Data center cooling: which startup is ahead?
In our data center market deck, you will find everything you need to understand the market
SUMMARY
ZutaCore is currently ahead in data center cooling, although Submer and Accelsius are close enough to keep the race open.
ZutaCore’s lead comes from combining more than 75 reported deployments with commercially available products, supported server integrations, credible customers, industrial partners, and roughly $111 million in funding. No other independent specialist currently matches that full package.
The market is not producing one winner across every cooling architecture. ZutaCore leads two-phase direct-to-chip cooling, Submer leads the broader immersion platform, Iceotope leads sealed chassis-level cooling, and Corintis leads the emerging in-chip microfluidic category.
Direct-to-chip cooling currently has the widest commercial opening. NVIDIA’s rack architectures, Open Compute Project standards, and existing data center layouts are pushing operators toward cold plates, manifolds, CDUs, and warm-water loops rather than complete immersion retrofits.
Submer is the strongest strategic challenger because it is building beyond cooling equipment. Its manufacturing footprint, certified systems, heat-reuse capabilities, and data center development projects could eventually make it a complete AI infrastructure company, but much of that capacity is still planned rather than operating.
Accelsius is catching up fastest. It publishes the most aggressive independent rack target, has raised at least $89 million, and has a planned role across DarkNX’s 300-megawatt Canadian campus, although the campus and the company’s fivefold deployment growth still need firmer operating numbers.
LiquidStack remains the strongest startup-origin commercial benchmark, even though it is no longer independent. Its 300-megawatt equipment order, mature CDU portfolio, and acquisition by Trane show the scale an independent startup still needs to reach.
The most impressive technical result does not automatically identify the market leader. Corintis removed up to three times more heat in Microsoft laboratory testing, while Accelsius targets 250-kilowatt racks, but ZutaCore currently offers the more mature and easier-to-procure commercial system.
Scale also takes different forms. Submer has the clearest owned manufacturing footprint, ZutaCore has the strongest partner-led delivery ecosystem, Iceotope has the broadest visible patent wall, and Corintis could create the deepest long-term technical lock-in.
The available data are too incomplete for honest market-share calculations, but they are sufficient for a ranking. Today, ZutaCore is first, Submer is second, and Accelsius is third, with Accelsius the most likely to change the order if its planned deployments become operating references.
This market map, featured in our data center market deck, highlights top companies and startups in the data center market
Data center cooling: which startup is ahead?
Which data center cooling startups should we actually compare?
The serious data center cooling startup field currently includes seven independent specialists, with LiquidStack retained as an acquired commercial benchmark.
We include companies whose main product removes heat directly from chips, servers, racks, or complete computing systems. We exclude conventional HVAC manufacturers, component suppliers, coolant producers, and large industrial groups that happen to sell liquid-cooling equipment alongside much larger product portfolios.
The companies fall into four technical groups. ZutaCore and Accelsius sell two-phase direct-to-chip systems. Submer, GRC, and Asperitas focus on immersion cooling. Iceotope seals liquid around individual server chassis. Corintis pushes microfluidic cooling inside or immediately next to the chip.
Those products compete for the same AI infrastructure budgets, although buyers cannot swap one for another without changing server architecture, facility design, or maintenance procedures.
LiquidStack would rank among the strongest companies in this comparison, but Trane completed its acquisition in March 2026. JetCool was acquired by Flex in November 2024, Chilldyne by Daikin in November 2025, and CoolIT has operated under KKR ownership since 2023. We use these acquired specialists to judge commercial scale, without ranking them as independent startups.
Funding totals are also less precise than they look. ZutaCore’s cumulative funding is relatively clear at roughly $111 million. Accelsius has announced at least $89 million. Submer reports approximately €80 million. Iceotope estimates vary from roughly $72 million to more than $100 million because databases treat grants, strategic investments, and older rounds differently. GRC has never published a reliable cumulative figure.
| Startup | What it does | Publicly tracked funding | Place in the comparison |
|---|---|---|---|
| ZutaCore | Waterless two-phase direct-to-chip cooling and CDUs | About $111M | Core competitor |
| Accelsius | Two-phase direct-to-chip NeuCool systems | At least $89M | Core competitor |
| Submer | Single-phase immersion cooling and data center infrastructure | About €80M | Core competitor |
| Iceotope | Sealed chassis-level precision liquid cooling | About $72M to $107M | Core competitor |
| Corintis | In-chip and microfluidic cooling | About $53M | Earlier-stage core competitor |
| GRC | Single-phase immersion cooling systems | Undisclosed | Commercial immersion benchmark |
| Asperitas | Immersion cooling driven by natural convection | More than $2M publicly tracked, plus undisclosed capital | Smaller specialist |
| LiquidStack | Direct-to-chip CDUs and immersion systems | At least $45M before acquisition | Acquired benchmark |
Is there a clear leader in data center cooling today?
