Data centers: what are the top startups now?

Why are data center startups attracting so much money now?

AI has turned data centers into one of the hottest infrastructure markets, and investors are putting billions into companies that can secure electricity, build facilities and get GPUs running quickly.

JLL's 2026 Global Data Center Outlook estimates that more than $300 billion has flowed into global data center mergers and acquisitions since 2020. The scale of individual deals has also changed. In July 2026, a consortium involving BlackRock's Global Infrastructure Partners, MGX and the Artificial Intelligence Infrastructure Partnership completed the $40 billion acquisition of Aligned Data Centers, alongside a further $5 billion commitment for expansion.

Private AI infrastructure companies are attracting similarly large amounts of capital. Crusoe announced a $3.9 billion financing at a $30.9 billion valuation, while Nscale secured $2 billion at a $14.6 billion valuation. These companies now raise amounts comparable to leading AI software developers, despite having to finance expensive buildings, power systems and computing equipment. Raising amounts that rival leading AI software developers is unusual, and we look at who has raised the most, and from whom, in our report covering data centers.

AI companies need thousands of high-performance GPUs operating together, often with enormous electricity and cooling requirements. A developer that can supply powered facilities within months has an advantage over a competitor promising the same capacity several years later.

What actually makes a data center startup one of the best?

The top data center startups currently combine real operating capacity, paying customers, strong financing and the ability to bring new infrastructure online. But comparing them requires separating several very different businesses.

Crusoe builds AI data centers and sells computing services. Nscale develops large AI infrastructure projects. Lambda and Fluidstack supply GPU computing capacity to AI developers. Vantage Data Centers and DataBank focus more heavily on physical facilities, while ZutaCore and Submer sell cooling systems.

These businesses have different financial requirements. Data center developers spend heavily on land, construction and electrical infrastructure, often relying on long-term leases to recover those investments. GPU cloud providers also need to buy expensive processors that can lose value as newer generations arrive. Cooling suppliers can sell equipment across many facilities without owning the underlying data centers.

We therefore give the most weight to delivered infrastructure, commercial contracts and access to capital. Company maturity matters too. Vantage is an established private infrastructure platform, while smaller cooling developers are closer to traditional venture-backed startups.

Which data center startups are leading right now?

Crusoe is our top private AI data center contender today, with Nscale and Lambda close behind. Fluidstack is one of the most promising challengers, while Vantage and DataBank are stronger choices when we focus specifically on physical data center development.

Crusoe has the clearest combination of operational delivery and huge contracted demand. Its reported 1 GW of delivered and operational gross capacity puts its infrastructure claims on a different footing from companies whose largest projects remain under construction.

Nscale has secured major Microsoft-related projects across multiple locations, including the United States and Norway. Lambda has built a substantial GPU cloud operation and arranged billions in infrastructure financing. Fluidstack has attracted major investment and works on computing deployments for advanced AI laboratories, although it publishes less detailed operating information.

Among physical developers, Vantage and DataBank have established financing relationships and actual facilities. The cooling market has its own leaders: ZutaCore stands out for commercial deployments and industrial backing, while Submer has expanded from immersion cooling into broader AI infrastructure.

Our ranking combines execution, scale, customer demand and the strength of each company's position. It deliberately includes mature private operators and specialist suppliers alongside venture-backed AI cloud companies.

Company	Main business	Evidence of position	Our assessment
Crusoe	AI campuses and GPU cloud	$30.9B valuation; 1 GW reported operational gross capacity	Overall private leader
Nscale	AI infrastructure and cloud	$14.6B valuation; major Microsoft projects	Strongest large-scale challenger
Lambda	GPU cloud	Large secured financing facilities and major AI customers	Established private contender
Fluidstack	Large GPU deployments	$830M Series A at a $7.5B valuation	Fast-growing challenger
Vantage Data Centers	Hyperscale facilities	Global campus portfolio; $2B development facility	Major private developer
DataBank	Colocation and hyperscale facilities	$2.65B Red Oak campus financing	Proven U.S. operator
Submer	Cooling and AI infrastructure	More than 500 MW in company-reported liquid-cooling deployments	Broad infrastructure specialist
ZutaCore	Direct-to-chip cooling	$100M Series C; more than 75 reported deployments	Cooling technology leader
Accelsius	Two-phase cooling	$65M Series B	Growing cooling challenger
Corintis	Microfluidic chip cooling	$25M Series A1	Early-stage technology contender

Is Crusoe the strongest private AI data center company?

Crusoe currently has the strongest case for being the leading private AI data center company because it has combined enormous customer commitments with substantial infrastructure already running.

