US45784M1080
1. the problem: Why cooling becomes a bottleneck
When talking about AI infrastructure, the same terms are usually used: GPUs, networks, power supply, data center capacity. However, one factor is still too rarely discussed - cooling. heat dissipation.
Each new GPU generation is more powerful than the last. However, more computing power also means more power consumption, and this power is almost completely converted into heat. The more densely the racks are packed, the more heat is generated in the smallest of spaces.
Air cooling is increasingly reaching its physical limits. It has worked for years, but with modern AI racks with 100-300+ kW power per rack, it is simply no longer enough to cool the room. You have to cool the chips directly.
This is not a future theory, it is physics, and it is already a reality today.
2. the technical solution: two-phase direct-to-chip cooling
The first step away from air cooling is single-phase liquid coolingA liquid flows through cooling plates directly on the chip, absorbs heat and transports it away. This works better than air - but even here there is a limit.
The decisive physical lever lies in the phase changeWhen a liquid evaporates, it absorbs many times more heat than when it is simply heated. This is precisely what two-phase cooling:
- Liquid hits the hot chip
- It evaporates and absorbs a lot of heat in the process
- The vapor is discharged and condenses back into liquid
- The cycle starts all over again
The denser future racks become - keyword: Vera Rubin and subsequent generations from H2 2026 - the more important this technology becomes. You can have land, you can have power, you can have GPUs - but if the rack cannot be cooled efficiently, the full computing power simply cannot be used.
3. the company: Accelsius and its product NeuCool
Accelsius is not a classic startup. The company was founded in 2022 by Innventure to commercialize a high-performance cooling technology that was originally developed at Nokia Bell Labs developed at Nokia Bell Labs. The core product is called NeuCool.
What NeuCool does
- 4,500 W cooling capacity per GPU socket - more than the specified upper limit of 4,000 W for next-generation single-phase systems
- No water in the IT rack - instead a dielectric, non-conductive refrigerant (safety class A1). In the event of a leak, there is no short circuit on the GPUs
- Almost no water consumption - important, as water consumption is now triggering political resistance to new data centers in the USA
The product range
MR250
Two-phase direct-to-chip solution for new buildings (greenfield)
up to 250 kW
IR150
Fully integrated rack solution incl. CDU, 42U IT area, distributor - plug & play, also suitable for retrofits
up to 150 kW
TSR
Thermal simulation rack for testing and validation without expensive AI servers
Simulates up to 252 kW
economic efficiency
According to Accelsius and an analysis by Jacobs Engineering, the following savings can be achieved compared to single-phase systems:
- 35-44% annual OpEx savings
- 8-17% TCO savings over 5 years
4 Validation: Who is behind it?
Accelsius competes with established players such as Vertiv, Schneider Electric and CoolIT. The competitive risk is real. This makes external validation all the more important.
Capital:
- Series B round of 65 million USDled by Johnson Controlswith participation from Legrand
- Post-money valuation: ~665 million USD
Ecosystem:
- Participant in the NVIDIA Inception Program
- Presentation at the NVIDIA GTC
- Technology partner in the Dell Customer Solutions Center
Commercial Deployments:
- DarkNX (Ontario): 300 MW AI campus to deploy NeuCool - the largest two-phase direct-to-chip deployment in the world to date, according to Innventure
- Validations underway with: Global Switch, Equinix, Nordik Data Centers, Computacenter
Hyperscaler discussions are taking place, but are still in the early stages. The management is communicating this transparently and expects a new product line with higher CDU capacity in the coming months.
5. the investment vehicle: Innventure (INV)
Accelsius is not listed on the stock exchange. The only public access is via Innventure ($INV
)the parent company, which also holds AeroFlexx (packaging) and Refinity (plastics recycling) in addition to Accelsius.
Innventure owns ~43% of Accelsiusbut retains the majority of voting rights, meaning that Accelsius is consolidated in the INV balance sheet.
Outlook of the management
- Accelsius order intake Q1 2026: > USD 50 million
- Cash flow break-even expected at Accelsius: End of 2026
- Annualized sales rate Accelsius end of 2026: ~100 million USD
- Positive consolidated cash flow Innventure: Targeted by 2028
- AeroFlexx and Refinity to conduct own capital rounds to relieve the burden on the holding company
Main risk: Historically, Innventure has diluted shareholders through capital increases at holding company level. The company communicates that future financing should primarily take place at subsidiary level. The SEPA (Standby Equity Purchase Agreement) remains a formal risk, but according to management it should not have to be activated.
6. valuation
As Innventure is not a traditional income company, a sum-of-the sum-of-the-parts approach (SOTP) is appropriate.
Starting point: Implied market value of Accelsius
Last private valuation Accelsius
*Conservative assumption of USD 50-100m for AeroFlexx + Refinity and deduction of the holding company's net debt.
The market is therefore currently valuing Accelsius at a moderate premium to the last private round - but not nearly as if Accelsius were a potential solution provider for a central AI infrastructure bottleneck.
Comparative value: CoolIT takeover by Ecolab
Ecolab acquired CoolIT for ~USD 4.75 billion with expected annual revenues of ~USD 550 million - this corresponds to a sales multiple of ~8.6x. CoolIT is commercially more mature, but this value serves as a realistic reference point.
Scenarios for Accelsius until the end of 2028
Recalculation to the INV share price
(Assumptions: 40% INV stake after potential dilution at subsidiary level, USD 50m for AeroFlexx + Refinity, 100m fully diluted shares)
7. risks (summarized)
- Dilution risk: Innventure has historically impacted shareholders through capital increases. SEPA continues to be active.
- Competition risk: Vertiv, Schneider, CoolIT and others are established and well capitalized.
- Commercialization risk: Incoming orders ≠ sales. Conversion needs to be proven first.
- Clump risk: Few large customers (e.g. DarkNX) make up a significant part of the prospect.
- Technology risk: It is not certain that two-phase cooling will become the standard. Improved single-phase systems could last longer than expected.
- Holding structure: INV is not a direct investment in Accelsius. The holding discount is structurally justified.
8. conclusion
Innventure provides public access to Accelsius, a private company developing two-phase direct-to-chip cooling, which could become increasingly necessary as AI rack densities increase.
The current market capitalization of ~$510m implies an Accelsius value of around $850-950m. This is a moderate valuation for a company with >$50m in new orders last quarter, a strategic investor round, partnerships with NVIDIA, Dell, Johnson Controls and Legrand, and a deployment that is expected to be the largest of its kind in the world.
This is not a low-risk stock. The holding structure, early stage development and dilution history are real risks. But this complexity is probably why the valuation does not yet reflect the potential value of Accelsius.
