Which AI data center startup is growing the fastest?
In our data center market deck, you will find everything you need to understand the market
SUMMARY
Crusoe is currently the fastest-growing private AI data center startup. CoreWeave remains the commercial growth leader across the wider AI infrastructure market, while Nebius is growing fastest in percentage terms.
The ranking changes depending on whether growth means earned revenue, live power, construction, signed capacity or a long-range project pipeline. We give the most weight to infrastructure that is operating or visibly being built.
Power has become the real constraint. GPUs still matter, but the companies pulling ahead are the ones that can secure land, grid access, transformers, cooling, financing and a large customer at the same time.
CoreWeave is already in a different league commercially. It added roughly $1.1 billion of quarterly revenue in one year and operates more than one gigawatt of active power, but it is now public and no longer fits a strict private-startup comparison.
Nebius has the sharpest reported growth rate. Its core AI cloud revenue rose 841% year over year, although the dollar increase remains much smaller than CoreWeave’s.
Crusoe’s advantage is the combination of 4.9 gigawatts of contracted infrastructure, an energized 200-megawatt phase in Abilene and several named campuses moving toward construction. Most of that capacity is not live yet, so the lead is real but still fragile.
Nscale is the closest challenger because its Microsoft-backed deployments cover about 200,000 GB300 GPUs, with another large Rubin wave planned. The catch is timing: much of the biggest capacity arrives in 2027 or later.
Fluidstack has the largest unmeasured upside through Anthropic’s planned $50 billion US infrastructure program. It could jump to the front quickly, but it has not disclosed enough contracted power, construction progress or expected revenue to support that conclusion today.
Customer concentration helps these companies finance enormous projects and also makes them vulnerable. Microsoft, Anthropic or Oracle can make a campus bankable, but a design change or delayed order can disrupt hundreds of megawatts at once.
Funding is no longer the deciding advantage because every serious contender can raise billions. The winner will be the company that repeatedly converts contracted power into energized buildings, installed hardware and paying customer workloads.
This market map, featured in our data center market deck, highlights top companies and startups in the data center market
Which AI data center startup is growing the fastest?
The answer depends on what we call growth. CoreWeave adds the most revenue, Nebius reports the fastest percentage growth, and several private companies are accumulating enormous future projects. Yet signed contracts, construction sites and working data centers are three very different things.
The private-company race currently comes down to Crusoe, Nscale, Fluidstack and Lambda. We give the greatest weight to infrastructure that is operational or clearly under construction, followed by contracted capacity with a named customer and delivery schedule. Vague development pipelines come last.
Using that approach, Crusoe is currently the fastest-growing private AI data center startup. CoreWeave remains the fastest-growing AI infrastructure company overall, while Nebius leads on disclosed percentage revenue growth.
If you want more recent data on this point, please see our latest data center market report.
Why is the AI data center startup race moving so fast now?
AI data center startups are expanding at an extraordinary pace because power-ready infrastructure has become nearly as scarce as advanced chips.
The latest International Energy Agency figures show that electricity use from data centers grew 17% in 2025. Consumption from AI-focused data centers increased 50%, roughly three times faster. The capital spending of five large technology companies also passed $400 billion in 2025, and the IEA expects another 75% increase in 2026.
That surge has changed the bottleneck. A few years ago, an AI cloud company mainly needed access to Nvidia GPUs and enough customers to rent them. These days, the hard part is finding a site with sufficient electricity, securing a grid connection, sourcing transformers and cooling equipment, financing construction and bringing everything online before the hardware becomes outdated.
The scale has moved beyond conventional data center development. McKinsey estimates that AI workloads could require 156 gigawatts of global capacity by 2030, including 125 gigawatts added between 2025 and 2030. One gigawatt equals 1,000 megawatts. A traditional data center may use 10 to 25 megawatts, while the projects now discussed by Crusoe, Nscale and Fluidstack can approach or exceed one gigawatt at a single campus.
The companies growing fastest are the ones that can secure power, construction capacity, financing and customers at the same time. Buying GPUs alone no longer creates a serious lead.
As this chart shows, and as featured in our data center market deck, search interest in data centers has increased significantly
Which companies actually count as AI data center startups?
The serious private AI data center startups today are Crusoe, Nscale, Fluidstack and Lambda.
We include companies that do more than rent a few GPU servers. A credible contender must develop or secure large amounts of power, deploy high-density computing infrastructure and sell that capacity to AI laboratories, hyperscalers or other large customers.
