🔎Cerebras is finally going public.
A quick update on possibly the hottest IPO of the year.
TLDR Summary
NVIDIA originally developed GPUs for video game graphics. It’s a coincidence that this technology turned out to be so powerful for the deep learning that we simply refer to as ‘AI’ today. Therefore, GPUs are likely not the most optimal hardware to run AI on.
Video game graphics require a ton of parallel processing. That’s helpful for AI as well. But it’s not necessarily the limiting factor for AI chip performance. Cerebras management contends that the bottleneck for AI is in fact latency. Since communication is thousands of times faster on-chip than across chips, they developed a huge chip that is made from an entire wafer.
Getting that done with an acceptable yield is extremely difficult. The probability of a chip defect scales exponentially chip size. Power and heat management is challenging as well. It seems that Cerebras has successfully overcome these challenges. Their chip is greatly outperforming NVIDIA in tests on popular open-source models. Partnerships with leading AI companies such as OpenAI and Amazon add to credibility.
The company is at a very early stage. But if they can scale the business properly, they are the first who can potentially challenge NVIDIA’s dominant position. A monster narrative might be in the making here with trillions of Dollars of market capitalization at stake.
They have started their investor roadshow today and they plan to raise up to $4bn at a $40bn valuation. This would value the company 10x higher than the original IPO plan from two years ago. Momentum is clearly on their side.
Related recent content from Fallacy Alarm
Original IPO plan
In 2024, Cerebras announced its intention to go public when they filed an S-1. I covered that attempt in the article below:










