Modular data centers: which startup is ahead?

Which startups actually belong in the modular data center race?

Seven startups deserve a place in this comparison, although only Crusoe and Armada currently look like broad category leaders.

We include companies that build complete modular data centers, operate modular facilities, or supply enough of the infrastructure to compete for the same AI projects. That gives us Crusoe, Armada, Nautilus Data Technologies, Submer, ECL, Flexnode and Metrobloks.

The group is not uniform. Crusoe and Armada sell integrated computing infrastructure. Nautilus and Submer built their positions around cooling before moving closer to complete data center delivery; Submer now has a dedicated modular arm, Corenix. ECL combines modular buildings with its own energy architecture. Flexnode sells prefabricated AI data halls, while Metrobloks develops permanent facilities from repeatable modular designs.

We leave out Vertiv, Schneider Electric, Eaton, ABB, Huawei and Rittal because they are established industrial groups. They remain serious competitors. Vertiv sells prefabricated systems from small modules to 50 MW campuses, and Schneider Electric can deliver complete data center pods through a global factory network.

We also exclude EPG from the startup ranking. It was founded in 2004, runs large manufacturing hubs in Asia and has supplied customers including GDS and Huawei. It works better as a private industrial benchmark.

Funding figures need care. Crusoe’s money supports cloud computing, energy projects and hyperscale campuses alongside Spark, so its total says little about Spark alone. Submer’s total varies between databases because some count convertible instruments and debt differently. Flexnode has raised at least $38.9 million across its disclosed seed and Series A rounds.

Startup	What it currently does	Funding raised
Crusoe	Builds and operates AI infrastructure, including Spark modular AI factories	$3.9B Series F (initial close, September 2026) at a $30.9B post-money valuation, on top of earlier equity and debt for the wider company
Armada	Builds rugged Galleon units, megawatt-scale Leviathans and the 10 MW Orion	Nearly $500M (company-stated), including a $230M Series B at a $2B pre-money valuation
Nautilus Data Technologies	Supplies prefabricated liquid-cooling infrastructure and operates a liquid-cooled facility	About $223M
Submer	Builds immersion-cooling systems and, through Corenix, Rubix and Radian Arc, modular data centers, campuses and GPU cloud	About $102M, with some databases reporting lower totals
Flexnode	Builds prefabricated high-density AI data halls with integrated power and cooling	At least $38.9M
Metrobloks	Develops modular, liquid-cooled data centers in urban and regional markets	About $20.8M
ECL	Builds modular AI data centers powered through grid, gas, batteries and hydrogen fuel cells	$17M

Is one modular data center startup clearly ahead today?

Crusoe is ahead in modular data centers today, and Armada is close enough to keep the race open.

Crusoe has more operating history, far more capital and a much larger infrastructure business behind its modular product. It says it has deployed more than 400 earlier modular units at energy sites. Its newer Spark system has a live customer site that has moved beyond its first trial, and Ollama now runs part of its cloud on Spark modules.

The wider company also grew sharply. When it announced its Series F in September 2026, Crusoe said it had more than 6 GW of gross contracted capacity and 1 GW delivered and operational. Most of that is hyperscale campus capacity, not Spark. It still means Crusoe can fill new modular units with its own cloud demand.

Armada has built a more focused modular company. Its Galleons have worked in Saudi industrial operations, on a U.S. Navy warship and on an offshore Aker BP installation. In August 2026, it added a firm Leviathan order from Fossefall and launched Orion, a 10 MW unit. In October 2026, Palantir named it its first certified modular data center partner.

The gap is proof of scale. Crusoe has disclosed operating units, a repeat customer expansion, more than $200 million committed to a factory and its first fleet, and a next-generation module on its roadmap. Armada has strong customers and a dated order book, but it still keeps its installed unit count and operating megawatts private.

Nautilus, Submer and ECL lead narrower parts of the market. Flexnode and Metrobloks remain further back: neither has publicly identified a commissioned commercial facility built on its current platform.

Who has actually put modular data centers into the field?

Crusoe has deployed the most modular infrastructure, while Nautilus, ECL and Armada give the clearest evidence that their systems handle real customer workloads.

Crusoe’s figure of more than 400 deployed modules gives it the deepest field experience, although most of those units came before Spark and served energy-site computing. The number shows operating experience, not 400 identical modern AI factories. Hundreds of units in the field are a real head start, and we line up who the top companies are today in our data center market report.

