NEW — Meridian Inference chooses Infinium Edge to deliver AI inference with less power, no water, and at a lower cost per token. Read the announcement →
Next generation cooling — built for AI

Next-gen compute,
at the thermal edge.

Infinium Edge designs and operates next-generation immersion-cooled data center platforms — engineered to run modern AI silicon at the edge of its thermal envelope, where density, throughput, and efficiency peak.

Explore the platform → Get in touch
~1.05
Target PUE
250 kW
Per rack — up to
Months
Not years to deploy
EDGE THERMAL VECTORING™ MAX GPU UTILIZATION : DEPLOYED

Running today's top silicon — no thermal compromises

NVIDIA
Blackwell · Hopper
AMD
Instinct GPUs · EPYC CPUs
Intel
Gaudi GPUs · Xeon CPUs
Samsung
DDR Memory · SSDs
Micron
DDR Memory · SSDs
+ more
Custom silicon welcome
Why now

Chip thermals are going exponential.
The old playbook is out of room.

AI accelerators have crossed a thermal threshold the industry hasn't seen before. Chip TDP is climbing exponentially — and each generation pushes harder against the limits of the previous cooling approach.

Air cooling already hit its ceiling. Direct-to-chip extended the runway, but the next wave of silicon is going to break through that, too. The industry needs a cooling layer engineered for what's coming — not what shipped last cycle.

1ST

Tipping point reached. Air cooling can't keep up with modern AI accelerators. Direct-to-chip becomes the new floor.

2ND

Tipping point approaching. Next-gen silicon TDP exceeds what DTC can handle. Edge Thermal Vectoring™ is engineered to solve it.

AIR COOLING DTC COOLING EDGE THERMAL VECTORING™ EARLY GPUs TODAY NEXT-GEN LOW EXTREME CHIP TDP (W) 1st tipping point 2nd tipping point

Chip TDP trajectory · cooling-regime overlay

The platform

A clean-sheet data center,
engineered for accelerated compute.

Every Infinium Edge module is factory-built and cooled by Edge Thermal Vectoring™ — engineered around the world's purest, most thermally stable dielectric fluids. No GPU throttling. No compromise.

Edge Module Infinium Edge module — full exterior

Edge Module — built for Edge Thermal Vectoring™

A self-contained immersion platform. Designed for densities legacy air-cooled halls can't touch — built in a factory, shipped to site, quick to energize.

Cooling Tank Edge Thermal Vectoring immersion cooling tank

Immersion-cooled compute

Single-phase dielectric immersion captures heat at the source. No chillers, no hot aisle, no compromise on rack density.

Control Edge module control panel and HMI

Integrated controls & safety

HMI, PLC, and fluid loop monitoring — engineered as one system. Commissioned at the factory, telemetry-ready on day one.

Rack density
250kW
up to per rack
Target efficiency
1.05PUE
industry-leading
Water use
Zero
to low — by design
Time to live
Months
not years to deploy
Solutions

One modular platform. Three ways in.

Deploy multiple 1 MW units at industrial or commercial sites, scale into 10 MW distributed edge deployments, and grow into hyperscale campuses — all on the same factory-built immersion platform.

Edge Thermal Vectoring™
Deploy at industrial / commercial
Infinium Edge / Vector 1
1MW

Modular 1 MW systems for industrial & commercial co-location.

FormFactory-built, plug-and-play module
SitingIndoor or pad-mounted outdoors
CoolingDry coolers, 47.5°C inlet — no water
ScalingAdd blocks to grow capacity in place
Best forBehind-the-meter, land-and-expand
Distributed edge
Infinium Edge / Vector 10
10MW

Distribution-substation systems for regional AI build-outs.

FormSubstation-class deployments
SitingPower-first, below transmission limits
Speed< 9 months build-to-power
BuildGreenfield or brownfield retrofit
Best forDistributed regional AI capacity
Campus scale
Infinium Edge / Hyperscale
100MW+

Campus-scale AI factories, co-developed with EPC partners.

Form100 MW+ campus designs
ModelWholesale capacity & tech licensing
LicenseEdge Thermal Vectoring™ + designs
DeliveryCo-developed via EPC partners
Best forWholesale, hyperscale, model labs
Why it matters

Built for the thermal demands of AI —
at every metric that counts.

Edge Thermal Vectoring™ delivers step-change improvements over both air-cooled and direct-to-chip platforms — in density, footprint, efficiency, and water use.

Compute density
3–4×
More compute per rack
vs. 30–50 kW/rack air-cooled
Footprint
70%
Less white space
~200 vs. ~1,280 sq ft/MW air
Efficiency
1.05
PUE — lowest in category
Water use: low or none*

Cooling technology comparison

PUE = Power Usage Effectiveness · WUE = Water Usage Effectiveness
Cooling method Rack density White space / MW Avg PUE Avg water use
Air cooling 30 – 50 kW ~1,280 sq ft 1.5 – 1.8 High
Direct-to-chip (DTC) 60 – 200+ kW ~350 – 500 sq ft 1.1 – 1.2 Medium
Infinium Edge™ Cooling 250+ kW ~150 – 250 sq ft ~1.05* Zero to low*

* Based on ideal design — actual PUE and water use depend on customer choice of secondary cooling and ambient conditions.

01

Thermal efficiency

20× greater thermal density than air. Modern GPUs run at max utilization with no thermal throttling.

02

Power use

PUE ~1.05 with waste-heat capture optionality. Industry average is ~1.5; today's state-of-the-art sits around 1.15.

03

Water use

Eliminate or dramatically reduce freshwater use. Water-free designs available at 47.5°C inlet.

04

Footprint & capex

Up to 70% smaller footprint and lower total capital cost — including IT infrastructure.

05

Deployment

Factory-built, pre-integrated modules. Faster to deploy, simpler to operate, easier to maintain.

06

Sustainability

Carbon-negative dielectric fluid and lower power draw — fewer emissions per useful FLOP.

Announcement · May 2026

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 — landing 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.

Compute offtaker
Meridian Inference
Platform & operator
Infinium Edge
Deployment model
Multi-site, rapid build
Talk to our team → Download press release
Infinium Edge platform tank with immersion-cooled servers
ScopeMultiple projects
SitesKey low-latency inference metros
PaceRapid deployment
EnergizationQ4 2026 — Q3 2027
OfftakerMeridian Inference
Operations

24/7 fleet operations,
telemetry-first by design.

Every module streams real-time health, thermal, and power telemetry to the Infinium NOC. Operators see every node, every loop, every kilowatt — and so do our customers.

01

Fleet-wide observability

GPU utilization, coolant temps, rack power, and node health — every site, every rack, sub-second resolution.

02

Predictive thermal control

Closed-loop coolant management automatically rebalances flow to keep silicon in its sweet spot under bursty AI workloads.

03

Customer API + dashboards

Customers see the same data we do. Streaming telemetry, SLA reporting, and incident timelines.

04

Tier-III equivalent uptime

Redundant power paths, dual coolant loops, hot-swappable everything. Built to keep training jobs running.

Infinium Edge NOC operator monitoring fleet dashboard
Get in touch

Looking for high-density
compute capacity?

We work with hyperscalers, model labs, and neoclouds to bring next-gen capacity online faster than anyone else in the industry. Tell us what you're building.

Contact partnerships → Tour the platform