Cerebras Defies Nvidia Dominance, Unveils $150 Million AI Behemoth Cloud for G42

UNITED STATES: In a remarkable twist against Nvidia’s GPU supremacy, Cerebras Systems, the AI computing startup, has revealed the completion of its colossal AI computer, Condor Galaxy 1 (CG-1). 

The $150 million machine, a fusion of 64 CS-2 computers, marks a paradigm shift in the industry. 

– Advertisement –

Andrew Feldman, Cerebras CEO, explained that their business model has evolved into a hybrid, emphasizing cloud services over individual system sales.

G42, an investment company based in Abu Dhabi, commissions the CG-1, which is based at Colovore in California. Feldman sees this shift as a boon, describing it as a “very good business,” with Cerebras having a “monster year” in sales. 

– Advertisement –

The machine’s unique structure, combining 576 CS-2 computers and 654,000 AMD CPU cores, is expected to reach an unprecedented 36 exaFLOPs in its final configuration.

Cerebras’s novel approach involves a specialized fabric switch, Swarm-X, and a dedicated memory hub, Memory-X, connecting CS-2 computers. The Wafer-Scale-Engine (WSE), the world’s largest AI chip with 850,000 individual cores, stands at the core of each CS-2 computer. 

– Advertisement –

Feldman highlighted the efficiency advantage, stating, “It’s very hard to get efficient use of that much distributed computing, it’s a very, very difficult problem. That’s one of the problems we fundamentally solve.”

This shift towards larger cloud-based solutions has proven successful for Cerebras, with pharmaceutical giant GlaxoSmithKline utilizing their cloud alongside on-premise hardware. 

Feldman envisions expanding this model, emphasizing not just hardware sales but also the rate of new customers renting capacity in the Cerebras cloud without any upfront purchases.

The industry is closely watching Cerebras as it gets ready to reveal the next iteration, CG-2, which will have an additional 64 computers and 4 exaFLOPS and has the potential to completely change the landscape of AI computing.

Also Read: Atom Computing Breaks Quantum Barrier with 1,000+ Qubits for Fault-Tolerant Computing