As artificial intelligence models become ever more complex, traditional hardware struggles to keep up. GPUs, designed originally for graphics, are increasingly seen as a bottleneck in scale and speed. Cerebras Systems confronted this head-on with its Wafer Scale Engine, a chip architecture purpose-built for AI workloads.
The company just closed a $1.1 billion Series G funding round at an $8.1 billion valuation, led by Fidelity Management with $700 million, marking one of the firm’s biggest private tech investments. Other major investors include Atreides Management, Tiger Global, and 1789 Capital, along with longtime backers Altimeter, Alphawave, and Benchmark.
This fresh capital will power the company’s relentless innovation, ramp up U.S. manufacturing, and expand its AI data centre footprint globally.
Breaking the limitations of legacy AI hardware
Andrew Feldman and Gary Lauterbach, Cerebras’ founders, bring years of experience in high-performance computing and chip design, having previously worked together on SeaMicro before its acquisition by AMD.
They launched Cerebras with a clear mandate: to break the limitations of legacy AI hardware and deliver a processor that could radically accelerate AI development without compromise. The vision is simple yet ambitious: to make AI “blisteringly fast,” unleashing new possibilities in research, medicine, defence, and industry.
At the heart of Cerebras’ offering is the Wafer Scale Engine 3 (WSE-3), a chip that dwarfs conventional GPUs by 56 times in size and delivers AI inference and training speeds over 20 times faster than the competition. It achieves this while being more power efficient and packing 7,000 times the memory bandwidth of typical GPUs.
The sheer scale of wafer-scale integration means entire AI models can reside on one chip, dramatically cutting latency and overhead. Since launching its cloud-based AI inference service in late 2024, Cerebras has quickly become the go-to choice for major players like AWS, Meta, IBM, Mistral, and government agencies.
The system now processes trillions of tokens every day on premises and across partner platforms, cementing its leadership in speed and scale.
What’s next?
With this substantial funding, Cerebras is set to expand its wafer-scale manufacturing capabilities within the U.S. and open additional AI data centres across North America and Europe. The company plans to push the envelope further with advances in co-packaged optics, new packaging technologies, and system design enhancements.
Although initially aiming for an IPO in 2024, Cerebras remains focused on public listing plans for 2025 amid regulatory hurdles. Its goal remains clear: to meet the surging global demand for high-speed AI processing and redefine the future of AI infrastructure.