The more useful Spark example sits at Redwood Materials in Nevada. Four units began operating beside a solar and second-life battery microgrid in 2025. Crusoe and Redwood later agreed to expand the site to 24 units and nearly seven times the original computing capacity.

Nautilus has a smaller footprint with a longer facility-level record. Its Stockton site provides 6.5 MW across four vaults and reports a PUE of 1.15. The company cites more than 500,000 operating unit-hours, although one section of its own website still says 400,000, so the exact figure is unclear. Nautilus has also delivered 12 MW of cooling capacity to Start Campus in Portugal.

ECL’s first commercial site is only about 1 MW, but it is fully leased and has hosted Lambda’s production NVIDIA systems. Armada’s smaller Galleons work in industrial, military and maritime conditions. Its first disclosed Leviathan order is due for delivery by early 2027, and it has published no total for live units or megawatts.

Startup	Strongest operating evidence	What remains unclear
Crusoe	400+ earlier modular units and a live Spark installation expanding from four units to 24	How many earlier units are comparable with Spark, and how many Spark units its factory has produced
Armada	Deployments with Aramco Digital, Microsoft, the U.S. Navy and offshore energy, plus a firm order for five Leviathans	Total operating units, installed MW and any live Leviathan capacity
Nautilus	A 6.5 MW facility, 500,000+ operating unit-hours (company-stated) and a 12 MW cooling delivery	Repeat sales of complete modular facilities beyond cooling systems
ECL	A fully leased 1 MW modular facility running production AI equipment	Replication across several sites and much larger capacities
Submer	More than 500 MW of liquid-cooled infrastructure deployed (company-stated) and a delivered 100 kW MegaPod	Named customers for Corenix and operation of the campuses it is announcing
Flexnode	No named commissioned customer facility found	Whether the Eaton-backed product performs as planned in live use
Metrobloks	Several development sites and partnerships	The company’s first completed and operating data center

Who has turned modular data centers into real customer demand?

Crusoe has the strongest evidence of repeat demand, while Armada now has both the most impressive mix of strategic customers and a disclosed megawatt-scale order.

Crusoe’s work with Redwood matters because the customer used the first four units before committing to another 20. That sequence tells us more than a large first announcement. The site is growing after real operation, on the same energy platform and location. In October 2026, Crusoe also named Ollama as a customer running dedicated workloads on solar- and battery-powered Spark modules, without saying how many.

Crusoe also signed a framework with Energy Vault for phased Spark deployments of up to 25 MW. A framework is a ceiling, not an order, so we discount it until units are commissioned. In July 2026, Crusoe and Aalo Atomics also agreed that an Aalo reactor would power a Spark unit as a proof of concept in 2027. That is a plan, not demand we can count yet.

Armada’s customer list is broader. Aramco Digital and Microsoft used its systems in Saudi Arabia. The U.S. Navy received a Galleon and tested one at sea. Aker BP agreed to deploy a unit offshore. In August 2026, Fossefall ordered five Leviathans totaling more than 9 MW, worth several hundred million Norwegian kroner, to be built in Dublin and delivered by the first quarter of 2027.

The Fossefall order changes the quality of Armada’s evidence. It has a size, a value range and a delivery date. Armada also names WinDC in Australia as a Leviathan customer, without a unit count. What it still lacks is a repeat order after operation, the kind of proof Crusoe has at Redwood.

Nautilus gives cleaner capacity figures. Its Stockton facility was reported as 86% leased when it was offered for sale, and Cerebras had signed for 2.5 MW. Start Campus deployed 12 cooling distribution units representing 12 MW.

ECL has the strongest demand relative to its size. Its 1 MW MV1 site reached full occupancy, and Lambda later doubled its footprint there. A company with $17 million in disclosed funding has persuaded customers to run dense production hardware on its infrastructure.

Which modular data center startup is growing fastest now?

Armada reports the fastest commercial growth, while Crusoe shows the physical expansion we can actually compare.

Armada said customer bookings grew 540% between fiscal 2025 and fiscal 2026, then rose 2,000% year over year in the first quarter of fiscal 2027. It has not published the dollar values behind those percentages. Percentages with no dollar figures behind them say little, so we gauge how far the challengers are behind in our report on data centers.

A company moving from $1 million to $21 million would also report 2,000% growth, so the starting point matters. The Fossefall order gives one concrete anchor: several hundred million kroner for more than 9 MW. The percentages still do not tell us total revenue or installed capacity.

