, is on the hook if that capacity doesn't generate the returns being underwritten. This is the core tension running through the entire neocloud sector right now: revenue growth rates that look almost unbelievable next to capital spending and debt levels that are, in absolute terms, enormous even for a well-capitalized infrastructure company.

CoreWeave's power numbers illustrate the second half of the growth story — the race between contracted capacity and active, revenue-generating capacity. The company reported 1.5 gigawatts of active power against more than 4.2 gigawatts contracted, meaning a substantial majority of its committed future capacity isn't yet built out and billing customers. That gap is normal for a company scaling this quickly, and it's also exactly where the execution risk lives: contracted power is a promise about the future, active power is revenue today, and the time and capital required to convert one into the other is the single biggest variable in whether a neocloud's growth story holds up over the next several years.

CoreWeave's customer base reinforces why this matters: Meta, Anthropic, and OpenAI are among its named customers, and the company carries a contracted backlog of roughly $104 billion — a backlog large enough to represent years of future revenue, assuming CoreWeave can build the capacity to deliver against it and assuming those customers' own AI spending plans hold up as currently projected. Microsoft's commitments to Nebius, separately, total $17.4-19.4 billion including options, a similarly outsized commitment relative to Nebius's current revenue base, and a strong vote of confidence from one of the largest AI infrastructure buyers in the world.

Ranking the Five Major Neoclouds on Price and Power

Rankings published in 2026, including analysis from MarkTechPost and Cryptopond evaluating the leading GPU neoclouds by published pricing and contracted power, give a useful, concrete comparison across the five most prominent players: CoreWeave, Nebius, Lambda, Crusoe, and Groq.

CoreWeave holds the distinction of being the only GPU cloud provider to earn a Platinum rating in SemiAnalysis's ClusterMAX 2.0 ranking — a system that evaluates GPU cloud providers on operational reliability, performance consistency, and infrastructure quality rather than just headline specs. That Platinum rating comes alongside premium pricing: CoreWeave lists H100 GPU-hour pricing around $6.16, notably higher than the other four providers compared here. The combination suggests CoreWeave is positioning itself at the premium, high-reliability end of the market — a reasonable strategy for a provider whose largest customers are running frontier-scale training workloads where infrastructure reliability failures are extremely costly, and where paying a premium for a Platinum-rated environment is a rational trade against that risk.

Nebius sits at a notably different point on the pricing spectrum, listing H100 GPU-hour pricing around $3.85 — among the most competitive of the group — while also publishing an H100 preemptible (spot) rate of $2.15 per hour, the lowest published spot rate among the five providers compared. Nebius has also distinguished itself as the only neocloud among these five publishing on-demand pricing for Nvidia's B300 GPU, at $7.85 per hour — a notable transparency move in a market where next-generation GPU pricing is often negotiated privately rather than published, and one that gives budget-conscious buyers a genuine, comparable reference point for planning around Nvidia's newest chip generation.

Lambda lists H100 pricing around $3.99 per hour, close to Nebius's on-demand rate, and has built a reputation particularly among AI researchers and smaller technical teams for straightforward, developer-friendly access to GPU capacity. Lambda operates 15 data centers across the US and has raised more than $1.5 billion in a Series E round, alongside a $1 billion senior secured credit facility, and is reportedly targeting an IPO in the second half of 2026 — a milestone that would make it the third of these five companies, alongside CoreWeave and Nebius, to become a public, SEC-reporting entity.

Crusoe prices H100 capacity around $3.90 per hour and stands out on this list as the only major provider among the five listing AMD's MI300X and MI355X GPUs alongside the more commonly offered Nvidia chips — a meaningful differentiator for customers specifically looking to diversify away from total dependence on Nvidia hardware, whether for cost, availability, or architectural reasons. Crusoe operates campuses in Texas, Missouri, and Wyoming, reported 4.9 gigawatts of contracted power as of June 2026, and has disclosed a pipeline exceeding 40 gigawatts of potential future capacity — a pipeline figure large enough to represent a multi-year growth runway well beyond what's currently built or even currently contracted.