ZutaCore currently leads the independent field, although Submer and Accelsius remain close enough to make this a three-company race.
ZutaCore has the best mix of operating deployments, products suited to current AI racks, named customers, server integrations, industrial partners, and fresh capital. The company now reports more than 75 deployments worldwide. Its references cover colocation, telecommunications, research computing, and university infrastructure rather than one narrow customer type.
Submer has built the strongest immersion platform. It combines cooling equipment, its own manufacturing footprint, validated server configurations, heat-reuse capabilities, and increasingly ambitious data center development projects. Its commercial weakness is disclosure. Submer announces large facilities and partnerships, but rarely gives a complete installed base measured in operating racks or megawatts.
Accelsius is climbing fastest. NeuCool only became broadly available recently, yet the company has already attracted strategic capital from Johnson Controls and Legrand, expanded its product family, and secured a planned role in DarkNX’s large Canadian campus.
Leadership splits further down the field. GRC has the deepest immersion operating history. Iceotope owns the broadest disclosed patent portfolio. Corintis has produced the most striking chip-level cooling result. None currently combines technical differentiation with the same breadth of commercial evidence as ZutaCore.
The gap between first and third remains manageable. One large operating deployment program from Accelsius, or the commissioning of Submer’s announced facilities, could narrow it quickly.
If you want more recent data on this point, please see our latest data center market report.
As this chart shows, and as featured in our data center market deck, search interest in data centers has increased significantly
Which cooling startup has the strongest funding position?
ZutaCore has the clearest funding lead, and its latest financing followed meaningful commercial deployments, which makes the capital unusually useful.
ZutaCore has raised roughly $111 million after announcing a $100 million Series C. That puts it around 25% ahead of Accelsius’s disclosed $89 million. The comparison with Iceotope is less certain because available estimates place Iceotope anywhere from approximately $72 million to $107 million.
The timing separates ZutaCore from several competitors. Its latest round came after the company had shipped products, completed customer installations, and established server and infrastructure partnerships. Investors are financing the expansion of an existing business rather than waiting for the first commercial product.
Accelsius raised a $65 million Series B shortly after launching its first generally available NeuCool products. The company now has enough capital to expand manufacturing, field service, engineering, and customer deployments, but it still needs to prove that demand extends beyond early projects.
Submer’s approximately €80 million has supported a different strategy. The company is spending beyond individual cooling tanks and moving into manufacturing, full facility design, and data center development. That broader ambition could produce a larger business, although it also requires more capital and exposes Submer to construction and infrastructure risk.
Corintis has raised about $53 million, much of it after Microsoft validated its microfluidic architecture. The funding is impressive for such an early company, but Corintis still faces semiconductor qualification, production, and reliability work before its technology reaches the same commercial stage as ZutaCore or Submer.
Funding supports ZutaCore’s lead today. It does not guarantee that ZutaCore will use the money more effectively than Accelsius or Submer.
Who has actually deployed the most data center cooling systems?
ZutaCore has the strongest disclosed deployment record among independent direct-to-chip cooling startups.
The company now reports more than 75 deployments worldwide. Its projects include work involving Equinix, SoftBank, the University of Münster, the University of Pisa, and the University of Chicago. That mix gives ZutaCore credible evidence across colocation, telecom, research, and high-performance computing.
GRC probably has more accumulated operating experience in immersion cooling. The company says its systems operate in more than 20 countries after roughly 16 years in the market. Geography alone cannot tell us how many systems, racks, or megawatts are running, but it shows that GRC has dealt with long-term fluid management, component compatibility, maintenance, and customer support across different environments.
Accelsius reported fivefold deployment growth during the first half of 2025. That is the strongest disclosed growth rate in the field, although the company did not publish the starting number. Growth from two deployments to ten would tell a very different story from growth from twenty to one hundred.