In September 2026, Crusoe announced a $3.9 billion Series F at a $30.9 billion post-money valuation. The company reported more than $140 billion in total contracted value, more than 6 GW of gross contracted capacity and 1 GW delivered and operational.

The capacity figures tell us more than the valuation. One gigawatt equals 1,000 megawatts, although Crusoe's gross capacity figures cannot be directly compared with every competitor's critical IT load. Nevertheless, the company's delivered infrastructure gives us concrete evidence of its ability to complete projects.

Crusoe is also developing its computing services. The company reported more than $100 million in contracted annual recurring revenue for its managed inference product and more than 20-fold year-over-year growth in Crusoe Cloud bookings during the year to date.

There is a catch with the $140 billion headline. That figure refers to total contracted value across different agreements and periods. It does not represent annual revenue, nor has the company already collected the full amount.

Can Nscale actually catch Crusoe?

Nscale has become one of Crusoe's most serious competitors, but we would still put it behind Crusoe today because much of its biggest announced AI infrastructure capacity remains to be delivered.

Nscale raised $2 billion in March 2026 at a $14.6 billion valuation. Shortly afterward, it announced the acquisition of the Monarch Compute Campus in West Virginia and a letter of intent with Microsoft covering up to 1.35 GW of AI compute capacity using NVIDIA Vera Rubin systems.

The company's expansion also reaches Europe. In April 2026, Nscale announced an enlarged Microsoft agreement involving more than 30,000 additional NVIDIA Rubin GPUs at its Narvik campus in Norway, associated with a 230 MW development. Nscale pushing into Norway with a bigger Microsoft deal is fast growth, and we track who is growing the fastest in our data center market report.

A 1.35 GW project would represent an enormous computing installation, and the Norwegian development offers access to a different electricity market from the largest U.S. AI campuses. But Nscale's announced projects are at different stages. A letter of intent does not guarantee that every planned GPU will be installed, powered and generating revenue.

Crusoe has already disclosed 1 GW of operational gross capacity, which gives it a stronger record of delivery. Nscale could narrow the gap considerably as its Microsoft projects come online.

Is Lambda still one of the top AI data center startups?

Lambda remains one of the leading private GPU cloud companies, and its recent financing shows that major lenders are increasingly comfortable backing its computing infrastructure.

Lambda originally became known for providing NVIDIA GPU access to developers and researchers. It has since moved toward much larger computing deployments, including dedicated infrastructure for companies training and running advanced AI models.

Its financing has grown accordingly. In May 2026, Lambda announced a $1 billion senior secured credit facility. It followed with a $926 million senior secured term loan in August and another $1 billion fixed-rate financing in October.

These are separate debt facilities rather than venture funding rounds. They allow Lambda to finance infrastructure and computing equipment against assets and commercial commitments.

The Wall Street Journal reported in October 2026 that Lambda was pursuing a potential $4 billion financing ahead of a planned 2027 IPO. The report described a proposed valuation of approximately $14.5 billion before the new investment and said the company's backlog had increased from $15 billion in June to $50 billion in September, helped by an Anthropic agreement.

Those reported figures suggest substantial commercial demand, although the proposed financing should not be treated as completed and backlog should not be confused with revenue already earned.

Is Fluidstack becoming a serious rival to Lambda and Crusoe?

Fluidstack has become a serious private AI infrastructure contender, although we still lack enough public financial and operating data to put it ahead of Lambda or Crusoe.

Fluidstack announced an $830 million Series A at a $7.5 billion valuation in January 2026, led by Situational Awareness. That is an unusually large financing for a Series A and reflects how much investors now value companies capable of supplying enormous GPU clusters. A $7.5 billion valuation at Series A shows how fast values can jump; we compare what each company is worth, then and now, in our data centers market report.

The company focuses on computing infrastructure for advanced AI laboratories. It competes in an increasingly busy market where developers may obtain capacity from hyperscalers, specialist GPU clouds or infrastructure providers building facilities specifically around large AI workloads.

The relationships between these companies can also be complicated. IREN, for example, has named Fluidstack among its customers. That means two businesses competing for AI infrastructure spending can simultaneously appear at different levels of the same supply chain.

Fluidstack's funding is impressive, but its publicly available information does not yet support a detailed comparison of revenue, owned infrastructure, operational capacity and profitability with more established competitors. For now, we would put it behind Lambda.

Which private companies are actually building the data centers?

Vantage Data Centers and DataBank are among the strongest privately held physical data center operators, while Aligned Data Centers shows just how valuable a successful developer can become.

Vantage operates a global hyperscale development platform and has continued securing capital for expansion. In September 2026, it announced a $2 billion development financing facility for its North American business.

DataBank provides a particularly concrete example. In April 2026, it secured $2 billion in construction financing for three already leased buildings at its Red Oak campus near Dallas. Together, those buildings account for approximately 180 MW of capacity and 600,000 square feet.

In June, DataBank added another $650 million to the financing package, bringing the total to $2.65 billion and supporting a fourth building with an additional 60 MW.

The distinction between leased and speculative capacity is important. DataBank's financing supports facilities with commercial commitments already attached, which makes it easier for lenders to assess how the investment will generate income.

Aligned reached an even larger scale before its acquisition. The transaction involved 51 campuses and more than 6.4 GW of operational and planned capacity.

These businesses are more mature than most venture-backed startups. Still, when the question is which private companies can actually construct and operate major facilities, they deserve more attention than many newer businesses with higher-profile AI branding.

Does Aligned Data Centers' $40 billion sale mean data center startups are worth a fortune?

Aligned Data Centers' $40 billion acquisition shows that large private data center platforms can command extraordinary prices, but we cannot apply that transaction value directly to smaller startups.

Aligned's portfolio included more than 6.4 GW of operational and planned capacity across 51 campuses. Using the approximate $40 billion enterprise value gives a rough figure of $6.25 million per megawatt.

That number needs careful interpretation. The portfolio included facilities at different development stages, and the purchase price covered an entire operating business with land, infrastructure, customer relationships and development capabilities. It was also an enterprise value, which differs from the equity valuations commonly reported for venture-backed companies.

We cannot therefore use the same ratio to estimate what Crusoe or Nscale should be worth. Their businesses include different combinations of campus development, GPU infrastructure and computing services.

Are CoreWeave, Nebius and IREN already ahead of the private startups?

CoreWeave, Nebius and IREN are among the most important public competitors in AI infrastructure, and their financial disclosures give us a better way to judge the promises made by private data center startups.

IREN is particularly interesting because its business has been changing so quickly. In August 2026, the company reported fiscal-year revenue of $707 million, including $128.8 million from AI cloud services. A business changing this quickly is why we go through what changed in the last 12 months in our report on data centers.

Its AI cloud revenue had increased from $16.4 million the previous fiscal year, representing nearly eightfold growth. However, Bitcoin mining still accounted for most of its reported revenue during the period.

IREN also disclosed approximately $4 billion in contracted annualized run-rate revenue linked to its 2026 capacity, compared with approximately $1 billion of operating ARR at the time.

That difference is revealing. Around three-quarters of the contracted annualized run rate had yet to become operational at the stated measurement point. The contracts were substantial, but customers could only be billed at the anticipated rate once the necessary infrastructure was available.

CoreWeave and Nebius provide further public-market comparisons because their revenue, financing and operating results can be examined through regulatory filings.

Does having more gigawatts mean a data center startup is winning?

The number of gigawatts a data center startup announces can be highly misleading because planned power capacity and operational computing capacity are often very different.

Crusoe reported more than 6 GW of gross contracted capacity alongside 1 GW delivered and operational. On its reported gross basis, that means the operational footprint represented approximately one-sixth of its contracted capacity.

IREN disclosed a development pipeline exceeding 5 GW, while targeting cumulative critical IT capacity of about 0.3 GW by the end of 2026 and 0.8 GW by the end of 2027. Those smaller figures represent a different stage and definition of infrastructure development.

Nscale's West Virginia project offers another example. The broader site has been described as having potential for more than 8 GW of microgrid power, while the initial Microsoft letter of intent concerned up to 1.35 GW of AI compute capacity.

The figures cannot simply be added together or compared as though every company uses the same definition. Site power, facility capacity, critical IT load and deployed GPU power can represent different things.

We also need to distinguish capacity supplied by equipment vendors from capacity operated by cloud providers. Submer's reported liquid-cooling deployments, for example, do not mean it owns all the data centers using its systems.

Capacity measure	What it tells us	What it doesn't prove
Land capacity	Space available for future construction	Electricity is secured
Power pipeline	Potential access to electricity	Power will arrive on schedule
Contracted capacity	Capacity committed under agreements	Infrastructure is operational
Gross facility capacity	Broad facility or site power scale	Equivalent usable GPU power
Critical IT capacity	Power allocated to computing equipment	All capacity is occupied
Energized capacity	Electrical systems are powered	Customers are running workloads
Operational capacity	Infrastructure is functioning	High utilization or profitability
Revenue-producing capacity	Paying workloads are being served	Attractive margins

Are Microsoft and other AI giants making these data center startups successful?

