Cerebras Condor Galaxy (Colovore Santa Clara)

The story

In Silicon Valley, standard AI data centers struggle to cool server racks that exceed 20 kW. In Santa Clara, liquid-cooling colocation specialist Colovore teamed with chip innovator Cerebras Systems and Abu Dhabi's G42 to host Condor Galaxy. Unlike standard clusters that wire together thousands of discrete GPUs, Cerebras uses the Wafer-Scale Engine (WSE-3)—an entire 300 mm silicon wafer holding 900,000 AI cores and 4 trillion transistors on a single chip. Circulating chilled water directly through custom high-density manifolds, the Santa Clara facility powers dinner-plate-sized AI chips.

Notable

• Houses Cerebras Wafer-Scale Engines (WSE), each wafer cut from a single 300 mm silicon ingot with 4 trillion transistors.
• Installed inside Colovore's liquid-cooled data center in Santa Clara, supporting extreme power densities of 35–50 kW per rack.
• Part of the Condor Galaxy network built in partnership with UAE AI holding company G42.
• Delivers exaflops of AI compute without the inter-chip communication bottlenecks of traditional discrete GPUs.

Expansions

• 2023: Condor Galaxy 1 announced and deployed with 64 Cerebras CS-2 wafer-scale systems
• 2024: upgraded with next-generation CS-3 wafer-scale engines delivering 4 exaflops of AI compute
• 2024: linked with sister clusters Condor Galaxy 2 and 3 in Texas and internationally