Submer has several credible installations, including high-density projects with data center operators and telecom groups. Iceotope also has real systems in the field through collaborations with Lenovo, Schneider Electric, Avnet, and other infrastructure partners. Neither company publishes a consolidated deployment total that can be compared cleanly with ZutaCore.
Private startup data remain frustratingly incomplete. ZutaCore gives us the clearest evidence of repeated deployment, GRC leads on years of experience, and Accelsius leads on recent pace.
This chart, featured in our data center market deck, illustrates yearly venture capital funding for data center startups
Who has won the biggest real data center cooling contracts?
LiquidStack still has the strongest startup-origin equipment order, while Accelsius has secured the largest independent commitment waiting to become operating capacity.
LiquidStack announced a 300-megawatt CDU order from a major US data center operator in January 2025. The customer remained unnamed, but the announcement concerned cooling equipment ordered for deployment. Among startup-origin companies, that remains the strongest public evidence of demand converted into a large purchase.
Accelsius has an agreement to provide NeuCool systems across DarkNX’s planned 300-megawatt campus in Ontario. The first two facilities are expected to provide 65 megawatts each. The project could transform Accelsius into a large supplier, but the full 300 megawatts represents planned campus capacity. It should not be counted as installed cooling today.
Submer has announced even larger development ambitions. These include an Indian initiative connected to as much as one gigawatt of capacity, a planned 56-megawatt facility near Barcelona, and a €1 billion data center development in Flix. These projects show that Submer wants to shape entire facilities rather than sell isolated cooling equipment.
A 300-megawatt order, a 300-megawatt development agreement, and a one-gigawatt project pipeline are not the same thing. LiquidStack has the strongest large equipment order. Accelsius has the most important independent cooling commitment. Submer has the largest infrastructure pipeline, but much of that capacity still sits on development schedules.
ZutaCore has disclosed more completed deployments, yet it rarely publishes contract values, backlog, or installed megawatts. That lack of financial detail makes the company look smaller on announced deal size than it may actually be.
Is direct-to-chip cooling beating immersion cooling?
Direct-to-chip cooling currently has the broader AI data center opportunity because it matches the rack designs now spreading through the server ecosystem.
NVIDIA’s GB200 NVL72 uses a liquid-cooled rack-scale architecture, and the Rubin generation continues the move toward dense liquid-cooled systems. Server makers, rack integrators, and data center operators are therefore organizing procurement around cold plates, manifolds, CDUs, and warm-water loops.
The Open Compute Project is also standardizing connectors, cold plates, manifolds, CDUs, and facility interfaces. Standards make it easier for operators to combine equipment from several suppliers and reduce the risk of committing to one proprietary cooling design.
Capital has followed that architecture. ZutaCore, Accelsius, Iceotope, and Corintis have collectively raised at least $325 million. Each brings liquid closer to the processor without placing the complete server inside an immersion tank.
Direct-to-chip systems also fit retrofit projects more easily. An operator can install compatible servers, piping, manifolds, and CDUs while preserving much of the existing rack layout and maintenance process. The work remains complex, but it is familiar enough for mainstream data center teams.
Immersion cooling is stronger in greenfield facilities built around high density, compact layouts, and heat recovery. It can remove server fans, cool more components, and save large amounts of space. Independent research has found energy reductions approaching 50% and space reductions of roughly two-thirds in suitable configurations.
Those gains come with new maintenance routines, fluid-compatibility questions, lifting equipment, and more difficult retrofits. Submer and GRC can win large purpose-built facilities, while ZutaCore and Accelsius can address a wider range of existing and new data centers.
If you want more recent data on this point, please see our latest data center market report.
This chart, featured in our data center market deck, shows how Equinix is capturing share in data centers
Which cooling startup has the most mature and practical product today?
ZutaCore currently offers the strongest combination of product maturity, retrofit practicality, and commercial support for mainstream AI data centers.
Its HyperCool platform is already deployed, available with several CDU configurations, and integrated into supported server systems. UNICOM Engineering has offered warrantied Dell PowerEdge XE9680 systems using ZutaCore cooling, while ASRock Rack has integrated the technology into NVIDIA HGX B300 platforms.