Groq takes a different technical approach entirely, built around its own custom chip architecture rather than reselling Nvidia or AMD GPUs, which positions it distinctly for specific inference-oriented workloads rather than as a direct like-for-like price comparison against the GPU-hour figures quoted for the other four. Groq's growth in 2026 has been notably fast on the funding side: the company raised $650 million and then a further $350 million Series A, reaching a $3.5 billion valuation, within roughly two months in the middle of the year, while scaling to 13 data centers and 54 megawatts of capacity, with a stated target of exceeding 200 megawatts by 2027.

Taken together, this pricing and capacity picture tells a coherent story: there is no single "best" neocloud in the abstract, only better and worse fits for a specific workload, budget, and risk tolerance — a theme the MarkTechPost/Cryptopost ranking makes explicit in its own use-case recommendations, covered further in the Q&A section below.

One caveat worth stating plainly: published GPU-hour rates are a genuinely useful starting comparison, but they're list prices, not necessarily the rate any specific customer ends up paying. Large, multi-year commitments typically unlock negotiated discounts well below the published on-demand figure, and availability at the published rate can shift quickly during periods of high demand for a specific GPU generation. Treat the numbers above as a reliable guide to relative positioning across providers — who's priced at a premium, who's competing on cost — rather than as a precise quote for any particular deal.

Nscale's Rapid Rise: Europe's Neocloud Contender

While CoreWeave and Nebius dominate the US-centric neocloud conversation, the fastest and arguably most unusual growth story in this category in 2026 belongs to Nscale, described in Futurum's coverage as "Europe's largest homegrown neocloud." Nscale's origin story is genuinely unusual for a company now valued in the tens of billions of dollars: it began as a 2024 spinout of Arkon Energy, a crypto-mining firm — meaning its founding team's core early expertise was in exactly the kind of power procurement, data center operations, and specialized hardware management that turned out to translate directly into building GPU infrastructure for AI, once that opportunity emerged at scale.

That translation happened remarkably fast. Nscale raised a $2 billion Series C in March 2026, reaching a $14.6 billion valuation — a figure that places it firmly among the most valuable neocloud companies globally, despite being, by a wide margin, the youngest company on this list. The scale of capital backing this valuation is matched by the scale of its infrastructure commitments: Nscale contracted roughly 200,000 Nvidia GB300 GPUs in a deal with Microsoft spanning facilities across Texas, Portugal, the UK, and Norway — a genuinely international footprint for a company that's barely two years removed from its crypto-mining origins.

Nscale has also moved to expand its capabilities beyond raw infrastructure through acquisition, agreeing in 2026 to acquire Anyscale, the commercial company behind the widely used open-source Ray framework for distributed computing. Futurum's coverage of the deal frames it as part of a broader "neocloud land grab," and the strategic logic is straightforward: owning Ray's commercial layer gives Nscale a software and developer-tooling foothold that pure infrastructure providers typically lack, potentially making it easier for customers to actually build and scale distributed AI workloads on Nscale's infrastructure rather than just renting raw GPU capacity and building all of that tooling themselves.

Nscale's most geographically distinctive project is "Stargate Norway," a joint venture with Norwegian industrial company Aker and OpenAI targeting 100,000 GPUs and 230 megawatts of capacity by the end of 2026. The project reflects a broader pattern in Nordic AI infrastructure investment — Norway and its neighbors offer some combination of cold climates that reduce cooling costs, established renewable and hydroelectric power infrastructure, and political stability, all attractive traits for power-hungry, long-horizon data center investments. Together, Nscale's Microsoft deal and its Stargate Norway venture make it one of the most geographically diversified neocloud operators of the group, spanning North America and multiple distinct European regions in a way none of the more US-concentrated competitors on this list currently match.