Microsoft and other major AI customers are helping private data center startups grow much faster by signing large contracts that can support billions in infrastructure financing.

Microsoft has become a particularly important customer across the sector. Nscale has announced major projects linked to Microsoft in West Virginia and Norway, while IREN signed a GPU cloud agreement with Microsoft worth approximately $9.7 billion over five years.

Lambda's reported backlog expansion has also been closely linked to a major Anthropic agreement. Leading AI developers and hyperscalers are increasingly using specialist infrastructure providers to obtain large amounts of computing capacity.

These contracts can make enormous projects financially possible. Banks are more willing to finance GPU purchases when a creditworthy customer has already committed to using the equipment.

IREN disclosed a particularly interesting example. Its Microsoft-related financing package amounted to $3.6 billion at a 6% rate. The company indicated that the financing and customer prepayments together covered approximately 96% of the associated GPU capital expenditure.

Customer concentration also deserves attention. If one or two AI developers account for a large share of a provider's future business, changes in those customers' spending plans can have major consequences.

Are AI data center startups becoming energy companies too?

Crusoe and Nscale are increasingly building their strategies around electricity because power availability has become one of the biggest constraints on large AI data center projects.

Crusoe has long connected energy development with computing infrastructure. Its business has evolved into large AI campuses where power access, cooling and computing equipment must be planned together.

Nscale is following a related approach. Its acquisition of the Monarch Compute Campus in West Virginia and the creation of Nscale Energy & Power show how directly it wants to participate in energy infrastructure. Companies that also build their own energy capacity are chasing a new opening, and we map where the biggest opportunities are in our data center report.

The West Virginia site has been presented as having a potential microgrid power runway exceeding 8 GW. That is an enormous development ambition, although the capacity will depend on construction, financing, infrastructure delivery and customer demand.

The change also creates opportunities for electrical equipment startups. In October 2026, Vantage announced an agreement reserving more than 1 GW of manufacturing capacity for Heron Power Electronics' solid-state transformer technology, supporting next-generation 800-volt direct-current infrastructure.

A shortage of transformers, switchgear or utility connections can delay a project even when the developer already has land and GPUs.

Which data center cooling startups are doing the best right now?

ZutaCore is currently our strongest independent cooling technology contender, while Submer leads a broader infrastructure strategy. Accelsius, Iceotope and Corintis are promising alternatives at different stages of development.

ZutaCore announced a $100 million Series C in June 2026 with backing from Mitsubishi Electric, Carrier Ventures and Samsung Ventures. The company also reported more than 75 deployments worldwide.

Its technology uses two-phase direct-to-chip cooling, allowing a dielectric fluid to absorb heat close to processors. It targets powerful AI chips that are becoming increasingly difficult to cool efficiently with conventional approaches.

Submer has historically focused on immersion cooling, placing computing equipment in specially designed systems using dielectric liquids. The company has reported a deployment base exceeding 500 MW of liquid-cooled infrastructure and expanded into wider AI data center services.

Among newer competitors, Accelsius raised $65 million in a Series B announced in January 2026. Iceotope secured $26 million in a Series B in May, while Corintis raised $25 million in late 2025 to advance microfluidic cooling technology.

These companies are pursuing different technical approaches, so comparing them purely by funding would be misleading. ZutaCore has demonstrated a combination of commercial deployments and strong industrial relationships. Submer has a broader position in infrastructure, while Corintis is still developing a more specialized opportunity around chip-level cooling.

Could cooling startups become more valuable than the data center operators?

Cooling startups could build highly valuable businesses with much less capital than the companies constructing entire AI campuses, although the largest physical infrastructure operators still have a substantial advantage in absolute scale.

Consider the financing difference. ZutaCore's $100 million Series C was enormous for a cooling specialist, yet Crusoe raised 39 times that amount in its $3.9 billion Series F.

A cooling supplier can sell equipment into facilities developed by multiple operators. It generally does not need to purchase the GPUs, build the entire campus or secure all the power required to operate the customer's infrastructure.

Cooling suppliers also face large established industrial competitors. Vertiv, Schneider Electric and other major manufacturers have global distribution networks, production capabilities and long-standing customer relationships.

There is already evidence that established companies want to acquire specialist technology rather than develop everything internally. Trane Technologies acquired LiquidStack, Flex acquired JetCool, and Daikin Applied acquired Chilldyne.