Crusoe’s clearest modular growth measure is the planned sixfold rise in Spark units at Redwood, with computing capacity expected to rise almost sevenfold because the newer units are denser. The wider company is growing much faster: Crusoe said in September 2026 that Crusoe Cloud bookings had grown more than 20 times year over year so far in 2026. That figure covers its whole cloud, not Spark.

Submer is growing by adding layers. In 2026 it acquired Radian Arc, launched Rubix Data Centers with more than 8 GW of powered land, committed more than €1 billion to an AI campus in Flix, Spain, and in September launched Corenix to build factory-tested modules. In October 2026, Submer India signed a memorandum with BSNL’s tower company to build AI-capable sites.

That breadth could reshape the ranking over time. Most Rubix capacity is still a development pipeline, and memoranda are not orders, so we would not place Submer above the operating leaders yet.

Whose modular data center product is ready to buy and deploy?

Crusoe has the most mature complete AI product, while Armada is the strongest option for systems that must work in remote or difficult places.

Crusoe says a Spark installation can be delivered in as little as three months. Each unit arrives with power distribution, cooling, fire protection, networking and GPU-ready racks. The Redwood site shows the design can run beside a non-traditional power system. In October 2026, Crusoe said a multi-unit Spark cluster in Nevada with NVIDIA HGX B300 systems met NVIDIA’s Exemplar Cloud training targets, with per-GPU throughput on par with its conventional data centers. Crusoe ran those tests itself, and no rival module has been measured the same way.

The next Crusoe product is not on sale yet. Spark Liquid, unveiled in October 2026, is designed for up to 3.5 MW per unit, about five times the density of earlier generations according to Crusoe, and targets NVIDIA GB300 and Vera Rubin racks. It joins Brighton production in 2027, so today it is a specification, not a deployed system.

Armada says a Galleon can be deployed in about 60 days, and its public deployments support that for smaller edge systems. Its range now runs from rugged edge units to the Leviathan, marketed with a 1.77 MW IT envelope per unit, and the 10 MW Orion, launched in August 2026, which supports up to 2,880 GPUs across 40 racks.

The larger products are less proven. Armada says Orion units are deliverable “in months from deposit”, but has named no Orion customer. Leviathan has an order but, as of now, no named operating site.

Nautilus offers a mature cooling product rather than a complete computing platform. It quotes delivery periods of roughly 12 to 16 weeks for some cooling distribution units, much shorter than waits for many conventional components.

Flexnode says its prefabricated process cuts schedules by an average of 35%. Eaton supplies power backup, racks, cabling and 800-volt direct-current equipment for its NX Compute Module. That makes the product more credible, but Flexnode has not published the projects behind the 35% figure.

ECL has proven its design at 1 MW. The next test is harder: repeating it across the planned 35 MW CSC-1 campus in Santa Clara while coordinating utility power, gas generation, batteries and hydrogen fuel cells.

How much are Crusoe’s and Armada’s modular factories producing?

Nobody outside the companies knows: Crusoe says its Spark factory is producing, Armada has not confirmed its line is running, and neither has published a volume.

Crusoe’s Spark Factory in Brighton, Colorado covers 352,000 square feet. The company committed more than $200 million to the site and its first fleet, and said in March 2026 that it expected the first factory-produced modules in the third quarter of 2026. That quarter has ended without a dated first delivery. In October 2026, Crusoe referred to existing Spark production at Brighton, while other October coverage described mass production as still to come. Neither gives a unit count.

Crusoe does have real manufacturing behind it. In September 2026 it opened a second plant in Tulsa, Oklahoma, adding 400,000 square feet for switchgear, electrical enclosures and busbar. That brings its national manufacturing footprint to roughly 1 million square feet. These plants make electrical equipment, not Spark units, but they cut its dependence on scarce suppliers.

Armada and Johnson Controls planned continuous production from summer 2026 at Galleon Forge One in Arizona, a site of up to 400,000 square feet expected to create about 500 jobs. Armada has not said whether that line is running. The lack of news does not prove a delay, but the summer target has passed without confirmation. Its Fossefall release says those five Leviathans will be built in Dublin, which suggests a second manufacturing base in Europe.

So the comparison still rests on statements, not counts. Crusoe’s factory is slightly smaller but is backed by hundreds of earlier modules, an in-house electrical supply chain and a 2027 product to add to the line. Armada’s is slightly larger and has a dated order to fill. Whoever first discloses steady output will hold the stronger claim.

