The perfect process flow sheet to turn electrons into AI calculations in the most efficient way possible. Made available globally at scale.
GPU infrastructure
AI training & inferencing instances, including full-rail H100 SXM nodes, connected via RDMA InfiniBand. Running within the HyperCube, SMC H100 instances are the most energy-efficient in the world. Up to 3kW of internal fans are not required, with further savings provided by DC power direct to the node.
Up to 36% of energy is saved at the node level
Sustainable AI Factories
HyperCube Sustainable AI Factories that increase density by >4x over legacy hosting infrastructure, and run at a sub 1.03 PUE. A complete GPU hosting platform, running NVIDIA's cutting-edge SXM platform, delivered with enterprise-grade SLAs.
A further 21% of energy is saved at the AI factory level.
Global availability
HyperCubes installed within a worldwide fleet of DCs provide security, connectivity and availability of SMC instances. Each DC with HyperCube sustainable AI factories provides the power, connectivity and waste condenser water required to deliver 24/7/365 GPU uptime.
Efficient GPU Clusters
Sustainable AI Factories
Globally Available
Zero-compromise, NVIDIA certified training & inference clusters.
SMC uses NVIDIA-certified HGX compute from global Tier 1 OEMs, optimized for maximum efficiency through dedicated engineering teams.
Thanks to the engineering of HyperCube, SMC servers, like the H100 SXM instance, require substantially less individual power to operate and cut e-waste in the process.
28% energy saving (3.5kW) from removal of fans
3% efficiency gain from 54v DC power direct to servers
35% (29kg) of e-waste (fans, cables, PSUs) saved per server
Slide controls left or right to see the before and after
Meet HyperCube
A purpose built, Sustainable AI Factory. The power behind SMC is engineered end-to-end to consume less energy, with zero compromise performance.
Sustainability, backed by facts.
Our technology-first approach to carbon reduction considers all power inputs, from the PUE of the data centre to the efficiency of the server itself. Reporting on PUE (Power Usage Effectiveness) is today easily gamed by the major clouds, as it fails to take into account power use in the server. SMC's platform saves energy both inside and outside the server. We report on total efficiency - down to the chip - to give a full picture of energy consumption.
Node level
H100 SXM-8 Instance.
Total power draw. Max Load.
10.04kW
Legacy
7.01kW
SMC
×
Data Centre Level
End-to-End PUE
1.47
Legacy
1.12
SMC
=
Total Power Consumption
14.05kW
Legacy
7.77kW
SMC
Cutting CO₂
Beware of greenwashers. This is technology-driven efficiency that delivers a nearly 50% carbon saving, with the data to prove it.
3,800kgper H100 per year
Case study
Singapore
At our launch AZ in Singapore, on average we’re experiencing a 47.5% reduction in energy consumption for H100 HGX-based workloads, versus a benchmark for running the exact same system in the legacy way (air cooled, in a typical Singapore-based data centre).
A 47.5% reduction in energy correlates 1:1 with the size of the workload’s carbon footprint. To see the scale of the issue, consider recent research on the total power required today to run AI workloads, which gives a current estimate of 29.3 terawatt-hours – the same as the entire power consumption of Ireland, or around 1.6m equivalent H100 GPUs.
Were that based in Singapore – where the direct consumption of power emits on average 417g of CO2 for every kWh of energy – this would generate 12.47mt (mega-tons) of CO2 per annum. Using common benchmarks, that’s the equivalent of over 2.7m passenger vehicles driving for a year, 635,000 US households, or the carbon-absorbing power of 577 million trees. That’s a lot of pollution.
Now, imagine this was all powered by SMC: that figure goes down to 6.3mt.
