Which data center startup is growing the fastest?

Why is the data center startup race so hard to read today?

The data center startup race is hard to read today because companies report growth in megawatts, GPUs, revenue, leases and future campuses, even though those figures describe very different stages of progress.

A signed 500-megawatt lease can look larger than a live 200-megawatt site, while the smaller number may represent far more work already completed. GPU totals create another problem. A company can reserve 100,000 next-generation chips for delivery over several years, but that tells us little about how many are installed, approved by the customer and earning revenue now.

The market has also changed quickly. AI laboratories and the largest cloud companies are signing contracts that once would have looked implausibly large for a young infrastructure company. Crusoe has disclosed several gigawatts under contract. Fluidstack is helping deliver Anthropic’s huge US build-out. Nscale has accumulated Microsoft commitments covering hundreds of thousands of GPUs. Lambda has signed a multibillion-dollar Microsoft agreement.

Several startups can now claim to be growing fastest, depending on whether we look at revenue, signed demand, live power or construction speed. A useful answer has to connect those measures instead of picking whichever headline produces the biggest number.

Which data center startups belong in this comparison?

We are comparing Crusoe, Fluidstack, Nscale, Lambda, TensorWave and Voltage Park as data center startups, while CoreWeave stays outside the ranking as a public-company reality check.

Crusoe and Nscale increasingly handle the whole chain, from securing power and building facilities to selling AI compute. Fluidstack combines its own engineering and operations with capacity developed through infrastructure partners. Lambda, TensorWave and Voltage Park lean more heavily toward the GPU cloud model, although all three now talk about larger data center operations.

CoreWeave would dominate the ranking if we still called it a startup. Its latest annual report showed more than $5 billion in yearly revenue, 168% growth, over 850 megawatts of active power and 43 data centers. CoreWeave has been public since 2025, so including it would turn a private-startup comparison into a contest with a company that has already crossed into a different scale.

Company	What it mainly builds today	Why it belongs here	Role in our ranking
Crusoe	AI campuses, power systems and GPU cloud	Large live campus plus multi-gigawatt contracts	Full contender
Fluidstack	Custom AI data centers and managed clusters	Rapidly expanding projects backed by large cloud customers	Full contender
Nscale	AI campuses and GPU infrastructure	Very large Microsoft and European commitments	Full contender
Lambda	GPU cloud and dedicated AI factories	Large revenue base and major customer contracts	Full contender
TensorWave	AMD-focused GPU cloud	Large live cluster and ambitious capacity plans	Emerging contender
Voltage Park	Nvidia GPU cloud across multiple sites	Large owned fleet after its Lightning AI merger	Emerging contender
CoreWeave	Public AI cloud and infrastructure platform	Best benchmark for proven hypergrowth	Benchmark only

What does “fastest-growing data center startup” mean here?

For this data center startup ranking, growth means turning customer demand into operating infrastructure quickly enough for revenue to follow.

Live or billable capacity deserves the most weight because customers can already use it. Signed capacity comes next, especially when we know the customer, the site and the expected delivery. Revenue also helps, although private companies rarely report it in the same way.

Land options, grid connection applications and a site’s theoretical maximum tell us much less. Projects often shrink, move or arrive years late. Fundraising can pay for construction, but it says nothing about whether the buildings will open on time.

The winner should have both large customer commitments and facilities that people can already use. A single spectacular announcement carries little weight on its own.

Which data center startup has the largest live operation now?

Crusoe currently has the largest live data center operation we can verify among the private startups in this comparison.

Crusoe’s first two Abilene buildings provide 200 megawatts and were energized within a year of construction starting. Oracle began installing Nvidia GB200 racks, and the campus started running training and inference workloads. Six more buildings have been under construction as part of the same 1.2-gigawatt site.

The other startups disclose less comparable information. Nscale has begun deploying more than 12,600 Blackwell Ultra GPUs for Microsoft in the first Sines building, while its Glomfjord operation gives it another established base in Norway. Voltage Park says its merged platform owns and operates more than 36,000 H100, B200 and GB300 GPUs across seven Tier III or better data centers. TensorWave has deployed 8,192 AMD MI325X accelerators in a liquid-cooled cluster.

Crusoe is the only startup here that lets us trace a large power figure all the way to named buildings, a customer and active workloads. Fluidstack and Lambda may operate more compute than they disclose, but the public figures leave that unproven.

If you want more recent data on this point, please see our latest data center market report.

Which data center startup is growing revenue fastest?

Fluidstack appears to have the fastest percentage revenue growth among the private data center startups we examined, while Lambda probably has the largest revenue base.

Fluidstack’s latest Companies House accounts show £52.7 million of revenue for 2024, around 140% above the previous year. Its newer Anthropic-backed projects are vastly larger than the business reflected in those accounts, so Fluidstack’s percentage growth could remain exceptional as the new sites begin generating revenue.

Lambda is bigger in absolute dollars. Sacra estimates that Lambda reached a $760 million annualized revenue rate in 2025, up 79% from $425 million at the end of 2024, with more than $520 million generated over the preceding twelve months. These estimates lack public-company audits, but they place Lambda well ahead of the older private estimates for Crusoe.

Crusoe’s last widely cited estimate was $276 million for 2024, up 82%. That number now looks dated beside the scale of Crusoe’s campus contracts. Revenue still leaves the overall ranking unresolved. Infrastructure revenue often arrives after years of site work, while cloud revenue can rise as soon as additional GPU clusters become available.

Data center startup	Best available revenue evidence	What we can safely conclude
Fluidstack	£52.7M in 2024 accounts, roughly 140% yearly growth	Strongest documented percentage growth from a smaller base
Lambda	Estimated $760M annualized rate in 2025, up 79%	Likely largest private revenue base
Crusoe	Estimated $276M in 2024, up 82%	Strong earlier growth, but current revenue is less transparent
Nscale	No reliable comparable revenue series	Contracts are much clearer than recognized revenue
TensorWave and Voltage Park	No reliable comparable revenue series	Too little disclosure for a revenue ranking

Which data center startup has signed the most capacity?

Crusoe has signed the largest disclosed pool of data center capacity among the private startups we examined.

We can tie Fluidstack to roughly one gigawatt of critical IT load, meaning power available to the computing equipment, across named projects before counting optional expansions. The total includes about 380 megawatts at TeraWulf’s Lake Mariner site, 207 megawatts at Cipher Mining’s Barber Lake campus, 168 megawatts at Abernathy and 245 megawatts at Hut 8’s River Bend site.

Nscale usually describes its growth through GPUs and site phases. Its Microsoft agreement originally covered about 200,000 GB300 GPUs across several locations, including approximately 104,000 GPUs at a 240-megawatt Texas campus. Nscale has since added more than 66,000 Rubin GPUs planned for Portugal and more than 30,000 for Norway, mostly scheduled for 2027.

The categories overlap. Crusoe combines campus projects with cloud capacity, Fluidstack reports leases and joint ventures, and Nscale counts GPUs across different chip generations and delivery years. It is not a clean comparison, but Crusoe still has the largest disclosed amount under contract today.

Data center startup	Clearest contracted or committed scale	How much is already proven
Crusoe	4.9 GW across five campuses and cloud capacity	First Abilene phase live, more buildings underway
Fluidstack	About 1 GW across named partner sites	Most of the large sites are being delivered now and into 2027
Nscale	Large Microsoft GPU package, plus later Rubin commitments	First Portugal deployment underway; largest phases remain ahead
Lambda	Tens of thousands of GPUs for Microsoft	Major contract, limited site-level disclosure
TensorWave	More than 2 GW of claimed long-term capacity	A large MI325X cluster forms the clearest live base
Voltage Park	Large owned and operated multi-site GPU fleet	Real infrastructure, but no comparable contracted MW total

Is Crusoe’s data center lead as solid as it looks?

Crusoe’s data center lead looks solid today because operating buildings, active construction and newly added sites all support the contracted growth.