Crusoe is the clearest vertically integrated developer. It works across energy, data center construction, electrical equipment and cloud services. Nscale combines data centers, GPU infrastructure and cloud software across several countries. Fluidstack started as an AI cloud platform but is now developing custom campuses for Anthropic. Lambda has operated GPU clouds for years and is moving deeper into large AI factories.
CoreWeave and Nebius remain essential comparisons because they publish revenue and capacity figures. However, both companies are publicly traded, so neither fits a strict definition of a private startup today.
Several other businesses sit around the edges of the category. Data center developers that rent space to every type of customer are too broad for this comparison. Small GPU clouds without major power commitments are too early. Hyperscalers such as Microsoft, Amazon and Google are customers, partners and competitors, but they are clearly not startups.
| Company | Current status | What the company is building | Best public evidence of growth |
|---|---|---|---|
| Crusoe | Private | AI campuses, energy systems and AI cloud services | 4.9 GW of contracted infrastructure |
| Nscale | Private | International AI campuses and GPU cloud capacity | About 200,000 GB300 GPUs contracted with Microsoft, plus newer Rubin commitments |
| Fluidstack | Private | Custom AI data centers and managed GPU infrastructure | Leading Anthropic’s planned $50 billion US infrastructure build |
| Lambda | Private | GPU cloud and large dedicated AI factories | Multibillion-dollar Microsoft agreement covering tens of thousands of GPUs |
| CoreWeave | Public benchmark | AI-native cloud infrastructure | $2.08 billion of quarterly revenue and more than 1 GW of active power |
| Nebius | Public benchmark | Owned AI factories and AI cloud services | AI cloud revenue up 841% year over year |
What should “fastest-growing AI data center startup” actually measure?
We should judge AI data center growth mainly by live capacity and revenue, then use contracted capacity to understand what may come next.
Revenue tells us that customers are already using and paying for infrastructure. Active power shows that computing equipment is operating. Connected power usually means that a completed facility can receive electricity, although it may not yet contain a full hardware deployment.
Contracted power sits one step earlier. It can represent a real customer commitment, secured land and an agreed power allocation, but construction and financing may still take years. A development pipeline is weaker because it can include sites under negotiation, optional expansions and projects without final customers.
GPU counts are useful when the hardware generation and delivery schedule are clear. They cannot be converted neatly into gigawatts because power use varies by chip, rack design, cooling system and supporting equipment. A deployment of 20,000 current-generation GPUs may consume much more electricity than the same number of older GPUs.
We therefore avoid adding every announced gigawatt together as though it were already operating. The strongest evidence combines a named location, customer, construction stage, delivery date and operational milestone.
| Growth measure | What it proves | Current leader |
|---|---|---|
| Revenue already earned | Customers are actively using the infrastructure | CoreWeave in dollars, Nebius in percentage growth |
| Active or connected power | Facilities have reached an operational stage | CoreWeave overall, Crusoe among the main private developers |
| Contracted capacity | Customers and infrastructure have been secured for future delivery | Crusoe among private startups |
| Named GPU deployments | A specific computing commitment exists | Nscale among private startups |
| Development pipeline | The company has identified possible future projects | Useful context, but too uncertain to decide the winner |
This chart, featured in our data center market deck, illustrates yearly venture capital funding for data center startups
Is CoreWeave still the fastest-growing AI infrastructure company?
CoreWeave currently leads the entire AI infrastructure category in absolute commercial growth, although the company is now public rather than a private startup.
CoreWeave generated $2.08 billion of revenue in its latest reported quarter, compared with $982 million one year earlier. It added almost $1.1 billion of quarterly revenue in twelve months. Most private competitors do not disclose total annual revenue, let alone produce that much additional revenue in one quarter.
The physical infrastructure is also working at scale. CoreWeave reported more than one gigawatt of active power and over 3.5 gigawatts of contracted power. Its revenue backlog reached $99.4 billion after the company signed or expanded agreements with Meta, Anthropic and other customers.
These figures put CoreWeave in a different stage of development. Crusoe and Nscale may announce larger future campuses, but CoreWeave already has enough operational capacity to generate billions of dollars every quarter.
The cost of that expansion is heavy. CoreWeave recorded a $740 million net loss in the same quarter, including $536 million of net interest expense. Its 56% adjusted EBITDA margin fell to only 1% at the adjusted operating-income level once depreciation and other operating costs were considered. Growth is huge, but financing it remains expensive.
CoreWeave is the overall commercial leader today. It simply falls outside our final private-startup ranking.
