How Carbice and DarkNX Target AI Data Centre Uptime

Carbice, a leader in vertically-aligned carbon nanotubes and manufacturing assembly solutions for advanced computing and power systems, has entered a strategic partnership with DarkNX to support the design and long-term performance of more than 300MW of AI-focused data centre capacity, beginning with a new campus in Ontario, Canada.

The collaboration positions Carbice as a system-level solutions partner within DarkNX’s digital infrastructure platform, bringing expertise in thermal interface materials, assembly processes and long term performance validation.

The companies say the focus is on improving reliability across chip-level cooling, power delivery and critical mechanical and electrical interfaces that influence uptime and hardware lifespan.

DarkNX is developing next-generation AI data centres optimised for high-density workloads. As compute densities increase, the integrity of thermal and electrical interfaces is becoming a constraint on sustained performance.

Addressing reliability at the interface

Under the partnership, Carbice will provide expertise and data collection through the Carbice Lab, which focuses on thermal interface reliability. The company will work alongside other technology partners selected for DarkNX’s 300MW campus, supporting performance from GPUs and CPUs through to switchgear and busbar joints.

"As AI infrastructure scales, reliability at the interfaces, including thermal, mechanical, and electrical, becomes a defining constraint," says Baratunde Cola, CEO and founder of Carbice. 

"This partnership is about bringing science-backed insight into how data centre systems perform on day one and how they continue to perform on day 1,000. Our US-manufactured solutions do not degrade with time; instead, they improve system reliability and extend the usable life of critical assets across GPUs, CPUs and power infrastructure."

Baratunde emphasises that interface degradation over time can lead to throttling, downtime and premature equipment replacement – issues that become more pronounced in high-density AI environments.

Carbice’s vertically aligned carbon nanotube technology is designed to maintain elasticity at the interface, enabling more accurate modelling and simulation during the design phase. The company says this allows engineers to predict real-world operating conditions more effectively and accelerate qualification cycles.

Supporting efficiency at 300MW scale

DarkNX’s Ontario campus is intended to deliver 300MW of AI capacity. At this scale, even marginal efficiency improvements can translate into significant energy savings.

According to Carbice, its cooling technologies can prevent performance drift and avoid nearly 30 GWh of unnecessary electricity consumption over five years. For a single 300MW AI campus, this equates to preventing approximately 10,500 metric tonnes of CO₂ emissions over that period.

By designing out degradation driven inefficiencies from the outset, DarkNX aims to reduce throttling, minimise rework and limit downtime. The companies state that this approach could avoid millions in energy and reliability related costs annually while lowering lifecycle emissions across the campus.

“Our approach is to be technology-agnostic while assembling the best available solutions to deliver performance, efficiency, and durability at scale,” says Isaac Islam, CEO of DarkNX. 

“Carbice brings a unique combination of materials science and system-level expertise that strengthens our ability to design and operate AI data centers built for long-term performance.”

Isaac highlights that maintaining performance consistency over years of operation is as important as achieving peak efficiency at commissioning.

Extending beyond chip cooling

While much attention in AI data centres centres on chip-level cooling, the partnership also addresses power delivery infrastructure. Carbice’s expertise includes mitigating torque loss and inconsistent torque application in switchgear and busbar joints – issues that can affect electrical reliability and safety.

By applying advanced interface materials and validation processes to these components, DarkNX aims to deliver a more resilient full-stack data centre solution. This approach aligns material science with manufacturing and qualification expertise to reduce operational risk.

Carbice will supply its Dow-Carbice SW and SA Series Pads, which have been qualified with OEMs across CPUs, GPUs and UPS systems. Through the Carbice Lab, the company provides testing datasets that replicate real-world operating environments, enabling customers and OEMs to validate solutions before large-scale deployment.

For DarkNX, integrating these capabilities into the design phase is intended to strengthen long-term asset performance across its AI infrastructure portfolio. As workloads intensify and rack densities continue to climb, attention to interface integrity – from chip to switchgear – is emerging as a key factor in sustaining uptime across large-scale AI data centre campuses.