As noted above, Crusoe’s latest capacity update put the company at 4.9 gigawatts under contract and more than 40 gigawatts in its broader development pipeline. The contracted portion covers five US campuses. At Abilene, two buildings are operating, six more are being completed for the original campus, and site preparation has started for a separate 900-megawatt Microsoft development.

Crusoe has also kept adding infrastructure around those contracts. A newly announced 1-gigawatt Childress campus is due to begin construction in 2026 for an unnamed large technology customer. A separate power-resilience agreement covers five gigawatts of battery-based systems across several campuses. Neither announcement proves that the full capacity is live, but together they show the company moving beyond one flagship project.

Most of Crusoe’s contracted gigawatts still have to be built. Even so, the company has already repeated the sequence of securing power, starting site work and bringing large buildings online. No private rival has shown the same combination at this scale.

If you want more recent data on this point, please see our latest data center market report.

Is Fluidstack catching Crusoe in data centers?

Fluidstack is catching Crusoe faster than any other data center startup, although it still trails on completed large-scale capacity.

Anthropic chose Fluidstack to help build custom data centers in Texas and New York within a wider $50 billion US infrastructure program. The projects listed above give Fluidstack a large, customer-backed construction program across TeraWulf, Cipher Mining and Hut 8 sites. Several of those contracts extend for ten to twenty-five years and carry multi-billion-dollar revenue commitments for the individual projects.

TeraWulf recently agreed to sell its 50.1% stake in the 168-megawatt Abernathy joint venture to an investor group led by Fluidstack. Fluidstack will continue leading the project, which moves the company closer to controlling an asset instead of simply leasing capacity from a partner.

Fluidstack now has to prove its claim that it can bring gigawatts online in roughly six months, compared with the industry’s usual 18 to 24 months. Its largest sites still lack enough independently confirmed operating capacity for us to accept that speed at full scale. The contracts have raced ahead of the finished facilities.

Has Nscale become the biggest threat in the data center startup race?

Nscale has become the biggest threat to Crusoe in projects planned for the next few years, but its current ranking still depends heavily on projects scheduled for later delivery.

The original Microsoft package covered roughly 200,000 GB300 GPUs in the United States and Europe. Nscale then added more than 66,000 Rubin GPUs at the Sines campus in Portugal, starting in late 2027, and more than 30,000 Rubin GPUs for Microsoft in Norway. It has also announced a 1.35-gigawatt West Virginia AI factory with Microsoft, Nvidia and Caterpillar.

At Sines, Nscale says more than 12,600 Blackwell Ultra GPUs are being deployed in the first building. The company also secured €790 million for its Norway build-out and recently closed a $900 million revolving credit line for projects across several regions.

A planned UK supercomputer site in Loughton was still at an early planning stage when reporters visited in early 2026, despite the large investment claims attached to it. That episode leaves the progress in Portugal, Norway and Texas intact, while showing why each campus needs to be checked separately.

Nscale could pass every private competitor if the Microsoft deployments arrive on schedule. For now, it has promised far more capacity than it currently operates.

If you want more recent data on this point, please see our latest data center market report.

Is Lambda growing faster than the data center builders?

Lambda probably beats several data center developers on AI cloud revenue growth, but its physical capacity is too poorly disclosed for first place.

Lambda’s multiyear Microsoft agreement covers tens of thousands of Nvidia GPUs, including GB300 systems. The company also increased an earlier $275 million credit facility to $1 billion, giving it more capital for servers and data center expansion. Lambda says it operates gigawatt-scale AI factories and serves tens of thousands of customers.

Lambda does not publish a simple total for active megawatts, billable capacity or buildings under construction. A cloud provider can grow by installing clusters in facilities owned by other companies, and that growth is commercially real. It tells us less about Lambda’s ability to develop campuses from the power connection to opening day.

Lambda currently looks strongest when we ask which private AI cloud has the biggest visible revenue base. Its claim weakens once the subject becomes physical data center growth.

Are TensorWave and Voltage Park growing fast enough as data center startups?

TensorWave and Voltage Park now matter as data center startups, but neither currently challenges the three leading builders.