Why Investors Are Paying Attention Now

The neocloud sector's growth numbers, described above, have made it an increasingly visible investment story in its own right, distinct from the broader "AI stocks" narrative that's dominated markets through the current AI infrastructure buildout. CoreWeave and Nebius are, notably, the only two neoclouds among the five covered here that are public, SEC-reporting companies — meaning they're the only two with the kind of disclosed, audited financials that let outside investors and analysts actually evaluate the numbers cited throughout this piece with reasonable confidence, rather than relying on company-reported figures without independent verification.

That transparency, combined with the sector's growth rates, led to a notable market development in August 2026: a new exchange-traded fund launched specifically to give investors diversified exposure across leading neocloud stocks, covered by 24/7 Wall St. under the framing that "CoreWeave & Nebius are soaring." The launch of a dedicated ETF is a meaningful signal in its own right — it typically reflects a fund provider's judgment that investor demand for exposure to a specific sector has reached a scale that justifies building and marketing a dedicated product around it, rather than leaving that exposure to be assembled manually stock-by-stock.

At the same time, the neocloud sector's financing structure connects directly to a broader concern covered in depth elsewhere: circular financing and bubble risk in AI infrastructure deals. Analysis from io-fund.com specifically examining Nvidia's relationships with CoreWeave and Nebius frames these neoclouds as participants in the same kind of circular financing pattern seen across the broader AI infrastructure market — Nvidia has financial ties to some of the same companies that are among its largest GPU customers, which raises the same fundamental question raised elsewhere about circular AI financing more broadly: how much of a neocloud's reported growth and contracted demand reflects independent, external demand for AI compute, versus demand connected to the same small, interconnected set of companies financing and buying from each other. This doesn't invalidate the very real growth these companies are reporting — CoreWeave's and Nebius's revenue figures are audited, disclosed numbers, not projections — but it's a genuinely relevant lens for understanding both the opportunity and the risk in this sector, not a separate, unrelated story.

Investors and business partners evaluating this sector are, in effect, being asked to hold two things true at once: the underlying growth numbers are real and independently verifiable for the two public companies in the group, and a meaningful share of the demand generating those numbers flows through a small, tightly interconnected set of counterparties whose own spending plans could shift together rather than independently. Neither fact cancels the other out — they simply both need to be part of the same evaluation.

The Global Picture

United States. The US is the primary operating base for four of the five major neoclouds covered here. CoreWeave operates 51 data centers with 1.5 gigawatts active and more than 4.2 gigawatts contracted. Lambda runs 15 US data centers and is targeting a US IPO in the second half of 2026. Crusoe's campuses in Texas, Missouri, and Wyoming account for its 4.9 gigawatts of contracted power and 40-plus gigawatt pipeline. Groq operates 13 data centers totaling 54 megawatts, targeting more than 200 megawatts by 2027.

United Kingdom. Nscale, described as Europe's largest homegrown neocloud, was founded in the UK in 2024 as a spinout of crypto-mining firm Arkon Energy. It reached a $14.6 billion valuation on a $2 billion Series C in March 2026, and its roughly 200,000-GPU Microsoft deal spans facilities including a UK site alongside Texas, Portugal, and Norway. The UK is also, by extension, the home base for Nscale's acquisition of Anyscale.

UAE / Dubai. No distinct regional-specific reporting on the neocloud GPU-rental model covered in this piece was found for the UAE. The UAE's most prominent compute buildout coverage centers on the separate Stargate UAE and G42 sovereign AI infrastructure project, which operates on a different model — sovereign, government-linked investment in national AI infrastructure — rather than the commercial, rentable GPU-cloud model that defines the neocloud category described throughout this piece, and it would be inaccurate to conflate the two.

Australia. No distinct regional-specific reporting was found. None of the five major neoclouds covered here were reported to have an Australian data center presence in the sources reviewed for this piece.

Germany. No distinct regional-specific reporting was found for Germany specifically, despite Germany's significant broader technology and industrial sectors.