These integrations remove a lot of friction. Customers can procure a supported server and cooling configuration instead of assembling cold plates, manifolds, pumps, controls, firmware, and warranty arrangements themselves.
ZutaCore’s closed, waterless chip-level loop can also be introduced without completely rebuilding an existing facility. That gives it an advantage over immersion systems, which require fluid-handling procedures, compatible components, lifting equipment, and different maintenance routines.
Accelsius follows a similar and increasingly credible path. Its IR150 packages cooling inside a 150-kilowatt rack, while the MR250 supports larger rack and row deployments. Installation and service partnerships with UNICOM Engineering and Park Place Technologies should make adoption easier.
GRC and Submer remain more mature in immersion cooling. GRC brings roughly 16 years of operating experience, while Submer combines commercial SmartPod systems, a Houston manufacturing site, and an Intel-certified configuration using Shell fluid and Supermicro hardware.
Immersion may deliver better economics in purpose-built facilities optimized for density, land use, and heat recovery. For most existing operators, though, ZutaCore currently asks for fewer operational changes and comes with the strongest supported purchasing path.
Who can cool the hottest AI racks?
Accelsius publishes the most aggressive rack specification today, while Corintis has produced the strongest chip-level laboratory result.
Cooling specifications remain difficult to compare. Companies alternate between watts per processor, kilowatts per rack, CDU capacity, fluid temperature, and reductions measured against different baseline systems.
Accelsius says NeuCool can support more than 4,500 watts per processor socket and as much as 250 kilowatts per rack. ZutaCore and Pegatron have certified a configuration supporting chips up to 2,800 watts. Iceotope demonstrated stable cooling of a 1,500-watt processor using its single-phase dielectric approach.
Corintis produced the most radical result through Microsoft’s laboratory testing. Microchannels integrated into silicon removed as much as three times more heat than conventional cold plates and reduced maximum silicon temperature rise by 65%.
That experiment could point toward the architecture needed for future chips, but it did not test a mass-produced product running across a full customer fleet. Data center buyers still need evidence on reliability, serviceability, manufacturing yield, and long-term operation.
LiquidStack remains the strongest startup-origin benchmark at complete system level. Microsoft has tested its equipment at up to 250 kilowatts per rack, and the company has validated CDU platforms at much larger facility scale.
| Company | Disclosed performance result | Type of evidence | What we can reasonably conclude |
|---|---|---|---|
| Accelsius | More than 4,500W per socket and up to 250kW per rack | Company product specification | Highest disclosed independent rack target |
| ZutaCore | Certified configuration for chips up to 2,800W | Partner certification | Strong commercial readiness for next-generation chips |
| LiquidStack | Testing up to 250kW per rack and CDU validation up to 14MW | Customer testing and certified product data | Strongest complete-system benchmark from a startup-origin company |
| GRC | 100kW per rack with warm water and close to twice that with chilled water | Customer project case study | Proven immersion performance in a real installation |
| Iceotope | Stable cooling of a 1,500W processor | Technical demonstration | Credible chip result with a narrower test scope |
| Corintis | Up to 3× more heat removed and 65% lower temperature rise | Microsoft laboratory testing | Most disruptive result, with commercial validation still ahead |
This chart, featured in our data center market deck, illustrates yearly funding for data center startups
Which cooling startup has the strongest path to commercial scale?
ZutaCore has the strongest partner-led route to scale, while Submer has built the clearest manufacturing operation of its own.
ZutaCore’s ecosystem already covers operators, server makers, engineering firms, building systems, cooling infrastructure, and field support. Its relationships with Equinix, SoftBank, UNICOM Engineering, Pegatron, ASRock Rack, Carrier, Mitsubishi Heavy Industries, and Samsung make larger deployments easier to sell, assemble, and service.
The model does not require ZutaCore to manufacture every part itself. That can support faster expansion, although it also leaves execution dependent on partners. Its latest $100 million financing gives the company room to secure components, hire field engineers, expand regional support, and hold more inventory.
Submer has taken the more vertically integrated route. Its Houston facility gives it direct control over the assembly, testing, and delivery of immersion systems, while its partnerships with Intel, Shell, Supermicro, Telefónica, AMD, and GIGABYTE support a broader infrastructure offering.
Accelsius is building a credible third route. Johnson Controls and Legrand provide industrial backing, its service partners extend its reach, and the planned DarkNX campus could become a major anchor deployment. The company also reports a roughly 500% expansion in Austin production capability, although it has not disclosed the starting capacity.
Corintis has the most ambitious production target, with plans for annual capacity above one million cold plates. That could change its position completely, but today it remains a target rather than verified output.
Submer leads on visible owned manufacturing. ZutaCore has the strongest route to market and the broadest delivery network. For now, that network gives ZutaCore the better chance of converting demand into repeated commercial installations.
If you want more recent data on this point, please see our latest data center market report.
What can cooling startups defend before industrial giants catch up?
Iceotope has the broadest visible patent wall, ZutaCore has the strongest commercial moat, and Corintis could build the deepest technical lock-in.
Iceotope reports 219 granted or pending patents around its precision liquid-cooling architecture. The portfolio reflects nearly two decades of work on dielectric fluids, sealed chassis, materials, serviceability, and system integration.
ZutaCore’s protection comes from a wider combination of assets. HyperCool combines two-phase heat transfer, refrigerant control, closed-loop operation, leak and pressure management, control software, server qualification, warranties, and partnerships. Competitors can reproduce individual components more easily than the entire supported package.
Corintis moves cooling into the semiconductor design itself. If chipmakers begin designing processors around microfluidic channels, competitors would need access to fabrication processes, packaging expertise, qualification programs, and long semiconductor development cycles. Microsoft’s test gives the approach technical credibility, while commercial lock-in still depends on mass production and adoption.
GRC and Submer rely heavily on operating knowledge. Fluid compatibility, server preparation, safety, maintenance, heat reuse, and facility integration improve through repeated deployments. These capabilities take time to build, although established industrial groups can buy them.
Recent acquisitions show how that pressure works. Flex bought JetCool, Daikin bought Chilldyne, and Trane bought LiquidStack. KKR acquired CoolIT in 2023 and later agreed to sell it to Ecolab for $4.75 billion after CoolIT’s revenue reportedly quadrupled and its workforce doubled.
Large industrial companies already own the factories, customer relationships, service networks, and facility equipment surrounding the startup products. A cooling specialist may gain more from partnering with an incumbent than from trying to replace one.
ZutaCore is currently best protected from that threat because Carrier, Mitsubishi Heavy Industries, Samsung, and other industrial groups sit inside its ecosystem. Accelsius is pursuing a similar route with Johnson Controls and Legrand. Submer is taking the more ambitious path of expanding toward complete infrastructure development.
This chart, featured in our data center market deck, compares the main business model options for hyperscale data center operators
Which data center cooling startup is gaining ground fastest now?
Accelsius is expanding fastest on the one disclosed deployment-growth metric, whereas ZutaCore is adding momentum across more parts of the business at once.
Accelsius reported fivefold deployment growth during the first half of 2025. Since then, the company has raised $65 million, moved two NeuCool configurations into general availability, expanded production capacity, and secured its planned role in the DarkNX campus.
The missing deployment baseline still limits the claim. Fivefold growth could represent a modest number of projects. DarkNX will become much more valuable evidence once the first facility begins operating.
ZutaCore’s recent progress is broader. The company raised $100 million after establishing a commercial footprint, added integrations for newer NVIDIA-based systems, expanded its CDU offering, and deepened relationships with server and industrial partners.
Lately, Corintis has also moved quickly. It announced two substantial financings in late 2025 and gained public validation from Microsoft. Its next milestones are less visible but more demanding: product certification, manufacturing yield, reliability data, and repeat orders.
Submer has the largest infrastructure ambition. Its announced projects cover facilities measured in tens, hundreds, and potentially thousands of megawatts. Construction schedules will decide whether that ambition becomes market leadership or remains a large development pipeline.
Accelsius is the company catching up fastest. ZutaCore still has stronger momentum overall because commercial deployments, financing, products, and partnerships are all moving in the same direction.
Who leads each part of the data center cooling market?
The data center cooling market currently has four distinct leaders because direct-to-chip, immersion, chassis-level, and in-chip cooling reward different strengths.
ZutaCore leads independent two-phase direct-to-chip cooling. It has more field evidence and a broader commercial ecosystem than Accelsius, although Accelsius publishes more aggressive rack specifications and is narrowing the gap.