TensorWave has taken a focused route around AMD, avoiding the same Nvidia supply race as almost everyone else. Its 8,192-GPU MI325X cluster gives customers a large liquid-cooled AMD option, and the company has started rolling out MI355X capacity. TensorWave has also discussed more than two gigawatts of long-term secured capacity.

Voltage Park took a different step by merging with Lightning AI. The combined platform says it owns and operates more than 36,000 Nvidia H100, B200 and GB300 GPUs, while Lightning contributes software used by more than 400,000 developers and organizations. This expands Voltage Park from a fleet of rentable chips into a fuller cloud platform.

TensorWave offers a clear alternative to Nvidia-heavy clouds, while Voltage Park joins infrastructure with a large developer funnel. Both still disclose too little for a clean comparison. Neither company has shown the contracted megawatts, construction schedule and named campus progression that would support a claim to the fastest data center build-out.

Which data center startup is turning contracts into working capacity fastest?

Crusoe remains the fastest data center startup at converting large contracts into working capacity.

Construction on the first Abilene phase began in mid-2024, and two 100-megawatt buildings were energized within a year. Oracle started installing GB200 racks soon afterward, and real workloads followed. Six more buildings are due to complete the original 1.2-gigawatt campus, while another Abilene site for Microsoft has already entered land-clearing and preparation.

Fluidstack’s biggest projects are being built at several partner sites, including the Abernathy campus that it now leads more directly. Nscale has begun its first large Sines deployment, but most of its new Rubin commitments start in 2027. TensorWave and Voltage Park have live GPU fleets, though their data does not reveal the same end-to-end campus progression.

Crusoe has already done this more than once. The company moved from one completed phase to additional buildings, another Abilene campus, sites elsewhere in Texas and a Missouri project. Fluidstack and Nscale can close the gap by completing their current wave, but the evidence available now still favors Crusoe.

Data center startup	Strongest delivery evidence now	Main delivery test ahead
Crusoe	First Abilene phase energized within a year	Finish the remaining Abilene phases and repeat at new campuses
Fluidstack	Multiple financed partner projects, plus greater control at Abernathy	Show several hundred operating megawatts across different sites
Nscale	First large Sines deployment and an established Nordic base	Deliver the much larger Microsoft and Rubin programs
Lambda	Large customer deployments and a growing AI cloud	Publish active power and campus-level progress
TensorWave	Live liquid-cooled AMD cluster	Turn long-term capacity claims into additional operating sites
Voltage Park	Large owned GPU fleet across seven data centers	Prove sustained fleet and revenue growth after the merger

Are data center startups too dependent on Microsoft, Anthropic and OpenAI?

Customer concentration is currently one of the biggest weaknesses in the data center startup boom.

Microsoft sits behind the largest disclosed programs at Nscale and Lambda, as well as Crusoe’s new 900-megawatt Abilene campus. OpenAI and Oracle drive the original Crusoe Abilene development. Anthropic anchors Fluidstack’s transformation and supports several projects financed through infrastructure partners.

These are excellent customers for a young data center company. Their long contracts can make billions of dollars of construction financeable, and their workloads justify unusually dense facilities. The same concentration can make several announcements look more independent than they really are. A customer contract, the funding raised against it and the campus leased to fulfill it all trace back to one buying decision.

The dependence is strongest at Nscale and Fluidstack because each company’s recent jump is tied closely to one main customer. Crusoe has spread its bets slightly wider across named projects linked to large cloud customers. Lambda says it serves tens of thousands of customers, although Microsoft dominates its largest disclosed contract.

Strong demand from a handful of buyers can support spectacular growth for years. It also means that a delayed customer deployment can move hundreds of megawatts at once.

Which data center startup has enough money to deliver its plans?

Nscale currently has the largest new financing package among the private data center contenders, while Crusoe, Fluidstack and Lambda can also fund a serious wave of construction.