Flexnode follows an asset-light path. Eaton supplies critical electrical equipment and manufacturing support, letting Flexnode offer data halls from 3.5 MW to 35 MW without building every part itself.

Submer has moved into this race through Corenix, led by a former Johnson Controls executive. It builds and factory-tests rack-ready modules at production sites it has not named. Nautilus relies on manufacturing partners, which shortens cooling deliveries but not full data center projects.

Metrobloks sits outside the factory race. Its proposed Liberty, Missouri campus would provide 150 MW across three permanent buildings, with other sites planned in several U.S. metros. It is scaling a development pipeline, not shipping standard units from a line.

Who has the best cooling and energy efficiency?

ECL reports the lowest PUE in an operating modular AI data center, while Nautilus has the more convincing long-term operating record.

ECL reports a PUE of about 1.1 at MV1, and its website claims “under 1.1”. Lambda has used the site for NVIDIA GB300 NVL72 systems at reported rack densities of about 142 kW, so the figure comes from a facility handling dense AI equipment. A PUE near 1.1 in dense AI racks is a strong result, and we follow which technologies are gaining traction in our data centers report.

Nautilus reports a PUE of 1.15 at Stockton that it says holds in every season. It also cites a long operating history and uses no consumptive water for cooling. That gives it a stronger reliability case, even with a slightly higher published PUE.

The gap with a conventional new facility is real. At 1 MW of computing load, a PUE of 1.3 needs about 300 kW for cooling, power conversion and other overhead. A PUE of 1.15 cuts that to 150 kW. ECL’s reported 1.1 would bring it to roughly 100 kW.

These figures come from different sites, climates and power systems, so this is not a controlled test. ECL has the better published number. Nautilus has more operating evidence behind its claim.

Submer remains the strongest specialist in immersion cooling. Its technology has been certified in a combination with Intel, Shell and Supermicro, and deployed with companies including Castrol and Macquarie Data Centres. Some Submer materials cite PUE as low as 1.03, a product-level claim rather than an audited campus result.

The two leaders are moving into this contest with new hardware. Crusoe says Spark Liquid will use closed-loop, non-evaporative direct liquid cooling that consumes no water. Armada markets Leviathan on a 1.77 MW IT envelope. Neither has published a measured PUE from a customer site, so both remain design claims.

Which modular data center startup offers customers the best economics?

Crusoe has the strongest economic case for urgent AI capacity, even though no startup publishes enough standard pricing for a clean cost-per-megawatt comparison.

Crusoe’s main advantage is speed to revenue. A customer can place Spark units beside an available energy source and start using expensive GPUs without waiting years for a traditional campus and a full grid connection.

That timing can outweigh a modest difference in construction cost. A delayed GPU cluster earns nothing while the hardware loses value. Crusoe also controls the energy relationship, the data center and the cloud service, which cuts the number of companies needed to bring a site online.

The speed argument now faces a standard rival. In October 2026, NVIDIA said DSX POD containers could cut installation at a prepared site from about six months to as little as 14 days, with partner deliveries planned for early 2027. That is a design target, not an observed result, and it covers installation only, not finding power or preparing land.

Armada now offers the first public price signal in this group. Fossefall’s order for five Leviathans totaling more than 9 MW is worth several hundred million kroner. The range is too wide for a precise cost per megawatt, and it may include more than the modules, so we treat it as an order of magnitude only.

Nautilus has the clearest operating-cost argument. A PUE of 1.15 implies about half the facility overhead of a site at 1.3. Zero-water cooling may also lower costs or remove a permitting obstacle where water use is restricted.

ECL could save customers from long grid-interconnection waits. Its FlexGrid design combines several energy sources on one site. The economics of hydrogen remain hard to judge because ECL does not disclose fuel, storage, conversion and maintenance costs.

A low PUE only measures how efficiently a facility uses the electricity it gets. It says little about what that electricity costs. ECL must show that hydrogen improves the full customer bill as it moves from one megawatt to tens of megawatts.

Armada’s strongest case is still remote sites, where a customer avoids sending raw data over weak networks and avoids building a permanent facility. The value depends on the workload and the location, which makes a universal price comparison unrealistic.

Can Crusoe and Armada stay ahead once NVIDIA’s DSX POD standardizes modules?

Yes, if their lead rests on more than the box: Crusoe has the hardest overall system to reproduce, and Armada the strongest specialist position in remote and sovereign computing.