Europe / France. Nebius operates locations in France alongside Finland, Estonia, Iceland, and Israel — giving it one of the more geographically distributed European footprints among the neoclouds covered here. Separately, Nscale's Stargate Norway joint venture with Aker and OpenAI, targeting 100,000 GPUs and 230 megawatts of capacity by the end of 2026, represents a significant Nordic milestone rather than a France-specific one, and is described here accordingly rather than folded into a broader "Europe" claim it doesn't specifically support.

China. No distinct regional-specific reporting was found. The neocloud model, as covered across the sources reviewed for this piece, is presented as a predominantly US and European phenomenon, and no equivalent Chinese domestic GPU-cloud competitor operating on a comparable commercial rental model was surfaced in this research pass — worth noting as a genuine research gap rather than evidence that no such activity exists in China.

Choosing Between Neoclouds, Hyperscalers, and Building In-House

For a business actually deciding how to source AI compute — rather than just following the investment story — the practical question isn't "which neocloud is best," it's "which sourcing model fits this specific workload, timeline, and risk tolerance." A few grounded distinctions help frame that decision.

Workload type matters more than brand recognition. Frontier-scale model training, high-throughput production inference, and short-term budget-conscious experimentation each favor different providers for different reasons, as the pricing and reliability comparisons earlier in this piece illustrate — a Platinum-rated, premium-priced provider makes sense for a workload where downtime is extremely costly, while a lower-cost, lower-frills option is entirely rational for exploratory work where occasional interruptions are an acceptable tradeoff for meaningfully lower cost.

Contract terms deserve as much scrutiny as pricing. Given how much of this sector's capacity is still contracted rather than active — CoreWeave's own 1.5 gigawatts active against more than 4.2 gigawatts contracted is a clear example — any business signing a multi-year capacity commitment with a neocloud should understand exactly what's guaranteed, on what timeline, and what recourse exists if a provider's own buildout falls behind schedule.

Total cost of ownership extends beyond the advertised GPU-hour rate. Data egress fees, minimum commitment terms, support tiers, and the operational overhead of managing infrastructure directly (versus a fully managed service) all affect the real cost of a neocloud relationship in ways a simple per-hour price comparison, like the one covered earlier in this piece, doesn't fully capture.

Geographic footprint and data residency requirements can rule out options before pricing even enters the conversation. A business with strict data residency requirements — a European customer base needing data to stay within the EU, for instance — needs to check which specific facilities, not just which company, a given neocloud can actually route a workload through. Nebius's presence across France, Finland, Estonia, Iceland, and Israel, or Nscale's footprint spanning Texas, Portugal, the UK, and Norway, means the honest answer to "can this provider meet our residency requirement" varies by facility, not just by brand, and is worth confirming directly rather than assuming from a company's general geographic description.

For most businesses building AI-powered products, the more relevant decision usually isn't which infrastructure provider to use directly at all — it's whether to build and manage that infrastructure relationship in-house, or to work with a technical partner who already has that evaluation experience and can architect the application layer to remain portable across providers as pricing and availability shift. Our AI agents and automation team and our custom software development practice both routinely make exactly this kind of infrastructure decision as part of scoping a build, and our comparisons hub covers how we think through vendor and platform tradeoffs more broadly — the same evaluation discipline that applies to choosing a neocloud applies just as directly to choosing a database, a framework, or any other foundational technical dependency.

Straight Answers on Neoclouds and the GPU Cloud Race

What is a "neocloud," and how does it differ from a traditional hyperscaler cloud provider?

A neocloud is a cloud infrastructure provider built specifically around GPU capacity for AI workloads, rather than the broad, general-purpose service catalog a traditional hyperscaler like AWS, Azure, or Google Cloud offers. Where a hyperscaler treats GPU cloud as one product line among hundreds, a neocloud's entire business — its capital raising, its power procurement, its data center construction, its customer relationships — is organized around securing and renting out AI compute as fast as possible. That focus lets neoclouds often move faster and price more competitively on pure GPU capacity, while typically offering a narrower range of managed services, compliance certifications, and enterprise integrations than an established hyperscaler provides. Neither model is strictly better; they suit different combinations of workload type, budget, and organizational need.

