Submer Technologies
Hardware systems sales (immersion pods, dielectric coolant, CDUs) with services and consumables attach, plus datacenter design-and-build (business unit added Feb 2025); adding infrastructure-as-a-service and NVIDIA Cloud Partner GPU cloud (InferX) and carrier-embedded edge GPU IaaS (Radian Arc) - a shift from capex product sales toward recurring infrastructure revenue
Only the Oct 2024 Series C carries a publicly disclosed post-money (~$500M). The Jan 2022 Series B (~$33.9M per Crunchbase) never disclosed a valuation; its point is an explicitly-labeled dilution-based estimate, not a reported figure. No priced equity round has been disclosed since Oct 2024; the Dec 2024 Santander facility was debt and the Feb 2026 Radian Arc acquisition had undisclosed terms, so the current mark is stale in both directions.
Earnings, margins, COGS & capex
Submer disclosed passing 150M euros of revenue in 2024, an unusually large figure for a ~$500M-valued, ~157-employee company and, if accurate, one of the largest immersion-cooling businesses globally. No margin, profitability, or balance-sheet data is public. The 2026 strategy shift - InferX (NVIDIA Cloud Partner neocloud) and the Radian Arc acquisition (telco-edge GPU IaaS, 70+ telecom and edge customers) - moves the model toward recurring, but capital-hungry, infrastructure revenue. Financial statements are filed privately in Spain and not publicly summarized.
Revenue trend
Margins
hardware systems margins typically 25-40% in this category; unverified for Submer
likely negative during neocloud buildout (GPU capex)
COGS structure
Not disclosed. Structurally: fabricated pod enclosures and tanks, pumps, heat exchangers and CDU components, dielectric coolant (synthetic single-phase fluids, both proprietary SmartCoolant and partner fluids such as ExxonMobil's), electronics, and assembly at its Barcelona facility and Houston, Texas 'Gigafactory'; post-pivot, GPU servers and datacenter capacity become the dominant cost line for the cloud segment.
Capex
Not disclosed. Historically moderate (Barcelona manufacturing and R&D plus the Houston Gigafactory); stepping up sharply with InferX GPU cloud deployments, Radian Arc edge nodes, and the up-to-1GW Madhya Pradesh (India) AI datacenter MoU (Jul 2025).
Latest earnings
not applicable
none published
- Last priced round
- $55.5M Series C at ~$500M valuation, Oct 2024 (M&G Catalyst lead; Planet First Partners, Norrsken VC, Mundi Ventures participating; Barclays Bank Ireland as placement agent)
- Prior round
- ~$34M Series B, Jan 2022, led by Planet First Partners with Mundi Ventures and Norrsken VC (post-money not disclosed)
- Total funding
- reported between ~$102M (Tracxn) and ~$131M (PitchBook), including ~20M euros Banco Santander debt (Dec 2024)
- FY2024 revenue
- 150M+ euros (company-disclosed, unaudited)
- Radian Arc footprint acquired
- 70+ telecom/edge customers, thousands of GPUs; acquisition announced Feb 10, 2026 (terms not disclosed)
- Implied EV/revenue at last round
- ~3x FY2024 revenue - low vs public thermal/AI-infra comps, though the round predates the revenue disclosure
Growth drivers
- AI rack power density (100kW+ racks) outrunning air cooling, forcing liquid-cooling adoption across new datacenter builds
- Sovereign AI demand — InferX plus Radian Arc positions Submer to sell in-country, telco-embedded GPU capacity (Middle East expansion Apr 2026, India MoU Jul 2025)
- Sustainability regulation (EU Energy Efficiency Directive datacenter reporting, PUE and heat-reuse mandates) favoring immersion's PUE ~1.05 and warm-water heat reuse
- 3M's exit from PFAS-based two-phase fluids removed the main rival cooling chemistry, consolidating demand into single-phase immersion and cold plates
- Channel expansion — UK partnerships (Boston Ltd, Hammer Distribution, Mar 2026), Stellium Datacenters OCP deployment, CleanSpark North America strategic partnership, AMAX and Hypertec integration partners, Intel validation
Bull & bear
A rare revenue-scale European AI-infrastructure asset priced at a 2024 valuation: ~3x reported revenue for the leading immersion-cooling franchise, with free optionality on the InferX/Radian Arc sovereign GPU cloud and on next-gen chip power densities forcing the market toward immersion.
- Valuation asymmetry: the ~$500M Series C was priced before the 150M-euro 2024 revenue disclosure, the NCP status, and the Radian Arc deal - a next priced round or strategic sale plausibly re-rates materially higher
- Physics tailwind: accelerator TDPs keep climbing; at 1.5-2kW+ per chip and 200kW+ racks, immersion and hybrid designs become economically compelling again
- Sovereign AI is a genuine wedge: 70+ telco and edge relationships via Radian Arc plus European identity wins deals US hyperscalers and Chinese vendors cannot
- Regulatory pull in the EU (energy-efficiency and heat-reuse rules) structurally favors immersion's PUE ~1.05
- Full-stack model compounds: cooling hardware sales seed relationships that convert into higher-LTV InferX cloud and edge-IaaS contracts, with 5GW+ of claimed land-and-power pipeline access across UK/US/India/Middle East partner consortiums (company-claimed, unverified)
Immersion cooling lost the AI-era standardization battle to direct-to-chip, the headline revenue is unaudited and possibly lumpy hardware sales, and the pivot into GPU clouds pits a ~$100-131M-funded company against multi-billion-dollar neoclouds and hyperscalers - with real down-round risk if either leg stumbles.
- NVIDIA's GB200/GB300 rack-scale reference designs use cold plates, not tanks; every major hyperscaler AI buildout followed suit, leaving immersion a single-digit share of liquid cooling
- The 150M-euro 2024 revenue is a single press-reported disclosure - no audited statement, no margin data; large one-off project hardware sales could mask weak recurring revenue
- GPU-cloud economics are brutal for sub-scale entrants: CoreWeave/Nebius-class rivals raise billions in debt against GPU fleets; InferX's capital base is a rounding error by comparison, and GPU price/perf deflation punishes late, small fleets
- Integration and focus risk: hardware manufacturer + neocloud operator + Australian edge-IaaS acquisition is three business models at once for a company of roughly 150-160 employees (PitchBook, 2026)
- No new equity since Oct 2024 during the hottest AI-infra funding market on record - either discipline or a signal that terms were not attractive; the Dec 2024 Santander facility was debt, not a validation of a higher equity mark
What it is worth
Last priced round anchor + EV/revenue cross-check vs public thermal and AI-infra comps
$200-400M (down round)
direct-to-chip fully owns the AI buildout, hardware orders prove lumpy/one-off, the GPU-cloud pivot burns cash against better-funded rivals, and new capital arrives at a discount with structure
$500-800M
hardware franchise grinds forward in its niche, cloud pivot progresses modestly; next round prices at or somewhat above the Oct 2024 mark
$1.2-2B
revenue holds/grows with real gross margins, InferX/Radian Arc adds recurring cloud revenue, and a 2026-27 growth round or strategic acquirer (Vertiv/Schneider-class) prices it as a sovereign AI-infrastructure platform at 6-10x revenue
Last mark: ~$500M (Series C, Oct 2024). Against the company-disclosed 150M+ euros (~$160M) FY2024 revenue, that is ~3x EV/S - a discount to Vertiv (~4-6x) and a fraction of neocloud multiples - but the revenue is unaudited, margins unknown, and the mark is 21 months stale. The Radian Arc acquisition and NCP launch have not been re-priced by any disclosed round.