Vertiv, Schneider Electric and other industrial groups already know how to manufacture prefabricated power, cooling and computing infrastructure. NVIDIA’s DSX POD, launched in October 2026, gives them a shared blueprint: its partner list includes Crusoe and Armada, but also Vertiv, Schneider Electric, Eaton, Dell, HPE and Foxconn. A startup needs an advantage beyond the box.

Crusoe combines energy development, modular manufacturing, its own electrical equipment plants, large data center construction and a cloud business. It can find power, build the computing environment and sell access through Crusoe Cloud, which signed a multi-year partnership with Perplexity in September 2026.

A competitor could build a similar module. Matching the energy relationships, operating experience, factories, cloud demand and contracts together would take much longer. Crusoe’s wider business strengthens Spark rather than distracting from it.

Armada combines rugged hardware with its own management platform, satellite connectivity and Azure Local integration. In October 2026, Palantir named Armada its first certified modular data center partner and said it will validate its sovereign AI software on Galleons, including fully air-gapped setups.

That Palantir link is a software and channel advantage, not yet a deployment. It gives Armada a route into government and defense buyers that most hardware makers lack. Large equipment makers can build a strong box, but not easily this mix of software, certification and field experience.

Nautilus has patents and years of operating data in facility-scale liquid cooling. Submer has immersion-cooling expertise, certified hardware and a growing ecosystem. Both face several well-funded incumbents in cooling.

ECL’s edge comes from how it combines and manages energy sources on one platform. Core fuel-cell hardware comes from partners such as PowerCell and Bosch, so its durable value depends on project execution and energy orchestration.

Does Crusoe’s $3.9 billion round translate into modular results?

Partly: Crusoe has turned the most capital into operating scale, but most of its new money goes to campuses and cloud, and ECL still gets the most physical proof per dollar.

Crusoe said in September 2026 that it had closed $3.9 billion as the initial close of its Series F, at a $30.9 billion post-money valuation. Bloomberg had reported earlier that month that the round was above $3 billion at roughly $30 billion; the company’s own figures are slightly higher. That single round is close to eight times Armada’s total of nearly $500 million. One round that dwarfs a rival's whole funding history shows how lopsided the money is; we map who is getting funded, and by whom, in our report covering data centers.

The totals are not directly comparable. Crusoe uses capital for cloud hardware, power projects and hyperscale campuses alongside Spark. Armada’s money is focused on modular products and its factory. Nautilus and Submer spend more on cooling, Metrobloks on site development, and ECL will need project financing for large campuses.

Crusoe can still point to the most output: hundreds of earlier modules, a repeat Spark customer, a funded factory, 1 GW of operating capacity across the company and a cloud business able to fill new units. What it has not shown is how much of the new money has turned into Spark units.

Armada has turned about half a billion dollars into credible deployments, a software platform, three modular product sizes, a factory partnership and a firm order. Its efficiency stays hard to measure without revenue, unit volumes or installed capacity.

ECL has done something unusual with $17 million. It opened a fully leased 1 MW facility that runs production AI hardware. That does not prove it can finance and deliver a 35 MW campus, but it gives ECL more operating credibility than its funding suggests.

Flexnode and Metrobloks have raised more than ECL while disclosing less live infrastructure. Their partnerships and pipelines may justify the spending later. Today, ECL has more physical proof per disclosed venture dollar.

Is Armada now gaining on Crusoe in modular data centers?

Yes, Armada has had the strongest recent run in modular products and orders, while Crusoe’s momentum is bigger but mostly sits outside Spark.

Armada’s run since August 2026 is the most modular-specific in the group. It received a firm order for five Leviathans from Fossefall, launched the 10 MW Orion and won Palantir’s first modular data center certification. The next milestone is plain: Leviathans delivered on time and running for customers.

Crusoe’s momentum is larger but broader. It raised $3.9 billion, reported more than 6 GW of gross contracted capacity, opened a second Tulsa plant and signed a 1 GW campus partnership with Lancium in Childress, Texas. On the modular side, its new steps are the B300 Spark training results, the 3.5 MW Spark Liquid design for 2027 and the Aalo plan to power a Spark unit with a small reactor in 2027.

Submer keeps changing shape. After Radian Arc, Rubix and the €1 billion Flix project, it launched Corenix in September 2026 and folded InferX into Radian Arc. Construction and customer commitments will decide whether this becomes an operating advantage.