How much did CoreWeave's revenue grow in Q2 2026, and how much power does it have active versus contracted?

CoreWeave's revenue grew 112% year-over-year in Q2 2026, reaching $2.575 billion for the quarter. On the infrastructure side, the company reported 1.5 gigawatts of active power against more than 4.2 gigawatts of contracted power — meaning the majority of its committed future capacity is not yet built and billing customers. That gap between contracted and active power is the central execution risk for CoreWeave and every other neocloud racing to scale: converting a signed commitment into a running, revenue-generating data center takes real time and capital, and the pace of that conversion is what will ultimately determine whether today's growth rates are sustainable.

How much did Nebius's revenue grow in Q2 2026, and what is its AI cloud annual recurring revenue?

Nebius's revenue grew 454% year-over-year in Q2 2026, reaching $582.3 million for the quarter — one of the fastest growth rates reported anywhere in the neocloud sector. Its AI cloud annual recurring revenue reached $3.0 billion, giving a clearer picture of the run-rate scale behind that quarterly figure. Growth at this rate, sustained over multiple quarters, is a major reason Nebius has drawn significant investor attention and a substantial financial commitment from Microsoft, discussed elsewhere in this piece, and it's part of why Nebius is one of only two neoclouds among the major players — alongside CoreWeave — that are public, SEC-reporting companies with fully audited numbers behind these figures.

Why is CoreWeave the only "Platinum-rated" GPU cloud in SemiAnalysis's ClusterMAX 2.0 ranking?

SemiAnalysis's ClusterMAX 2.0 ranking evaluates GPU cloud providers on operational reliability, performance consistency, and infrastructure quality, not just headline pricing or specs, and CoreWeave is the only provider among the major neoclouds to earn its top Platinum rating. That distinction aligns with CoreWeave's positioning and pricing — it lists H100 GPU-hour pricing around $6.16, notably higher than competitors like Nebius, Lambda, and Crusoe — suggesting CoreWeave has deliberately built toward the premium, high-reliability end of the market rather than competing primarily on price. That positioning fits its customer base of frontier AI labs like Meta, Anthropic, and OpenAI, where infrastructure reliability failures during large-scale training runs are extremely costly, making a reliability premium a rational trade for those specific customers.

How do H100 GPU-hour prices compare across CoreWeave, Nebius, Lambda, and Crusoe in 2026?

Published 2026 pricing puts CoreWeave at roughly $6.16 per H100 GPU-hour, Nebius at roughly $3.85, Lambda at roughly $3.99, and Crusoe at roughly $3.90 — meaning CoreWeave prices meaningfully above the other three, which cluster relatively close together in the high-$3 range. That spread reflects real positioning differences rather than a simple efficiency gap: CoreWeave's premium aligns with its Platinum ClusterMAX rating and enterprise-grade reliability focus, while Nebius, Lambda, and Crusoe compete more directly on price for customers whose workloads don't require, or can't justify paying for, that reliability premium. Buyers comparing these numbers should also account for factors the headline rate doesn't capture, like minimum commitment terms, support tiers, and data egress costs.

Which neocloud offers the lowest published spot/preemptible GPU pricing in 2026?

Nebius publishes the lowest spot rate among the major providers, offering H100 preemptible capacity at $2.15 per hour. Preemptible or spot pricing offers a substantial discount versus on-demand rates in exchange for accepting that the provider can reclaim that capacity with limited notice, typically for workloads that can tolerate interruption — batch processing, non-time-critical training runs, or experimentation. At $2.15 per hour, Nebius's spot rate represents one of the most cost-effective ways to access H100-class GPU capacity currently published anywhere in the neocloud sector, making it a natural fit for budget-conscious teams whose workloads can be architected around the possibility of interruption.

Why is Nebius the only neocloud publishing on-demand pricing for Nvidia's B300 GPU?

Nebius publishing on-demand B300 pricing at $7.85 per hour is notable because next-generation GPU pricing is often negotiated privately between providers and large customers rather than published openly, particularly soon after a new chip generation becomes available. Nebius's decision to publish this rate is a meaningful transparency signal relative to its competitors, giving smaller buyers and researchers a genuine, comparable reference point for budgeting around Nvidia's newest chip generation rather than having to negotiate blind or rely on a sales conversation to learn approximate pricing. Whether other neoclouds follow with their own published B300 rates is a reasonable thing to watch as that GPU generation becomes more broadly available across the sector.

Why is Crusoe the only major neocloud listing AMD MI300X and MI355X GPUs?

Crusoe's decision to list AMD's MI300X and MI355X GPUs alongside the Nvidia hardware more commonly offered across the sector reflects a deliberate diversification strategy, distinguishing it from CoreWeave, Nebius, Lambda, and Groq, all of which center their offerings on Nvidia (or, in Groq's case, entirely custom silicon). For customers concerned about Nvidia supply constraints, pricing power, or simply wanting architectural flexibility across chip vendors, Crusoe's AMD offering is a genuine differentiator. It also reflects Crusoe's broader infrastructure strategy — with 4.9 gigawatts of contracted power as of June 2026 and a pipeline exceeding 40 gigawatts, the company has clearly built enough scale to support offering meaningful capacity across multiple chip architectures rather than betting entirely on one vendor's supply and roadmap.

What is Nscale, and how did it grow from a 2024 crypto-mining spinout to a $14.6 billion valuation?

Nscale is a UK-founded neocloud, described in Futurum's coverage as Europe's largest homegrown player in the category, that began in 2024 as a spinout of Arkon Energy, a crypto-mining firm. That origin turned out to be a meaningful advantage rather than a liability: the operational skills required to run large-scale, power-intensive crypto-mining infrastructure — securing power agreements, managing specialized hardware at scale, running data centers efficiently — translated directly into the requirements for building AI-focused GPU infrastructure once that opportunity emerged. Nscale raised a $2 billion Series C in March 2026, reaching a $14.6 billion valuation in roughly two years from its founding — an unusually fast rise even by the standards of the current AI infrastructure boom, reflecting both strong execution and the sheer scale of capital currently available to credible AI infrastructure plays.

What GPU deal did Nscale strike with Microsoft, and across which countries does it span?

Nscale contracted roughly 200,000 Nvidia GB300 GPUs in a deal with Microsoft, spanning data center facilities across Texas, Portugal, the UK, and Norway. A deal of this scale and geographic spread is significant on two fronts: it represents a substantial vote of confidence from Microsoft, one of the largest AI infrastructure buyers globally, and it establishes Nscale as one of the more geographically diversified neocloud operators covered in this piece, with a genuine multi-country footprint spanning North America and several distinct European markets rather than concentration in a single region.

Why is Nscale acquiring Anyscale, the company behind the Ray open-source framework?

Nscale's move to acquire Anyscale, covered by Futurum as part of a broader "neocloud land grab," reflects a strategic push beyond pure infrastructure into the software layer that sits on top of it. Ray is a widely used open-source framework for distributed computing, and Anyscale is the commercial company built around it — owning that layer gives Nscale a developer-tooling foothold that pure infrastructure providers typically lack. Practically, this could make it meaningfully easier for Nscale's customers to actually build, scale, and manage distributed AI workloads directly on Nscale's infrastructure, rather than renting raw GPU capacity and having to independently assemble and maintain their own distributed-computing tooling on top of it.

What is Nscale's "Stargate Norway" joint venture, and what GPU/power targets does it have for 2026?

