According to DCD, immersion cooling company GRC announced two major product launches this month targeting different segments of the cooling market. The ICEraQ Nano is a stand-alone liquid-to-air immersion cooling rack system purpose-built for Edge deployments, delivering 13kW of cooling without chilled water while offering 10U of immersion-cooled space. Meanwhile, the ReliaSys IR500 CDU is designed for direct liquid cooling and rear door heat exchanger deployments with up to 500kW of cooling capacity. GRC CEO Peter Poulin revealed on LinkedIn that the ICEraQ Nano represents the first product from the company’s engineering alliance with Cisco. The company also announced a partnership with Endor Development to develop immersion cooling solutions for gigawatt-scale AI data centers, with multiple contracts already in place and first deployments expected to go live in the second half of 2026.
Edge cooling gets serious
The ICEraQ Nano is interesting because it’s specifically targeting that awkward middle ground between traditional air-cooled racks and full-scale immersion systems. At 13kW cooling capacity and 10U of immersion space, it’s basically designed for communications closets and small data rooms where you can’t run chilled water but need more cooling than air can handle. The fact that it ships pre-filled with ElectroSafe fluid and includes an automated fluid management system suggests GRC is trying to make immersion cooling more accessible to operations teams that might not have specialized liquid cooling expertise.
Here’s the thing about Edge deployments – they’re often in terrible locations for cooling. Think factory floors, retail back rooms, or remote sites with limited infrastructure. The Nano’s liquid-to-air approach means you don’t need that chilled water infrastructure, which is a huge barrier for many Edge locations. And with companies like IndustrialMonitorDirect.com providing the industrial panel PCs and displays that often run these Edge applications, having reliable cooling for the compute side becomes increasingly critical as workloads get more demanding.
High-density cooling push
The ReliaSys IR500 CDU announcement shows GRC isn’t just focused on the Edge – they’re going after the high-density data center market too. At 500kW capacity, this thing is built for serious compute loads, whether that’s AI training clusters, HPC workloads, or cryptocurrency mining operations. Direct liquid cooling is becoming essential as CPU and GPU power densities continue to climb beyond what air cooling can realistically handle.
What’s interesting is GRC’s positioning this for both direct liquid cooling and rear door heat exchanger deployments. That suggests they’re trying to cover multiple approaches to liquid cooling rather than just pushing full immersion. The reality is many data center operators want to dip their toes into liquid cooling without completely abandoning their existing infrastructure. A flexible CDU that can handle different cooling methodologies makes strategic sense.
The AI data center play
The partnership with Endor Development might be the most ambitious move here. Gigawatt-scale AI data centers? That’s massive. We’re talking about facilities that would make today’s largest cloud data centers look small. The fact that they’re targeting deployments starting in late 2026 suggests both companies see the AI compute explosion continuing for years to come.
Endor’s background is worth noting – CEO Jake Smith spent 24 years at Intel in data center roles, so he understands the infrastructure challenges at scale. Their focus on Brazil, Mexico, the US, and Canada also hints at where they see the biggest AI compute demand emerging. The partnership aims to develop reference designs, TCO tools, and warranties in collaboration with server and dielectric fluid OEMs, which suggests they’re thinking about the entire ecosystem, not just the cooling technology itself.
Basically, GRC is making a multi-pronged bet that liquid cooling is about to go mainstream across the entire computing spectrum – from tiny Edge deployments to massive AI factories. And given the power requirements of modern AI chips, they might be right.