SWOT
Strengths
- Category leadership in single-phase immersion cooling with a decade of deployments (SmartPod/MegaPod), Intel validation, and its own dielectric coolant chemistry
- Reported 150M+ euros 2024 revenue - real commercial scale, not a science project
- Full-stack positioning after InferX + Radian Arc — cooling hardware, datacenter design, and sovereign GPU cloud under one roof
- Backed by patient sustainability-focused capital (M&G Catalyst, Planet First Partners, Norrsken VC) plus a bank debt relationship with Santander
- European origin is an asset for EU/Middle East/India sovereign-AI procurement that avoids US/China vendors
Weaknesses
- Immersion cooling lost the first AI buildout round — NVIDIA GB200/NVL72 reference designs standardized direct-to-chip cold plates, and hyperscalers followed - immersion remains a niche of the liquid-cooling market
- No public audited financials; the headline revenue figure is single-source and unverified; margins unknown
- Server OEM warranty and serviceability friction for immersed hardware still slows enterprise adoption
- Sub-scale balance sheet (~$102-131M lifetime funding) for a GPU-cloud ambition where rivals deploy billions
- Integration risk — simultaneously running a hardware business, a neocloud, and a newly acquired Australian telco-edge platform with roughly 150-160 employees (PitchBook, 2026)
Opportunities
- AI inference at the telco edge (Radian Arc's carrier-embedded GPUs) - low-latency sovereign inference is an emerging, less contested niche than training clouds
- 1.5kW+ next-gen accelerators may exceed what cold plates alone can handle, reopening the door for immersion and hybrid immersion+DTC designs
- Heat-reuse and PUE regulation in the EU creating mandated demand for immersion-grade efficiency
- Madhya Pradesh up-to-1GW MoU and Middle East sovereign cloud deals as anchor projects for the full-stack model
- Consolidation — as a scaled survivor, Submer can roll up smaller immersion players or be a strategic target for Vertiv/Schneider-class acquirers
Threats
- Direct-to-chip ecosystem (Vertiv, nVent, Schneider/Motivair, CoolIT, Boyd) entrenching as the default AI thermal architecture, capping immersion TAM
- NVIDIA design-cycle dependence — each GPU generation's reference design dictates which cooling architectures win
- Neocloud price competition from CoreWeave, Nebius, and hyperscalers compressing GPU-cloud economics
- AI capex cycle risk — a digestion phase would hit both cooling hardware orders and GPU-cloud utilization simultaneously
- Large-cap thermal incumbents can bundle immersion offerings at lower margins if the segment inflects
Moats, dependencies & bottlenecks
Moats
Deployment track record and reference base in single-phase immersion (10 years, HPC/colo/telco installs, Intel validation, Stellium OCP showcase) Credibility matters in mission-critical cooling, but incumbents can buy their way in
pods + proprietary dielectric coolant + monitoring software + now cloud platform System integration raises switching costs per site; weaker across new-build decisions
Radian Arc's carrier-embedded footprint (70+ telecom/edge customers) Telco integration cycles are slow to win and slow to displace - the strongest recurring moat in the group if retained through integration
weak-to-moderate Valuable access, but NCP status is shared with many rivals and is NVIDIA's lever, not Submer's
weak-to-moderate M&G's Catalyst mandate explicitly backs IP-rich businesses, but competing immersion patents are plentiful
Dependencies
technology/platform GPU allocation for InferX, NCP program terms, and - critically - whether NVIDIA reference designs ever bless immersion
fluid used in the Stellium deployment; Shell SHEL, Castrol/BP BP, TotalEnergies TTE, Lubrizol/Berkshire BRK.B as category suppliers) Single-phase hydrocarbon fluids are multi-sourced, unlike the defunct 3M two-phase chemistry
Supermicro SMCI, HPE HPE, Lenovo) Immersion voids or complicates many standard warranties; OEM buy-in gates enterprise deals
Both hardware orders and GPU-cloud utilization are levered to the same cycle
Neocloud buildout needs substantially more capital than raised to date