Stargate Norway is a joint venture between Nscale, Norwegian industrial company Aker, and OpenAI, targeting 100,000 GPUs and 230 megawatts of capacity by the end of 2026. The project fits a broader pattern of AI infrastructure investment gravitating toward Nordic locations, which offer a combination of naturally cold climates that reduce data center cooling costs, well-established renewable and hydroelectric power infrastructure, and political and regulatory stability — all attractive traits for the kind of long-horizon, capital-intensive investment a data center represents. Alongside its Microsoft deal, Stargate Norway reinforces Nscale's position as one of the most geographically ambitious neocloud operators currently scaling in Europe.

How large is Lambda's committed capital as of 2026, and is it planning to go public?

Lambda has raised more than $1.5 billion in a Series E funding round and secured a $1 billion senior secured credit facility, and the company is reportedly targeting an initial public offering in the second half of 2026. If that IPO proceeds as targeted, Lambda would become the third neocloud among the major players covered in this piece — alongside CoreWeave and Nebius — to become a public, SEC-reporting company, which would meaningfully increase the amount of independently disclosed, audited financial detail available about the broader neocloud sector's actual economics, beyond what's currently knowable about privately held competitors like Crusoe and Nscale.

How much contracted power does Crusoe have, and how large is its future pipeline?

Crusoe reported 4.9 gigawatts of contracted power as of June 2026, alongside a disclosed pipeline exceeding 40 gigawatts of potential future capacity. That pipeline figure is substantially larger than Crusoe's current contracted position, indicating a company planning for a multi-year growth trajectory well beyond what's currently built or even currently under contract. Crusoe's campuses span Texas, Missouri, and Wyoming, and its willingness to offer AMD GPUs alongside Nvidia hardware, discussed elsewhere in this piece, is part of a broader strategy of building flexible, large-scale capacity rather than optimizing narrowly around a single chip vendor or a single facility type.

How fast has Groq scaled its data center footprint and funding in 2026?

Groq raised $650 million and then a further $350 million Series A, reaching a $3.5 billion valuation, within roughly two months in the middle of 2026 — an unusually fast sequence of fundraising events even relative to the broader AI infrastructure funding environment. Over the same period, Groq scaled to 13 data centers and 54 megawatts of active capacity, with a stated target of exceeding 200 megawatts by 2027. Groq's approach is architecturally distinct from the other four neoclouds covered here, since it's built around its own custom chip design rather than reselling Nvidia or AMD GPUs, positioning it specifically for certain inference-oriented workloads rather than as a direct, apples-to-apples price comparison against GPU-hour rates quoted by GPU-based competitors.

Which major AI labs are CoreWeave's biggest customers, and how large is its contracted backlog?

CoreWeave's named customers include Meta, Anthropic, and OpenAI, and the company carries a contracted backlog of roughly $104 billion. A backlog of that size represents years of prospective future revenue, assuming CoreWeave successfully builds out the capacity needed to deliver against those contracts and assuming its customers' own AI compute spending continues at roughly the scale currently planned. The concentration of that backlog among a small number of very large frontier AI labs is worth noting alongside the broader circular financing conversation covered elsewhere in this piece — CoreWeave's growth is, to a significant degree, tied to the continued spending trajectory of a handful of major AI companies rather than a broad, diversified customer base.

How much has Microsoft committed to Nebius, and what is the total potential value including options?

Microsoft's commitments to Nebius total $17.4-19.4 billion, including options that could increase the total value further. A commitment of this scale from one of the world's largest AI infrastructure buyers is a significant vote of confidence in Nebius specifically, and it helps explain the extraordinary revenue growth rate — 454% year-over-year in Q2 2026 — that Nebius has been able to report, since a commitment this large from a single major customer can drive outsized growth relative to a smaller company's prior revenue base.

What is CoreWeave's 2026 capital expenditure guidance, and how much recourse debt does it carry?

