AI Chip Sector Adds Heavyweight Player, Nvidia Rival Cerebras Plans to List Next Week, Targeting Largest US IPO of the Year

TradingKey - AI chipmaker Cerebras is reportedly planning to list on the Nasdaq on May 14 under the ticker "CBRS." This IPO is poised to become the largest new listing in the U.S. this year, bringing this formidable challenger to Nvidia ( NVDA) in the AI computing power race officially into the capital markets spotlight.

According to prospectus filings submitted this week, Cerebras plans to issue 28 million shares in its IPO with a price range of $115 to $125 per share. If priced at the upper limit, the offering would raise $3.5 billion, valuing the company at $26.6 billion. Including stock options and other equity, the fully diluted valuation would be approximately $33 billion.

Notably, sources familiar with the matter revealed that due to robust investor demand, Cerebras may raise its pricing range to $125–$135 per share as early as May 12. The IPO is currently more than 20 times oversubscribed.

This marks Cerebras' second attempt at an IPO. Last October, it withdrew its listing application due to a U.S. government review of an investment from Abu Dhabi-based client G42. The relaunch coincides with a surge in AI infrastructure investment and a recovery in the U.S. IPO market, significantly boosting market attention.

Why Cerebras dares to challenge Nvidia?

Founded in 2015 in California, this chip company has targeted a completely new technical route beyond GPUs since its inception—using an entire silicon wafer to build giant chips directly, rather than the traditional method of cutting wafers into multiple dies and then stitching them together. This wafer-scale architecture is precisely what gives it the confidence to challenge Nvidia.

Several of the company's core founders previously held key technical positions at leading chipmakers such as AMD, providing them with a profound understanding of GPU technical limitations and industry pain points. In a market environment where GPUs remain the mainstream solution for AI training and inference, Cerebras has taken an alternative path, addressing the core bottlenecks of traditional GPU architectures at the physical level through its Wafer-Scale Engine (WSE) chips.

Taking the latest WSE-3 chip as an example, it integrates 4 trillion transistors and 900,000 dedicated AI compute cores. With a chip area of 46,225 square millimeters, it is 56 times larger than Nvidia’s B200 chip; its on-chip SRAM memory is 250 times that of the B200, and its memory bandwidth is 2,625 times higher.

This design allows entire AI models to be stored completely on a single chip, fundamentally eliminating the latency caused by frequent data transfers between multiple chips in a GPU cluster. This provides a significant advantage when handling large model inference and ultra-scale computing demands.

Third-party test data shows that in Llama 4 Maverick 400B model inference scenarios, the Cerebras CS-3 system achieves a single-user response speed of 2,522 tokens per second, which is 2.4 times that of Nvidia’s B200. In the Llama 3.1 8B small model scenario, its speed is 20 times that of Nvidia’s H100, while the cost per token can be reduced by up to 80%.

Beyond hardware performance breakthroughs, Cerebras has built a comprehensive solution ranging from dedicated hardware to a software stack, launching supercomputing systems such as the CS-2 and CS-3. These support both on-premise deployment for customers and pay-as-you-go cloud services.

This flexible service model has attracted numerous heavyweight clients. In January 2026, the company signed a three-year cooperation agreement exceeding $20 billion with OpenAI to deploy 750 megawatts of computing power. In March, it partnered with AWS to launch the CS-3 system on Amazon Web Services, making it the first non-GPU AI accelerator to enter the supply chain of a major cloud provider. Furthermore, GlaxoSmithKline (GSK), the U.S. Department of Energy, and several national laboratories are also clients, providing multi-dimensional validation of its technical prowess.

Looking at market trends, the AI industry is shifting from a focus on training to a focus on inference. The global AI inference market reached $106.2 billion in 2025 and is expected to grow to $255 billion by 2030. Cerebras' technical advantages align precisely with this market demand.

Leveraging its outstanding performance in low-latency inference scenarios, the company has ranked first in inference calls on the HuggingFace developer platform. In 2025, revenue from its inference cloud services accounted for 30% of its total, successfully carving out a niche in the market dominated by Nvidia.