Submer leads the full-stack immersion strategy. The company combines cooling products, manufacturing, heat reuse, facility design, and data center development. GRC remains stronger on years of field operation.
Iceotope leads sealed chassis-level cooling. No other independent specialist has built the same combination of intellectual property, packaged infrastructure products, and long-term focus around that architecture.
Corintis leads the emerging in-chip microfluidic category. The technology has produced an exceptional laboratory result, but the segment remains much earlier than direct-to-chip or immersion cooling.
LiquidStack still provides the best startup-origin commercial benchmark across large CDUs and high-density cooling systems. Trane ownership removes it from the independent ranking.
| Cooling subcategory | Current leader | Closest challenger | Why the leader is ahead |
|---|---|---|---|
| Two-phase direct-to-chip | ZutaCore | Accelsius | More disclosed deployments and broader commercial integrations |
| Full-stack immersion platform | Submer | GRC | Broader manufacturing, facility, and development strategy |
| Field-proven immersion operations | GRC | Submer | Longer operating history across many countries |
| Sealed chassis-level cooling | Iceotope | No close independent peer | Largest visible patent portfolio and a mature packaged architecture |
| In-chip microfluidic cooling | Corintis | No comparable independent specialist | Microsoft-validated architecture with strong technical upside |
| Startup-origin commercial benchmark | LiquidStack, now owned by Trane | ZutaCore | Large equipment order, mature CDU portfolio, and industrial acquisition |
If you want more recent data on this point, please see our latest data center market report.
This chart, featured in our data center market deck, shows the revenue mix across customer segments in the data center market
How reliable is the data behind this startup ranking?
The available evidence is strong enough to name a data center cooling leader, but too uneven to calculate honest market shares.
Private cooling startups publish financing rounds, selected customers, product specifications, and major partnerships. They usually withhold revenue, gross margins, installed megawatts, unit shipments, average prices, factory utilization, renewal rates, and repeat-order volumes.
Completed acquisitions, named customer deployments, certified products, generally available systems, and operating manufacturing sites deserve the highest confidence. Those events can usually be checked through customers, investors, certification bodies, or transaction announcements.
Company-reported deployment figures and performance tests require more care. ZutaCore’s deployment total comes from ZutaCore. Accelsius’s growth rate comes from Accelsius. Corintis’s technical result is more credible because Microsoft carried out the testing, but it still describes controlled laboratory work.
Development agreements sit lower in the evidence hierarchy. A planned 300-megawatt campus, a memorandum covering one gigawatt, or a target of one million annual cold plates may become commercially important. None represents operating output today.
Energy-savings claims are especially difficult to compare. Vendors use different climates, loads, baselines, facility boundaries, and definitions of cooling energy.
Revenue would settle several debates, but the leading independent companies do not publish audited figures. We therefore give more weight to completed deployments, product availability, certifications, customer quality, and the ability to support installations.
That approach leaves some uncertainty in the middle of the ranking. It does not change the identity of the current leader.
Which data center cooling startups are actually ahead?
ZutaCore is currently ahead overall, with Submer as the closest strategic challenger and Accelsius as the company most likely to change the ranking.
ZutaCore wins on the combination that matters most in this market today. Its cooling architecture fits the direction of AI server design, its products are commercially available, and its ecosystem spans operators, server manufacturers, engineering firms, and industrial cooling groups.
ZutaCore’s more than 75 reported deployments give it the clearest independent field record in direct-to-chip cooling. The company also has enough fresh capital to expand production support and pursue much larger installations.
Submer ranks second because it is building something broader than an immersion-cooling product company. Its manufacturing facility, certified hardware combinations, heat-reuse capabilities, and large development projects give it a credible path toward complete AI infrastructure.
That wider strategy also creates more execution risk. Submer needs its planned facilities to become operating references using its cooling systems. Until then, ZutaCore has stronger evidence of repeated commercial deployment.
Accelsius ranks third and has the clearest path to moving higher. The company is expanding quickly, its systems target extremely dense AI racks, and the DarkNX agreement could give it a major operating reference. The fivefold growth figure still lacks an absolute deployment base.
GRC ranks fourth because long field experience carries real weight, especially in immersion. Its weaker public disclosure and narrower positioning around immersion keep it below the top three.
Iceotope ranks fifth. The company has valuable intellectual property and a differentiated chassis-level architecture, but public evidence of current deployment growth remains thinner.