Nscale closed a $2 billion equity round and later added a $900 million revolving credit facility, alongside dedicated financing for Norway. Crusoe raised $1.375 billion at a valuation above $10 billion and has also used credit facilities and joint-venture financing for individual projects. Fluidstack raised $830 million, while project partners and customer-backed leases carry much of the campus construction burden. Lambda has raised more than $1.5 billion in equity and now has a $1 billion secured credit facility.

No balance sheet in this group can comfortably self-fund every announced gigawatt. Data centers can cost several million dollars per megawatt before GPUs, and the largest projects run into tens of billions. The winning model will combine corporate equity, debt, project finance and long-term customer commitments.

Money separates these four leaders from lightly funded cloud startups, but it no longer tells us which leader will win. The hard part now is building on time.

Which data center startup has the most fragile growth story?

Nscale has the most fragile data center growth story today because its promised scale has expanded much faster than what it runs now.

Nscale is trying to deliver huge deployments across Texas, Portugal, Norway, the UK and West Virginia within a short period. The company has real customers, financing and major industrial partners behind the build-out. The difficulty comes from running many complicated projects at once while moving from tens of operating megawatts toward several gigawatts of live capacity.

Fluidstack also faces a steep jump, although its partner-led model spreads construction across companies that already control land and power. That lowers some development risk and creates more coordination risk. Crusoe is taking on several giant campuses, but its Abilene record gives us a stronger baseline for judging the next projects.

TensorWave carries a different risk. Its long-term capacity claim is far larger than its live AMD fleet. Voltage Park must prove that its merger with Lightning AI creates more usage instead of merely a larger combined product story. Lambda’s risk is harder to measure because its physical operations remain poorly disclosed.

If you want more recent data on this point, please see our latest data center market report.

Can Fluidstack become the fastest-growing data center startup?

Fluidstack can take first place if its current contracts become several hundred megawatts that customers have approved and started paying for.

The company already has enough demand to challenge Crusoe. As shown above, about one gigawatt can be tied to named partner projects, and Anthropic has said that more US sites will follow Texas and New York. Fluidstack’s move to lead the Abernathy ownership group also suggests that it wants more control over delivery and economics.

The ranking would change once several large Fluidstack sites open close to schedule, pass customer checks and show that the company can repeat the process. Another contract announcement would add little by itself because demand is already obvious.

Nscale has a similar path, but most of its biggest new commitments arrive later. Lambda would need much clearer megawatt reporting. Crusoe can defend its position by finishing the original Abilene campus and showing that Childress, the Microsoft site and its other contracted campuses follow the same construction pattern.

Which data center startup is growing the fastest?

Crusoe is currently the fastest-growing private data center startup overall.

Crusoe leads because it combines the largest disclosed amount under contract with the strongest proof that a young company can build and energize a hyperscale AI campus quickly. The first Abilene phase is operating, the rest of that campus is advancing, a separate Microsoft development is underway and newer campuses have extended the model beyond one site.

Fluidstack wins a narrower contest. Its contracts have grown faster relative to its previous revenue and live capacity, making it the most aggressive challenger. Nscale has accumulated the biggest set of future GPU commitments, but too much of its scale still depends on 2027 delivery. Lambda probably leads the remaining private companies on revenue, while its physical growth is too opaque for first place.

The answer changes when we isolate one measure. Fluidstack leads on the speed of new contracts, Nscale on GPUs promised for later and Lambda on estimated private revenue. The combined evidence still points to Crusoe because operating capacity deserves more weight than capacity promised for later.

The lead is real and could change quickly. Fluidstack could overturn it during the next wave of openings, and Nscale could become much larger if its Microsoft projects arrive as planned. Today, Crusoe has done the best job of turning AI demand into data centers people are already using.

If you want more recent data on this point, please see our latest data center market report.

Who is the author of this content?

NEW MARKET PITCH TEAM

We track new markets so founders and investors can move faster

We build living "market pitch" documents for emerging markets: AI, synthetic biology, new proteins, and more. Instead of outdated PDFs or hallucinated LLM answers, our clients get a clean, visual, always-updated view of what's really happening: key players, deals, regulations, and signals that matter. Learn more about us.