Churn during ownership transition would gut the edge thesis
Advantages
- Best-known brand in single-phase immersion with real revenue scale
- European/neutral flag for sovereign AI procurement
- PUE ~1.05-class efficiency and heat-reuse fit with tightening EU regulation
- Full-stack offer (cool + build + operate + cloud) that pure cooling vendors and pure neoclouds each lack
- Sustainability-mandated investor base aligned with long build cycles
Weaknesses
- Immersion is currently the minority architecture in AI liquid cooling versus direct-to-chip
- Unaudited, single-source financial disclosure; opaque margins
- Thin capitalization for the cloud ambition; likely needs a large raise
- Post-acquisition integration risk (Radian Arc, Australia-based, different business model)
- Valuation mark is stale (Oct 2024) - genuine current worth is uncertain in both directions
Bottlenecks
- Capital: GPU fleet expansion for InferX is the binding constraint versus billion-dollar-funded neoclouds
- OEM warranty and serviceability acceptance for immersed servers
- Skilled deployment/integration capacity as projects scale to campus size (India up-to-1GW MoU) with a ~157-person team
- Power and site availability for sovereign cloud builds in the Middle East and India
- Management bandwidth across three concurrent business models post-acquisition
Top signals & trends
Top signals
Decisive strategic pivot to full-stack sovereign AI; execution now the question
NVIDIA relationship secured, though NCP is non-exclusive
If accurate, implies the last round undervalued the business; unaudited
Could signal discipline, an imminent larger round, or difficulty pricing up
Non-dilutive but adds fixed obligations ahead of a capex-heavy pivot
MoUs are not contracts - watch for signed anchor tenants
Early evidence of the full-stack strategy predating the InferX/Radian Arc moves
Structural headwind to the core immersion hardware franchise
Trends
Each generation strengthens the case for liquid, and eventually hybrid/immersion, cooling
Immersion relegated to edge, retrofit, harsh-environment, and ultra-density niches for now
Core demand driver for InferX + Radian Arc
Killed two-phase immersion rivals; single-phase hydrocarbon fluids (Submer's approach) are the survivor
Mandated efficiency reporting favors immersion PUE and warm-water heat reuse
Compresses returns for sub-scale GPU clouds like InferX
Ecosystem & competitor graph
Suppliers feed the company; customers pull from it. Line thickness shows the strength of each tie (supply-chain dependency, customer earnings contribution). Hover to isolate a tie.
GPUs for InferX/Radian Arc fleets; NVIDIA Cloud Partner program
Immersion fluid supplier (DC 3235 Super) in the Stellium Newcastle deployment
Category supplier of single-phase immersion dielectric fluids
Dielectric fluid supplier to the immersion market
Castrol ON immersion cooling fluids
Immersion-ready server hardware ecosystem
UK colo running Submer immersion (OCP ORv3 proof of concept, Jan 2025) at its Newcastle campus
70+ telco/edge customers running GPU edge IaaS incl. cloud gaming and AI inference
MoU for up to 1GW of liquid-cooled AI datacenters (Jul 2025)
Named individual end-customers largely not disclosed
Dominant datacenter thermal incumbent; CoolChip CDU direct-to-chip line is the architecture winning AI deployments
Liquid cooling and enclosure systems for AI datacenters
US OTC ADR; primary listing Euronext Paris (SU). Acquired Motivair (cold plates/CDUs); full datacenter power+cooling portfolio
US OTC; primary listing Nasdaq Stockholm (MTRS). Swedish datacenter cooling supplier (air and liquid)
Private US/NL immersion (two-phase and single-phase) and CDU maker; Tiger Global-backed
Private US single-phase immersion pioneer; closest like-for-like rival
Private UK precision liquid cooling (chassis-level immersion)
Private Dutch immersion cooling vendor
Private (KKR-owned) direct-to-chip cold plate leader
Private (Goldman Sachs-owned) thermal solutions incl. cold plates and CDUs
Competes with InferX on GPU cloud, at vastly greater scale
European-rooted neocloud competing for the same sovereign/EU AI workloads