If you want more recent data on this point, please see our latest data center market report.
Is Nebius growing faster than CoreWeave right now?
Nebius currently wins on percentage revenue growth, while CoreWeave still adds far more money in absolute terms.
Nebius reported $399 million of quarterly group revenue, up from $50.9 million one year earlier. That represents 684% growth. Its core AI cloud business reached $389.7 million, an 841% increase, and accounted for almost all group revenue.
CoreWeave’s revenue increased by 112% over the same type of year-on-year comparison. That percentage looks modest beside Nebius, but CoreWeave added approximately $1.1 billion of quarterly revenue. Nebius added about $348 million. Nebius is growing faster from a smaller base; CoreWeave is creating more new business in dollars.
The infrastructure numbers support Nebius’s acceleration. The company says contracted capacity now exceeds 3.5 gigawatts, with more than 75% associated with owned facilities. It expects 800 megawatts to one gigawatt of connected power by year-end and has secured gigawatt-scale sites in Missouri and Pennsylvania.
Nebius also reported a 45% adjusted EBITDA margin for its AI cloud business. That suggests recently installed capacity is being absorbed at strong utilization and pricing, rather than sitting empty while the company waits for customers.
Nebius deserves the title of fastest-growing public AI cloud in percentage terms today. It cannot answer the article’s narrower private-startup question.
This chart, featured in our data center market deck, shows how Equinix is capturing share in data centers
Which private AI data center startup is growing fastest today?
Crusoe currently has the strongest claim among private AI data center startups.
Crusoe says it has contracted 4.9 gigawatts of AI infrastructure across five US campuses and Crusoe Cloud. Crucially, the company separates those commitments from its development pipeline, which exceeds 40 gigawatts but includes sites still being negotiated or prepared.
That 4.9-gigawatt figure is the largest consolidated contracted total disclosed by any of the main private contenders. Nscale provides impressive GPU counts and individual project sizes, but it has not published one directly comparable total. Fluidstack refers to gigawatts of Anthropic infrastructure without giving a precise capacity figure. Lambda describes gigawatt-scale ambitions while disclosing much less about its current power footprint.
Crusoe has also moved beyond paper capacity. The opening 200-megawatt phase of its first Abilene campus has been energized, and construction continues on six additional buildings that take the site toward 1.2 gigawatts. The company is developing another 900-megawatt Abilene campus for Microsoft, a 900-megawatt campus in Claude, Texas, and a new one-gigawatt campus in Childress.
We cannot verify Crusoe’s revenue growth because the company keeps its financial results private. That prevents a complete comparison with CoreWeave and Nebius. Still, the combination of contracted capacity, an energized hyperscale project and several named campuses gives Crusoe the best-supported private growth story today.
If you want more recent data on this point, please see our latest data center market report.
How much of Crusoe’s AI data center growth is already live?
Only a minority of Crusoe’s planned AI data center capacity is live today, but the startup has already proved that it can deliver a large project quickly.
The first phase of Crusoe’s flagship Abilene campus contains two buildings with more than 200 megawatts of combined capacity. Construction started in 2024, and Crusoe says the buildings were constructed and energized in under one year. The campus began supporting Oracle Cloud infrastructure in 2025.
Six more buildings are being added to bring that campus to 1.2 gigawatts. The expansion is expected to reach completion around the end of 2026. Until those buildings are energized and filled with computing equipment, most of the campus should still be treated as capacity under construction.
The rest of Crusoe’s portfolio is even earlier. The second Abilene campus for Microsoft is under development. Construction on the one-gigawatt Childress campus is expected to start in the third quarter of 2026. Other Texas and Missouri projects remain at different stages of development.
Crusoe’s current advantage comes from having completed the first large step and secured several more. The company has not yet turned its entire order book into operational infrastructure, and its position would weaken quickly if the next campuses suffer major delays.
This chart, featured in our data center market deck, illustrates yearly funding for data center startups
Has Crusoe’s AI data center lead become stronger lately?
Crusoe’s private-startup lead has strengthened lately because the company keeps adding different pieces of the same expansion plan.
The newest campus announcement covers one gigawatt in Childress, Texas. Lancium owns the land, secures the grid connection and manages the energy system, while Crusoe designs, builds and operates the data center. That repeats the partnership structure first used in Abilene instead of forcing Crusoe to invent a new delivery model for each site.
