What you need to know about Nvidia competitor Cerebras after wild IPO

Cerebras Systems made headlines with a spectacular debut on the stock market, marking one of the largest IPOs in the tech industry. The company's impressive launch underscores the escalating demand for chips designed to power artificial intelligence, as major tech firms search for alternatives to the increasingly scarce and expensive graphics processing units (GPUs) produced by Nvidia. By the end of its first trading day, Cerebras achieved a market capitalization nearing $100 billion, placing it among an elite group of companies that have surpassed this milestone, alongside giants like Meta and Alibaba. However, the stock experienced a decline on its first full trading day, raising eyebrows among investors. Cerebras distinguishes itself from Nvidia by manufacturing an unconventional chip, which is notably larger than traditional options. "We build the largest chips in the semiconductor sector," stated Andrew Feldman, CEO and Co-Founder of Cerebras, during an interview on CNBC. He emphasized that larger chips can process more data in shorter timeframes, enhancing performance. Historically, Nvidia has dominated the AI chip landscape due to its versatile GPUs, which are adept at handling the parallel computations necessary for training expansive models. Yet, the landscape is shifting towards agentic AI, where inference becomes increasingly critical. While training involves teaching AI to recognize patterns from vast datasets, inference is about leveraging that knowledge to make decisions based on new input. This process can be efficiently executed on less powerful, specialized chips such as Cerebras' WSE-3, which belongs to the category of application-specific integrated circuits (ASICs). Cerebras reports that its WSE-3 chip is 57 times the size of the largest available GPU and contains 50 times more transistors. The most advanced AI chips are produced using Taiwan Semiconductor Manufacturing Company’s (TSMC) cutting-edge 2-nanometer technology, which is currently exclusive to Taiwan. In contrast, Cerebras utilizes TSMC's 5-nanometer technology, which, while less advanced, still enables significant performance capabilities. Founded in 2016 in Silicon Valley, Cerebras initially aimed for an IPO in 2024 but retracted its application due to concerns regarding its dependence on a single client, the Microsoft-backed AI firm G42 in the UAE. Following the successful IPO, Feldman and co-founder Sean Lie have reportedly become billionaires based on their equity stakes in the company. Cerebras has transitioned from primarily selling chips to operating them within its own data centers as a cloud service, positioning itself in direct competition with cloud service providers like Google, Microsoft, Oracle, and CoreWeave. Notably, Cerebras has entered into a $20 billion cloud agreement with OpenAI, set to last until 2028, and Amazon Web Services announced its integration of Cerebras chips in March. Bob Komin, the CFO of Cerebras, highlighted the overwhelming demand for their fast inference products, stating that their primary challenge lies in meeting this demand, with orders already sold out through 2027. While many hyperscale companies are developing their own ASICs, Cerebras competes closely with specialized firms like Groq, which Nvidia recently acquired for $20 billion. Other competitors such as SambaNova and D-Matrix are also eyeing the burgeoning AI chip market, with SambaNova attracting significant clients, including Hugging Face and Meta. Cerebras' successful IPO not only solidifies its place in the market but also opens the door for other ASIC startups, such as Rebellions, which recently raised $400 million in preparation for its own IPO. The race in the AI chip sector is heating up, and Cerebras is poised to be a key player as demand continues to surge.