Creating the most efficient AI infrastructure. Maximizing AI token output.
We build AI factories, utilizing peak energy efficiency to drive down TCO and maximise AI token output.
Up to
%
less capital required to build a Firmus AI Factory against a legacy DC
Up to
%
better energy TCO when running AI infrastructure and workloads
The AI Factory is a new class of data center. One built to maximise token output, with maximum profitability, with managed services built in.
Our physical approach is based on advanced, liquid everywhere data halls - utilizing our own immersion, 1-phase and 2-phase cold plate cooling platforms - with space and density optimized designs that cost up to 50% less per MW to build and deploy than legacy colocation facilities.
We overlay all this infrastructure with AI FactoryOS™, a GPU-to-energy grid management suite that controls all aspects of the physical AI factory from hardware and cluster control, maximising throughput and uptime.
We design, build and operate liquid-everywhere, TCO-optimized AI Factories. A complete managed platform.
An example configuration of a 2,048 GPU cluster in a legacy data center retrofit utilizing our 1-phase immersion solution. Using 61% less energy per FLOP than comparable air-cooled deployments, with 100% of the performance output - independently verified. See how it comes together.
The immersion data halls connect back to central air-cooled spine & leaf networking racks, with space for storage & infrastructure nodes. In a standard AI factory deployment, 2 x 128 node (256 node total) HyperCubes (for a total of 2,048 H100/H200/MI325X accelerators) share this common area.
World-first 400-800GBPS Firmus immersion-warranted OSFP & QSFP transceivers deliver full-rail cluster designs with ease.
We're the world leader in integrated, liquid-everywhere AI Factories.
37% more FLOPS per unit of energy vs free cooling. >50% more cost effective to build standalone than legacy colocation data centers. Future proof rack densities for multi-KW chips.
Our platform and suite of managed services are verifiably the most cost-effective way to build and operate AI Factories at scale.
Using advanced immersion or direct-to-chip cooling technologies, we've mastered service delivery to guarantee high uptime for any type of CPU or GPU cluster, unlocking cost and environmental benefits of our platform for everyone to use.
Flexible ways to access and benefit from our technology, today.
- Retrofitting
- New builds
- Managed infrastructure
- NCP GPU Cloud
Turn a legacy data center into a liquid-cooled hyperscale AI factory. Firmus Infrastructure Services can assess, design and execute retrofit projects to uplift aged or unsuitable data halls into capable AI Factories.
Greenfield Firmus AI Factories are future-proof, energy-dense, using half the energy, and costing half as much to build against legacy AI data centers. Our Global Engineering and Construction (GECO) organization can work with you to build ground-up AI data centers.
For retrofits or new builds, Firmus AI Factories come with managed infrastructure services that deliver frictionless access to our platform for operators and users. Firmus AI FactoryOS guarantees SLA uptime in order to maximise the throughput and profit of your AI Factory - AI tokens and model output.
SMC by Firmus is our globally available GPU cloud service, delivering AI services to innovators with performance, reliability and cost-efficiency.
From Availability Zones (AZs) in Singapore & Australia, we're enabling South-East Asia's access to an NVIDIA NCP reference architecture cloud stack, featuring H200 SXM GPUs, fast networking, and high security. Visit SMC to learn more.
