We design our own servers and racks in-house to optimize for efficiency, reliability, longevity and ease of operation. Compared to similar off-the-shelf models, our servers consume around 20% less energy under the same conditions — adding up to nearly 10 GWh of energy saved in 2024 alone.

Servers are cooled with a smart air-based system that isolates airflow for CPUs and GPUs, with intelligent fan controls that dial speed up or down depending on thermal needs. As a result, they run reliably at temperatures up to 40°C (105°F) and are operated without mechanical cooling, reducing power demand and eliminating the use of water and refrigerants. That extra headroom also means we can fit more servers into each megawatt of connected power, making better use of available capacity.

By focusing on simplified, durable architecture, we extend the lifespan of components and reduce the need for replacements, lowering demands for new materials and helping cut emissions across the value chain. For example, our cableless design eliminates an average of 7 meters of cable per server, reducing both material use and assembly complexity.

The hardware is designed not only for performance, but for long-term operational simplicity. Tool-free assembly and cableless connections make maintenance quicker and safer, while reducing the risk of errors.

All of these elements come together to lower wear and tear, extend component life and help lower the cost and environmental footprint of compute over time.