Crusoe has also signed a roughly 750-megawatt power-generation agreement with Bergen Engines. Another recent agreement covers five gigawatts of battery-based uninterruptible power systems across several hyperscale campuses. Earlier in the year, Crusoe agreed to deploy 12 gigawatt-hours of long-duration iron-air batteries.
Taken together, these announcements show a company working through recurring bottlenecks: finding sites, securing primary power, handling short-term grid disturbances, storing energy and standardizing construction. The group is more convincing than another giant campus announcement on its own.
Crusoe is also building manufacturing capacity for prefabricated electrical and data center components. Its smaller Spark facilities are designed to be deployed in roughly 90 days where power and local inference demand are available.
The recent pattern strengthens our conclusion. Crusoe is expanding the machinery needed to deliver several campuses, rather than betting everything on one famous project.
Can Nscale overtake Crusoe in AI data centers?
Nscale could overtake Crusoe, but its largest AI data center deployments still sit further in the future.
Nscale has contracted approximately 200,000 Nvidia GB300 GPUs with Microsoft across the United States and Europe. The program includes about 104,000 GPUs at a 240-megawatt Texas facility, 12,600 in Portugal, roughly 23,000 in the United Kingdom and approximately 52,000 in Norway.
Some delivery has already begun in Portugal, while the Texas deployment is expected to start during 2026. The UK project is scheduled for 2027. That gives Nscale a real near-term program, although a large part of the total remains ahead.
The next wave is even larger. Nscale plans to supply more than 66,000 Rubin GPUs at a second 200-megawatt building in Portugal and over 30,000 additional Rubin GPUs in Norway. Those systems are expected from 2027. Its proposed West Virginia campus begins with roughly 1.35 gigawatts for Microsoft and could ultimately expand much further.
Nscale also has exceptional access to money. It raised $2 billion in equity at a $14.6 billion valuation, arranged a $1.4 billion GPU-backed loan, secured $790 million for Norway and recently added a $900 million revolving credit facility.
The company may be accumulating future commitments faster than Crusoe. We still rank Crusoe first today because more of its case rests on consolidated contracted power and a hyperscale campus that has already been energized.
| Nscale project | Initial disclosed scale | Expected timing |
|---|---|---|
| Texas for Microsoft | About 104,000 GB300 GPUs and 240 MW | Phased delivery beginning in 2026 |
| Portugal for Microsoft | About 12,600 GB300 GPUs | Initial delivery underway |
| UK for Microsoft | About 23,000 GB300 GPUs and 50 MW | Expected in 2027 |
| Norway for Microsoft | About 52,000 GB300 GPUs, followed by more than 30,000 Rubin GPUs | GB300 deployment followed by Rubin expansion |
| Portugal expansion | More than 66,000 Rubin GPUs and a second 200 MW building | Beginning in late 2027 |
| West Virginia | Initial collaboration covering about 1.35 GW | First tranches expected from late 2027 |
This chart, featured in our data center market deck, compares the main business model options for hyperscale data center operators
Does Fluidstack’s $50 billion Anthropic project make it the AI data center leader?
Fluidstack has the boldest customer-backed expansion story today, but the public evidence remains too thin to rank it first.
Anthropic announced a planned $50 billion investment in American computing infrastructure built with Fluidstack. The program includes custom data centers in Texas and New York, with more locations expected. The first sites are supposed to come online throughout 2026.
The size of Anthropic’s business makes the commitment credible. Anthropic recently said its annualized revenue had passed $30 billion, up from about $9 billion at the end of 2025. The number of customers spending more than $1 million annually reportedly doubled from over 500 to more than 1,000 in less than two months. Anthropic clearly needs additional computing capacity quickly.
However, the $50 billion belongs to Anthropic’s overall infrastructure investment. It should not be presented as Fluidstack revenue, Fluidstack backlog or a fully funded construction budget controlled by Fluidstack.
Fluidstack says the project will involve gigawatts of power and that its approach can deliver gigawatt-scale computing in six months. It has not publicly disclosed the exact power contracted, the number of GPUs ordered, the capacity already under construction or the revenue Fluidstack expects to receive.
The company has serious operational experience. Fluidstack previously reported more than 100,000 GPUs under management and recently raised $830 million at a $7.5 billion valuation. Its hiring plans now span data center design, construction, manufacturing, power systems and operations, which fits a company preparing for a much larger physical build.
Fluidstack may eventually grow faster than Crusoe. The missing numbers stop us from saying that it already has.
If you want more recent data on this point, please see our latest data center market report.
Is Lambda really becoming an AI data center developer?
Lambda is becoming a serious AI data center developer, although its disclosed physical footprint still trails Crusoe and Nscale.
Lambda spent much of its history selling GPU servers and cloud access. Its Microsoft agreement marked a change in scale. The multiyear contract covers tens of thousands of Nvidia GPUs, including GB300 systems, and Lambda describes the total value as several billion dollars.
The company has since raised more than $1.5 billion in equity and expanded a credit facility from $275 million to $1 billion. Lambda says the financing will pay for new accelerator servers and additional data center capacity tied to contracted revenue.
There is also physical evidence outside the funding announcements. EdgeConneX is developing a 23-megawatt single-tenant facility for Lambda in Chicago, alongside two existing sites in Atlanta and Chicago. Lambda’s private-cloud products now cover dedicated clusters ranging from roughly 1,000 to 64,000 GPUs.
Even so, Lambda has not disclosed a company-wide figure for active power, connected power or contracted capacity. “Gigawatt-scale AI factories” remains more of a direction than a measurable current footprint.
Lambda is clearly in the race, especially on cloud revenue and customer diversity. Its disclosed data center footprint is still too limited for the top position.
This chart, featured in our data center market deck, shows the revenue mix across customer segments in the data center market
Could one big customer derail an AI data center startup?
A single customer could seriously disrupt the growth of today’s AI data center startups.
Fluidstack’s leap into custom campus development is closely tied to Anthropic. Nscale’s largest disclosed GPU deployments are overwhelmingly connected to Microsoft. Lambda’s biggest announced infrastructure contract is also with Microsoft. Crusoe has a somewhat broader customer base, including Oracle, Microsoft and another unnamed hyperscale technology company, but a handful of large projects still accounts for most of its visible expansion.
Large customers make these projects financeable. Banks and infrastructure investors are far more willing to fund a billion-dollar campus when Microsoft, Oracle or Anthropic has committed to using it. The customer contract can effectively support the debt raised to buy GPUs and construct the facility.
The same structure creates dependence. A hardware change, design revision, construction delay or reduction in customer demand can affect an entire campus. Replacing a customer that intended to use tens of thousands of identical GPUs is much harder than finding several smaller cloud clients.
CoreWeave shows how a provider can eventually diversify. Its latest customer announcements include Meta, Anthropic, Jane Street, Cohere, Mistral and Perplexity. The private challengers have not yet demonstrated that breadth at comparable scale.
Customer concentration does not overturn our Crusoe conclusion, but it lowers the certainty. Crusoe’s lead will become much stronger when several campuses are live for unrelated customers.
Does funding decide which AI data center startup wins?
Funding no longer separates the serious AI data center startups because every leading contender can now raise billions.
Nscale has assembled the largest recent corporate financing package. Its $2 billion Series C was followed by a $1.4 billion GPU-backed loan, $790 million of financing for Norway and a $900 million revolving facility. These instruments cannot all be treated as unrestricted cash, but together they provide substantial flexibility across equipment purchases and construction.
Crusoe raised $1.375 billion at a valuation above $10 billion. It also secured a $750 million Brookfield credit facility and participates in a $15 billion joint venture funding the first Abilene campus.
Lambda raised more than $1.5 billion and now has its $1 billion credit facility. Fluidstack’s $830 million Series A is smaller, although it arrived unusually early in the company’s funding history and valued the business at $7.5 billion.
The real financing test comes after the announcement. AI data centers require repeated spending on land, power equipment, buildings, cooling, networking and GPUs. A company can raise several billion dollars and still lack enough capital to complete a multi-gigawatt portfolio.
We treat financing as permission to compete, rather than proof that a company is winning. Delivery will decide what that money was worth.
This chart, featured in our data center market deck, shows how hyperscale AI-ready campus technology has evolved over time
Which AI data center business model can scale fastest?
Crusoe currently has the most convincing private AI data center model because it controls more of the journey from electricity to computing.
Crusoe works on power sourcing, energy storage, electrical equipment, campus construction, cloud services and infrastructure operations. The company can prepare power and construction in parallel, manufacture some long-lead components itself and reuse a partnership model across several sites.
Nscale is building a similarly broad stack, including power, data centers, GPUs and cloud software. Its international portfolio may eventually become an advantage because it can serve sovereign AI demand across Europe and North America. Coordinating many countries, partners and hardware generations also creates more opportunities for delays.
Fluidstack is rapidly moving toward an integrated model. Its public description now covers power acquisition, data center design, construction and operation. The Anthropic program will provide the first large test of whether it can perform those tasks across several gigawatts.
Lambda’s model is more cloud-led. That gives Lambda experience serving many different customers, but it remains more dependent on third-party data center developers for parts of its physical expansion.
None of the private companies publishes enough financial information for a clean profitability comparison. CoreWeave and Nebius show why that question will eventually become critical. AI clouds can produce impressive adjusted EBITDA while still spending billions on equipment, depreciation and interest.
For now, Crusoe’s model looks best suited to the current bottleneck: turning scarce power into working AI infrastructure quickly and repeatedly.
If you want more recent data on this point, please see our latest data center market report.
Which AI data center startup is growing the fastest today?
Crusoe is currently the fastest-growing private AI data center startup.
The company has the largest clearly disclosed contracted infrastructure total among the private contenders, one major AI campus already energized and several additional campuses moving through construction or development. Its latest agreements also cover power generation, energy storage, grid stability and modular manufacturing, giving us more than one type of evidence.
Nscale is the closest challenger. It has accumulated an enormous number of Microsoft-backed GPU commitments and may have the strongest chance of taking the lead once its 2027 projects begin operating. Today, too much of that capacity remains scheduled for later delivery.
Fluidstack has the greatest unmeasured upside. Anthropic’s planned infrastructure spending could make Fluidstack one of the world’s largest AI data center operators, but the company has not disclosed enough contracted power, construction progress or expected revenue to justify first place.
Lambda is growing into a larger infrastructure company and may have stronger cloud revenue than the public evidence suggests. Its disclosed data center footprint is still too limited for the top position.
Outside the private-startup definition, CoreWeave remains the absolute commercial growth leader. Nebius holds the clearest percentage-growth lead. Neither result changes the private-company answer.
Our judgment is direct: Crusoe leads today, Nscale could catch it, and Fluidstack could surprise both. Crusoe will keep the title only if the next wave of contracted campuses becomes operational on schedule.
In our data center market deck, we identify pain points entrepreneurs should prioritize
OUR METHODOLOGY
This analysis tests which AI data center startup is growing fastest based on evidence available today. We compare earned revenue, active and connected power, construction progress, contracted capacity, named GPU deployments, customer commitments, financing and the ability to repeat a delivery model across several projects.
We give the greatest weight to infrastructure that is already operating or clearly under construction. A named project with a customer, location, delivery schedule and physical milestone counts more than a broad development pipeline, even when the pipeline is much larger.
Revenue and active power show what has already been delivered. Connected power usually shows that a completed facility can receive electricity, although it may not yet contain a full hardware deployment. Contracted capacity is useful for measuring what comes next, but it can still face financing, permitting, construction and hardware delays.
GPU commitments are treated as strong evidence only when the hardware generation, customer and delivery timing are clear. We do not convert GPU counts directly into gigawatts because power use varies by chip, rack design, cooling system, networking and supporting equipment.
We keep public companies in the comparison as benchmarks rather than private-startup candidates. CoreWeave provides the cleanest view of absolute commercial growth, while Nebius provides the clearest recent percentage-growth comparison.
No single figure determines the result. We look for several pieces of evidence moving in the same direction, especially the combination of secured power, a credible customer, financing, construction progress and an operational milestone.
We also separate a company’s own revenue or backlog from a customer’s broader spending plan. Anthropic’s planned $50 billion infrastructure investment, for example, is evidence of Fluidstack’s opportunity, but it is not treated as Fluidstack revenue or a fully disclosed Fluidstack backlog.
Key sources include the International Energy Agency on data-center electricity growth and AI power demand, McKinsey on projected AI data-center capacity through 2030, CoreWeave’s first-quarter 2026 results, and Nebius’s first-quarter 2026 results.
For the private-company ranking, the main company sources are Crusoe’s disclosure of 4.9 GW of contracted infrastructure, Crusoe’s Abilene operational milestone, Nscale’s Microsoft-backed GB300 program, Anthropic’s infrastructure announcement with Fluidstack, and Lambda’s multibillion-dollar Microsoft agreement.
Additional project evidence includes Crusoe’s 900 MW Microsoft campus, the planned 1 GW Childress campus, Nscale’s planned Rubin deployment in Portugal, Nscale’s Norway expansion, and Lambda’s $1 billion credit facility.
This chart, featured in our data center market deck, shows the revenue mix by region across Europe, Asia, North America, Africa, and South America in the data center market
Related blog posts