ECL’s progress is more concrete. In July 2026, PowerCell announced a firm purchase order for fuel-cell systems for ECL’s 35 MW CSC-1 campus in Santa Clara, plus a non-binding memorandum covering about 300 MW of possible future capacity.

Metrobloks has a proposed 150 MW campus in Missouri, a joint venture with Lincoln Property Company and a potential 100 MW project with Soluna. Its pipeline grew fast in early 2026, while its completed capacity remains at zero.

Can we trust the modular data center numbers?

Operating megawatts, repeat deployments and firm orders deserve the most weight; bookings growth, planned gigawatts and memoranda tell us much less on their own.

The strongest evidence involves equipment that can be located and customers that can be named. Nautilus’ Stockton facility, ECL’s MV1 site, Armada’s Navy delivery and Crusoe’s Redwood units all pass that test.

A repeat order carries even more weight because the customer has seen the product work. Crusoe’s Redwood expansion therefore counts for more than Armada’s Fossefall order, which is a first order with delivery still ahead.

Firm orders sit one level below operating infrastructure. Fossefall’s Leviathans and ECL’s PowerCell order both qualify. ECL’s 300 MW memorandum and Submer’s Indian memoranda describe possible future business and deserve a much larger discount.

Development pipelines can be real assets when the company controls land and power. They can also take years to finance, permit, pre-lease and build. Submer’s 8 GW Rubix portfolio and Metrobloks’ 150 MW Liberty project show ambition and site access, not commissioned capacity.

Growth percentages have a different problem. Armada’s 540% and 2,000% and Crusoe’s 20-fold cloud bookings growth sound enormous, but neither company gives the base. We use them for direction, not market share.

Company-reported performance numbers need context too. Crusoe’s 400-plus modules include systems built before Spark. Submer’s 500 MW of deployed liquid cooling and its very low PUE claims are company-stated. Nautilus’ own site gives two different unit-hour totals. Flexnode’s 35% schedule cut lacks a published project sample. Spark Liquid’s 3.5 MW and DSX POD’s 14 days are design figures for products not yet delivered.

The evidence is uneven, so operating sites, repeat customers and disclosed capacity sit at the top of our ranking.

Which modular data center startups are actually ahead?

Crusoe is still ahead overall in modular data centers, Armada is the closest challenger and gaining, and the rest lead narrower parts of the market.

Crusoe wins because it combines operating experience, repeat customer demand, a funded factory, access to energy and a cloud platform that can sell the capacity. Its September 2026 raise makes that combination even harder to match, and Spark Liquid gives it a denser product for 2027. Its weak point is that Spark output from the new factory is still not public.

Armada has the best chance of changing the ranking. It has enough money to compete, serious industrial and government customers, a Palantir certification, a factory partnership and now a firm megawatt-scale order. Delivered Leviathans, a running Arizona line and disclosed operating megawatts would narrow the gap quickly.

Nautilus keeps third place on years of operating evidence. Its focus on cooling makes it more specialized than Crusoe or Armada, but that focus also makes its efficiency claims more credible.

Submer ranks fourth and is closing in. Its cooling technology is established, it says it has deployed more than 500 MW of liquid cooling, and Corenix and Rubix give it reach across the value chain. It now needs named modular customers and running campuses.

ECL ranks fifth with a small but real operation. Its first facility, the Lambda deployment and a firm fuel-cell order give it more credibility than many larger announcements. CSC-1 will show whether it can repeat the model at commercial scale.

Flexnode ranks above Metrobloks because its product is standardized and supported by Eaton’s supply chain. Metrobloks has the larger pipeline, but it looks more like a developer than a repeatable manufacturer.

Rank	Startup	Why it ranks here
1	Crusoe	The broadest mix of operating experience, customer expansion, manufacturing investment, energy access and cloud demand, backed by a $3.9B Series F
2	Armada	The strongest focused challenger, with high-quality deployments, a firm order for five Leviathans, the 10 MW Orion and Palantir certification
3	Nautilus Data Technologies	The best long-term operating evidence among the specialists, including a live 6.5 MW facility
4	Submer	The strongest immersion-cooling ecosystem, 500 MW+ of liquid cooling deployed (company-stated) and a new modular arm, with most campus capacity still ahead
5	ECL	A small but operating competitor with dense AI workloads, full occupancy and a firm order supporting its 35 MW campus
6	Flexnode	A credible modular data hall design backed by Eaton, held back by the absence of a named operating facility
7	Metrobloks	A fast-growing development pipeline with large planned capacity, but no commissioned facility yet