Cerebras Systems, the AI computing startup, has announced a partnership with Abu Dhabi-based investment firm G42 to build “the world’s largest supercomputer for AI” called Condor Galaxy. The machine, focused on training neural networks, will be built in multiple stages and is expected to generate hundreds of millions of dollars in revenue for Cerebras. The initial configuration, CG-1, consists of 32 of Cerebras’s special-purpose AI computers, the CS-2, which collectively hold 27 million compute cores, 41 terabytes of memory, and 194 trillion bits per second of bandwidth.
Condor Galaxy runs at 2 exa-flops, making it capable of processing a billion billion floating-point operations per second. The machine is powered by Cerebras’s Wafer-Scale-Engine (WSE-2), the largest chip ever made, and is supplemented by the Swarm-X switch and Memory-X hub. While it claims to be the largest supercomputer for AI, there is no general registry for AI computer size, and comparisons to conventional supercomputers are not direct due to differences in precision. The machine is physically located at Colovore’s facilities in Santa Clara, California.
The partnership between Cerebras and G42 will scale through version CG-9, with phase 2 expected to double the CG-1’s footprint and phase 4 aiming to create a “constellation” of interconnected systems, resulting in a total capacity of 36 exa-flops. This partnership marks the first time Cerebras is not only building a clustered computer system but also operating it for the customer, providing multiple avenues for revenue.
Cerebras CEO Andrew Feldman expressed his optimism about the partnership, stating that it is larger than all other startup contracts combined and is expected to grow significantly in the future. The collaboration between Cerebras and G42 highlights the global nature of the AI phenomenon and the increasing demand for powerful AI computing solutions worldwide.