Corintis ranks sixth because commercial readiness still trails its technical potential. A successful move into qualified mass production could push the company several places higher.
Asperitas remains a credible immersion specialist, although its visible funding, customer scale, and commercial footprint are smaller than those of the other companies in the ranking.
LiquidStack would probably rank first if acquired companies were eligible. Its large CDU order, mature product family, and acquisition by Trane provide stronger commercial-scale evidence than any independent startup has published. The comparison gives ZutaCore a clear next test: winning larger repeat orders before industrial ownership becomes the main source of scale.
| Rank | Startup | Why it holds this position | What could change the ranking |
|---|---|---|---|
| 1 | ZutaCore | Strongest overall mix of deployments, AI-rack relevance, customers, integrations, industrial backing, and funding | Slow conversion of its new capital into larger repeat orders |
| 2 | Submer | Strongest immersion platform, visible manufacturing, certified systems, and broad infrastructure strategy | Commissioning its announced data center projects could move Submer into first place |
| 3 | Accelsius | Fastest disclosed growth, high-density products, fresh capital, and a major planned campus deployment | Operating DarkNX capacity and a larger disclosed installed base could move Accelsius ahead |
| 4 | GRC | Longest immersion field history and credible industrial deployments | Faster commercial growth or expansion beyond immersion |
| 5 | Iceotope | Strong patent portfolio and differentiated chassis-level cooling | More transparent deployment, revenue, and hyperscale adoption data |
| 6 | Corintis | Strongest emerging in-chip architecture and Microsoft validation | Certification, mass production, and recurring customer orders |
| 7 | Asperitas | Credible immersion technology and industry experience | More capital, larger customer deployments, and clearer commercial scale |
If you want more recent data on this point, please see our latest data center market report.
This chart, featured in our data center market deck, shows how hyperscale AI-ready campus technology has evolved over time
OUR METHODOLOGY
We assessed which data center cooling startup is ahead by comparing the areas that most clearly separate technical promise from an operating business: deployments, contracts, product maturity, cooling performance, adoption practicality, customers, manufacturing readiness, defensibility, and recent momentum.
We included independent specialists whose main product removes heat directly from chips, servers, racks, or complete computing systems. Conventional HVAC companies, component suppliers, coolant producers, and diversified industrial groups were excluded from the startup ranking.
LiquidStack was retained as a commercial benchmark because it originated as a cooling startup and has disclosed equipment orders and system-scale validation that independent companies have not yet matched. It was not ranked because it is now owned by Trane.
We gave the most weight to evidence showing technology moving into real use: completed deployments, commercially available products, supported server integrations, certifications, equipment orders, operating manufacturing sites, named customers, and partnerships that can help deliver and service systems.
Planned campuses, development agreements, manufacturing targets, and infrastructure pipelines were treated as evidence of future potential rather than current operating scale. We kept equipment orders separate from announced campus capacity because the two can look similar in a headline while representing very different levels of commercial progress.
Cooling-performance claims were compared according to the type of evidence behind them. Product specifications, partner certifications, customer installations, and laboratory tests answer different questions, so a record chip-level result did not automatically outweigh field history or commercial availability.
We also separated category leadership from overall leadership. Direct-to-chip, immersion, sealed chassis-level, and in-chip cooling require different infrastructure and reward different strengths. The overall ranking reflects the combined pattern across those architectures rather than one exceptional technical result.
Key sources include ZutaCore’s financing and deployment update, ZutaCore’s customer deployments, Accelsius’s Series B announcement, the Accelsius and DarkNX campus agreement, Microsoft’s microfluidic cooling results, Open Compute Project standards, NVIDIA’s GB200 NVL72 architecture, NVIDIA’s Rubin platform, Submer’s product and infrastructure materials, GRC’s immersion systems and deployments, Iceotope’s precision liquid-cooling materials, and Corintis’s microfluidic technology materials.
Acquisition and scale comparisons also use announcements from Trane on LiquidStack, Flex on JetCool, Daikin on Chilldyne, and Ecolab on CoolIT Systems.
In our data center market deck, we identify pain points entrepreneurs should prioritize
Related blog posts
- Which startup is redefining how data centers are powered?
- Who has the edge in modular data centers?
- Is power replacing chips as AI’s bottleneck?
