Meridian Inference chooses Infinium Edge to deliver AI inference with less power, no water, and at a lower cost per token.
A multi-project deployment agreement between Meridian Inference and Infinium Edge lands Edge Thermal Vectoring™ capacity at key low-latency inference sites for rapid build-out. First sites energize in late 2026, with capacity expanding across multiple U.S. metros through 2027.
The Challenge
AI inference workloads demand sustained, high-throughput compute at the lowest possible cost per token — and the economics are tightly tied to energy efficiency. Conventional air-cooled and direct-to-chip approaches impose a ceiling on rack density and drive up cooling overhead, making it harder for inference providers to operate competitively at scale.
Meridian Inference evaluated the full range of infrastructure options and identified immersion cooling as the path to meaningfully lower power draw, zero water consumption, and a better cost structure on every inference request served.
The Solution
Infinium Edge™ delivers a factory-built, immersion-cooled compute platform optimized for exactly this workload profile. Key platform characteristics that drove Meridian Inference's selection:
Less power per token
Target PUE of ~1.05 vs. the industry average of ~1.5 — meaningfully more of every watt goes to useful compute rather than cooling overhead.
No water
Closed-loop immersion cooling eliminates evaporative water consumption entirely, enabling siting in water-constrained markets and simplifying environmental permitting.
Higher rack density
Up to 250 kW per rack enables far more compute capacity per square foot, reducing the total real estate and infrastructure footprint required to serve a given inference load.
Speed to market
Factory-built, modular Edge modules deploy in months, not years — letting Meridian Inference add capacity in step with demand without waiting on long construction cycles.
Deployment Details
The agreement covers multiple projects across key low-latency inference metros in the United States. First sites are scheduled to energize in late 2026, with capacity expansion continuing across additional U.S. locations through 2027. Each deployment uses Infinium Edge's modular platform, sized and configured for the compute density and power profiles Meridian Inference requires.
Looking Ahead
This partnership reflects a broader shift in how inference infrastructure is being procured and built. As cost-per-token economics become the defining competitive variable for AI inference providers, the efficiency, density, and speed advantages of immersion cooling move from “interesting” to “necessary.”
Infinium Edge is built for exactly this moment — and the Meridian Inference partnership is the first of multiple deployments being brought online across the platform.
Interested in a deployment?
Talk to our team about bringing Edge Thermal Vectoring™ to your infrastructure.
Contact partnerships →