← AI INFRASTRUCTURE
2-PHASE DIRECT-TO-CHIP

Engineering the Future of Thermal Management.

DARKNX is removing the thermal barriers to AI scale by deploying the leading edge of cooling architecture across our entire data center portfolio, regardless of state of matter.

In partnership with Accelsius, the leader in advanced thermal technologies, every DARKNX facility is purpose-built to handle the extreme heat profiles of next-generation GPU and AI clusters.

View Partners →
Honeywell 515B Refrigerant Specifications
PDF ↓
Single-Phase vs 2-Phase Cooling Comparison
PDF ↓
250kW
PER RACK, MR250
4,500W+
CHIP TDP TESTED
1.08
ACHIEVABLE PUE
35–44%
OPEX SAVINGS VS SINGLE-PHASE
STRATEGIC PARTNERSHIP

Accelsius × DARKNX

As the physical requirements of high-performance computing evolve, the limits of air cooling have become the limits of innovation. DARKNX is removing those barriers across our entire portfolio of data center projects by partnering with Accelsius, the leader in advanced thermal technologies.

Together, we are moving away from traditional infrastructure to the leading edge of cooling architecture, ensuring that every DARKNX facility is purpose-built to handle the extreme heat profiles of next-generation GPU and AI clusters.

"DARKNX and Accelsius: Redefining the thermal envelope to power the global AI economy."

Accelsius
TECHNOLOGY PLATFORM
Accelsius NeuCool® Two-Phase Direct-to-Chip Cooling System
DEPLOYMENT SCOPE
All DARKNX greenfield, brownfield, and modular data center projects
TECHNOLOGY PLATFORM

Introducing the NeuCool® MR250

Accelsius's first row-based Coolant Distribution Unit purpose-built for AI and HPC at scale. The MR250 delivers 250kW of liquid cooling per rack in flexible configurations, 1×250kW or 2×125kW, with industry-leading thermal resistance and zero ozone depletion potential.

Accelsius NeuCool Two-Phase Cooling
Accelsius 2-Phase Cooling
Accelsius 2-Phase Cooling
NeuCool MR250
ACCELSIUS ↗
The future is liquid. Two-Phase Direct-to-Chip.
THERMAL RESISTANCE
0.020°C/W
Engineered to cool the thermal load of the Rubin GPUs
CDU CAPACITY
250kW / RACK
Configurable as 1×250kW or 2×125kW per rack
5-YEAR TCO SAVINGS
8–17%
vs. equivalent single-phase liquid cooling deployments
ENVIRONMENTAL
ZERO ODP
Waterless, low-pressure dielectric refrigerant, zero ozone depletion potential
SYSTEM CYCLE

How the Closed-Loop Refrigerant Cycle Works

01
VAPORIZE

Dielectric refrigerant flows through vaporators (cold plates) mounted directly to GPU and CPU hot spots. Heat from the chip causes the refrigerant to nucleate and vaporize.

02
TRANSPORT

The vapor, now carrying the chip's heat as latent energy, travels through insulated lines from the server rack to the row-based Coolant Distribution Unit (CDU).

03
CONDENSE

Inside the NeuCool MR250 CDU, the refrigerant vapor condenses back into liquid, releasing its heat to facility cooling infrastructure via water-cooled doors, dry coolers, or other heat rejection methods.

04
RETURN

The cooled liquid refrigerant returns to the vaporators at the chips, completing the closed loop. No water ever contacts the IT equipment, zero corrosion risk, zero leak contamination.

Learn more about NeuCool® technology directly from Accelsius: VISIT ACCELSIUS.COM →
TECHNOLOGY

The Power of Two-Phase Cooling.

Unlike standard liquid cooling, which simply moves heat through a fluid, two-phase cooling utilizes the latent heat of vaporization. A specialized non-conductive dielectric refrigerant boils at the chip surface, absorbing exponentially more energy than water alone, enabling a new standard for data center efficiency and power density.

MEGAWATT-LEVEL DENSITIES

Our infrastructure is engineered to support megawatt-level rack densities, catering specifically to the extreme demands of hyperscalers and enterprise AI leaders.

By supporting rack densities of 250kW and beyond, up to multi-MW, we enable the consolidation of massive compute into a significantly smaller, more efficient footprint.

SUPERIOR THERMAL STABILITY

The NeuCool® system demonstrates industry-leading cooling capability of up to 4,500W per socket. By using a two-phase cooling process, the system leverages the latent heat of vaporization to remove significantly more heat from the chip than conventional liquid or air-cooling approaches.

THERMAL RESISTANCE
0.020°C/W
SUSTAINABLE SCALING

Two-phase systems operate effectively with facility water temperatures up to 8°C higher than competing solutions, broadening conditions for chiller-less, free-cooling operation. PUE of 1.08 is achievable, with near-zero water consumption eliminating evaporative towers entirely.

ACHIEVABLE PUE
1.08
CAPABILITIES

Standardized. High-Performance.

SYSTEM SPECIFICATIONS
PRODUCTNeuCool® MR250 CDU
COOLING TECHNOLOGYTwo-Phase Direct-to-Chip
CDU CAPACITY250kW per rack (1×250 or 2×125)
THERMAL RESISTANCE0.020°C/W at 700W+ TDP
CHIP TDP TESTED4,500W+ (CPU, GPU, hot components)
COOLANTWaterless, low-pressure dielectric
OZONE DEPLETIONZero ODP
ACHIEVABLE PUE1.08
ANNUAL OPEX SAVINGS35–44% vs single-phase
5-YEAR TCO SAVINGS8–17% vs single-phase
FREE COOLING ADVANTAGE+8°C facility water tolerance
GREENFIELD & BROWNFIELD READY

From retrofitting existing spaces to massive greenfield developments, the Accelsius partnership allows DARKNX to deploy modular, future-proof cooling that scales as chip wattages continue to climb.

STANDARDIZED ACROSS THE PORTFOLIO

Across our diverse project landscape, the integration of Accelsius NeuCool® technology provides a standardized, high-performance cooling stack, consistent across every DARKNX deployment for simplified operations and maintenance.

SMALLER FOOTPRINT, GREATER DENSITY

Two-phase cooling allows DARKNX to consolidate massive compute power into a significantly smaller, more efficient footprint, unlocking new possibilities for urban edge deployments and dense hyperscale campuses.

TECHNICAL DOWNLOADS
Honeywell 515B Refrigerant Specifications
PDF — Refrigerant spec sheet
DOWNLOAD →
Single-Phase vs 2-Phase Cooling Comparison
PDF — TCO infographic
DOWNLOAD →
GET IN TOUCH

Ready to Deploy Next-Generation Cooling?

AI workloads are pushing chip wattages beyond what legacy infrastructure can handle. DARKNX and Accelsius are engineering to that future today.

Ready to help, when you need it.