CoreWeave's 2026 capital expenditure guidance sits at $35-39 billion, and the company carries $31 billion in recourse debt as part of funding that buildout. Recourse debt is a meaningful detail worth understanding specifically: it means CoreWeave itself, as a company, is directly obligated to repay that debt regardless of how any specific data center project performs, as opposed to non-recourse, project-level financing where lenders' claims are limited to a specific asset. That distinction matters for evaluating CoreWeave's overall financial risk profile — a large recourse debt load raises the stakes if the pace of converting contracted power into active, revenue-generating capacity doesn't keep up with the assumptions underlying that debt.

Why do neoclouds like CoreWeave and Nebius report growing losses even as revenue accelerates?

Neoclouds report growing losses alongside accelerating revenue because they're spending aggressively on data center construction, power procurement, and GPU purchases ahead of when that capacity becomes fully active and billable — exactly the dynamic behind CoreWeave's 1.5 gigawatts active versus more than 4.2 gigawatts contracted. Building this kind of capital-intensive infrastructure at the pace the current AI compute market demands requires spending well ahead of confirmed revenue, and depreciation, financing costs, and construction spending on not-yet-active capacity all weigh on reported profitability even while top-line revenue is growing at rates most industries never achieve. Whether this resolves into durable profitability, once contracted capacity fully converts to active capacity, is one of the central open questions investors are watching closely across the entire sector.

What new investment vehicle launched in August 2026 to give investors diversified exposure to neocloud stocks?

A new exchange-traded fund launched in August 2026 specifically to give investors diversified exposure across leading neocloud stocks, covered by 24/7 Wall St. under the framing that CoreWeave and Nebius "are soaring." The launch of a dedicated ETF for this sector is a meaningful market signal in its own right, reflecting a judgment by the fund's provider that investor demand for neocloud exposure has reached a scale that justifies a purpose-built product, rather than requiring investors to assemble that exposure manually by researching and buying individual neocloud stocks one at a time — a sign of how quickly this category has moved from a specialist industry term to a mainstream investment theme.

How does the "circular financing" concern apply specifically to Nvidia's relationships with neoclouds like CoreWeave and Nebius?

Analysis from io-fund.com examining Nvidia's relationships with CoreWeave and Nebius frames these neoclouds as participants in the same circular financing pattern seen across the broader AI infrastructure market: Nvidia has financial ties to some of the same companies that are also among its largest GPU customers. This raises a version of the same question relevant to circular financing concerns more broadly — how much of a neocloud's reported growth and contracted demand reflects genuinely independent, external demand for AI compute, versus demand connected to the same small, interconnected group of companies financing and purchasing from each other. It's worth being precise that this is a relevant lens for interpretation, not evidence that the disclosed revenue figures themselves are inaccurate — CoreWeave's and Nebius's numbers are audited, disclosed figures as public companies — but it's a genuinely important factor for understanding how much of the current growth trajectory is externally validated.

Which use cases (frontier-scale training vs. budget experimentation vs. token-based inference) best fit each of the five ranked neoclouds?

The MarkTechPost/Cryptopond ranking of the leading 2026 GPU neoclouds includes specific use-case recommendations that map roughly to each provider's positioning described throughout this piece. CoreWeave's Platinum ClusterMAX rating and premium pricing suit frontier-scale training workloads where reliability failures are extremely costly and worth paying a premium to avoid. Nebius's competitive on-demand and spot pricing, including the lowest published preemptible rate among the group, fits budget-conscious experimentation and workloads that can tolerate interruption. Crusoe's scale, multi-chip-vendor flexibility, and large contracted pipeline suit customers wanting architectural flexibility across Nvidia and AMD hardware. Lambda's developer-friendly access suits smaller technical teams and researchers. Groq's custom chip architecture is positioned specifically around high-throughput, token-based inference workloads rather than training, reflecting a genuinely different technical approach than the GPU-reselling model the other four providers share.

If you're evaluating GPU infrastructure options for an AI product or internal tool, our AI agents and automation team can help scope the right sourcing model for your actual workload, and our pricing page shows how we structure engagements transparently — the same transparency this piece points to as a genuine differentiator among the neoclouds themselves.