These deals give smaller companies a potential route to a valuable exit even if they never become independent billion-dollar platforms.

Can smaller data center startups really compete with Amazon, Microsoft and Google?

Specialist data center startups are already winning major contracts involving Microsoft and leading AI developers, but we expect most of them to succeed as infrastructure suppliers rather than become direct replacements for AWS, Azure or Google Cloud.

Microsoft's agreements with IREN and Nscale demonstrate that even hyperscalers with enormous infrastructure budgets sometimes buy capacity from outside providers.

Lambda, Fluidstack and Crusoe compete for related opportunities by supplying computing infrastructure designed for demanding AI workloads. Their customers may need very large, dedicated GPU clusters rather than the full range of services available through a traditional cloud platform.

For these buyers, delivery speed, hardware availability and the ability to operate thousands of GPUs together can matter more than having a broad product catalog.

However, major cloud providers enjoy advantages that smaller competitors cannot easily match. AWS, Azure and Google Cloud have global customer relationships, distribution networks, software ecosystems and financial resources.

The startups often rely on the same hyperscalers as customers, while purchasing much of their computing hardware from NVIDIA and borrowing from large financial institutions.

Are data center startup valuations getting too high?

Some AI data center startups are now valued on years of expected growth, and the biggest risk is that construction delays, hardware costs or weaker margins prevent them from meeting those expectations.

Crusoe's valuation reflects the potential of its large contracted infrastructure business. Nscale's valuation depends partly on major projects that still need to be delivered. Fluidstack's $7.5 billion valuation is difficult to assess independently because detailed revenue and profitability figures are not widely available.

The public market provides a useful reminder of how complicated the financial results can become. IREN reported $707 million in fiscal-year revenue but also a $702.6 million net loss, substantially affected by noncash impairments associated with its transition from Bitcoin mining toward AI infrastructure. Moving from Bitcoin mining to AI infrastructure is a model bet; we compare which business models are winning in our data centers report.

Financing costs make the picture even more demanding. IREN has disclosed major infrastructure financing packages with interest rates of 6% and 9%.

For every $1 billion borrowed, that difference represents $30 million in additional annual interest expenses, before principal repayments. A company with several billion dollars of debt can therefore see a meaningful part of its operating cash flow absorbed by financing costs.

There is also the issue of GPU depreciation. Newer processors arrive regularly, and operators have to recover their investments over the useful commercial life of each generation.

So, what are the top data center startups today?

Crusoe is our number-one private AI data center startup today, followed by Nscale, Lambda and Fluidstack. Vantage and DataBank are among the strongest private physical developers, while ZutaCore and Submer lead our selection of specialist cooling companies.

Crusoe earns first place because it combines major customer commitments with a substantial reported operational footprint. Nscale could become a much closer competitor as its Microsoft projects are completed, while Lambda has built a strong position through its GPU cloud business and access to financing.

Fluidstack is also worth watching closely. Its enormous early-stage financing and exposure to leading AI customers make it a serious contender, but the limited public information about its operating results prevents us from ranking it higher with the same confidence.

Vantage and DataBank bring a different kind of strength through proven data center development and financing. They are more mature private operators, but their experience building and leasing facilities makes them important benchmarks.

In cooling, ZutaCore currently offers one of the more convincing combinations of funding, industry partnerships and deployments. Submer has broader infrastructure exposure, while Accelsius and Corintis are younger companies with interesting technologies.

Our final ranking reflects the evidence available today rather than assuming every announced project will be completed.

Position	Company	Why it ranks here	Main issue to watch
1	Crusoe	Large operational footprint and contracted demand	Profitability and backlog delivery
2	Nscale	Major projects and Microsoft relationships	Converting plans into live infrastructure
3	Lambda	Established GPU cloud business and financing access	Borrowing costs and hardware economics
4	Fluidstack	Major funding and AI laboratory demand	Limited financial transparency
5	Vantage Data Centers	Proven global data center development	Capital-intensive expansion
6	DataBank	Financed and leased U.S. campus developments	Construction costs and delivery
7	Submer	Broad cooling and infrastructure activities	Separating existing deployments from future projects
8	ZutaCore	Industrial partnerships and commercial cooling deployments	Wider technology adoption
9	Accelsius	Strong financing and two-phase cooling technology	Competing with established suppliers
10	Corintis	Promising microfluidic chip cooling	Demonstrating commercial scale

The ranking could change considerably as new AI campuses come online. Crusoe is ahead on the evidence available now, but the next few years will reveal which companies can repeatedly complete large projects, keep expensive equipment busy and earn attractive returns.