The largest AI chip IPO in history is here! Soars 1,09% in intraday trading, market cap over $450bn

On May 15, U.S. AI chip companyCerebras SystemsOfficially listed on NASDAQ under the ticker symbol “CBRS”.

Cerebras was offered at $185 (roughly Rs. 1,255) per share and opened up89%to $350 (roughly Rs. 2,375) per share, spiking to $386 (roughly Rs. 2,619) per share at one point during the session, rising109%, with a total market capitalization of $83.2 billion (about Rs. 564.2 billion).

As of today's U.S. stock close, its shares were trading at $311 (about Rs. 2,111) per share, up 68.151 TP4T, with a total market capitalization of67 billion dollars (about 454.3 billion yuan).

this isThe largest so far this yearIPOAnd alsoLargest IPO by a US tech company since Uber's 2019 IPOas well asLargest pure AI IPO in US stocks to date.

Cerebras is offering 30 million shares to raise $5.55 billion (RMB 37.66 billion). The underwriters also have the option to exercise their placing rights for an additional 4.5 million shares, bringing its total fundraising to a possible $6.38 billion (RMB 43.29 billion).

This year is ushering in a wave of historic AI-focused IPOs, with Wisdom Spectrum and MiniMax landing on the Hong Kong Stock Exchange at the start of the year, and the listing of Cerebras today further igniting the AI IPO boom, with heavyweight AI giants such as MuskSpaceX, OpenAI, and Anthropic scheduled to go public this year in the follow-up.

Compared to other Nvidia challengers, as of yesterday's close, the market value of A-share-listed Cambrian, Moore Threads, and Mu Xi shares were 792 billion yuan, 320.6 billion yuan, and 296.4 billion yuan, respectively, and the market value of Hong Kong-listed Tianshi Zhixin, and Wall Ren Technology were 127.1 billion Hong Kong dollars (about RMB 110.1 billion yuan), and 126.3 billion Hong Kong dollars (about RMB 109.4 billion yuan), respectively.

Cerebras was founded in 2015 and is credited with developing the dinner plate-sizedThe world's largest chipAfterward it became known.

The official website shows that its investors includeQualcomm, AMD, TSMCand other chip giants, with Trump's eldest son as a partner in the1789 Capitalas well asSam Altman, Co-Founder and CEO, OpenAI; Lip-Bu Tan, CEO, Intel; Adam D'Angelo, Co-Founder and CEO, Quoraand other well-known entrepreneurs.

OpenAI, AWS, UAE tech company G42, UAE Artificial Intelligence University MBZUAIAll are important customers of Cerebras.

Cerebras has built the world's first and only commercially available wafer-scale processor, the Wafer-Scale Engine (WSE), and calls its third-generation AI chip, the WSE-3, “The biggest and fastest AI chip ever brought to market in high volume".

WSE-3 utilizes5nmProcess technology, stuffed at 46225mm².4 trillion.Transistors, integratedNine hundred thousand.Computing Core,44GBOn-chip memory and21PBof memory bandwidth, an area that is as large as the NVIDIA B200's58 times.

Compared to the NVIDIA B200 package, which contains two separate chips, the WSE-3 has a19 timesof the transistor,250 timesThe on-chip memory capacity and2625 timesof memory bandwidth.

According to the IPO document, its revenue increased from $24.6 million (roughly Rs. 167 million) in 2022 to $510 million (roughly Rs. 3.5 billion) in 2025, a growth of more than19 times.

In 2025, Cerebras achieves a turnaround and a net profit ofUS$238 million (about RMB 1.6 billion).

First, the former AMD vice president founded, had refused to be acquired by Musk Masayoshi Sun

Founded in 2015, Cerebras was co-founded by Andrew Feldman, Sean Lie and others and is based in California.

Andrew Feldman, a Stanford graduate who sold server company SeaMicro to AMD in 2012 and then served as AMD's vice president and general manager, is now CEO of Cerebras.

A graduate of MIT, Sean Lie has served as Chief Hardware Architect for AMD's data center server solutions, and since April 2022 has been CTO of Cerebras.

Cerebras COO Dhiraj Mallick is a graduate of Stanford University and has served as Vice President of Architecture for Intel's Data Center Business Unit and Vice President and General Manager of AMD's Server Solutions Division.

Back in 2017, Musk considered acquiring Cerebras while he was at the helm of OpenAI, and the original plan was for OpenAI and Tesla to work together on the deal, but it was rejected by Cerebras' founding team.

According to foreign media this week, citing sources familiar with the matter, Masayoshi Son's SoftBank Group of Japan and its holding of British semiconductor IP giant Arm, had made an offer to Cerebras a few weeks before the launch of its IPO, and the offer was also rejected by Cerebras.

Second, built the largest AI chip in history, running large models at speeds far exceeding GPUs

While most chip design companies choose to cut wafers into hundreds of individual chips, Cerebras takes a different approach by making an entire wafer into a single chip.

The entire compute and storage resources are placed on a single giant chip, and the communication between the cores is interconnected on-chip, with latency so low as to be almost negligible. This cracks the bottleneck of time-consuming and energy-consuming data handling.

In August 2019, Cerebras“ first wafer-scale chip, the WSE, set the global tech world ablaze with the title of ”World's Largest Chip," which is made from a single wafer and utilizes TSMC's 16nm process, integrating 400,000 AI cores and 1.2 trillion transistors in an area of 46,225mm².

The second-generation WSE-2, launched in 2021, goes one step further by adopting TSMC's 7nm process and integrating 850,000 AI cores and 2.6 trillion transistors, and CS-2, an AI supercomputing system equipped with the WSE-2 chip, is also released at the same time.

In March 2024, the company unveiled its third-generation wafer-scale chip, the WSE-3, and AI supercomputer, the CS-3. The WSE-3 is built on TSMC's 5nm process and has 900,000 AI cores and 4 trillion transistors.

The WSE-3 is deployed within the CS-3 system, and multiple CS-3 systems can be connected to form a Cerebras AI supercomputer that can be used as a single logic computer for large-scale training and reasoning.

The WSE chip is made by manufacturing giantTaiwan Semiconductor Manufacturing Company (TSMC)Foundry production, packaging and testing manufacturerssun and moonlightSpecial processes are handled, culminating in final wafer packaging, assembly and testing at Cerebras' facility in California, USA.

The IPO filing shows that for many workloads, Cerebras runs open source models such as Llama, GPT-OSS, GLM, MiniMax, Qwen, etc. faster than leading GPU-based solutions, according to benchmarks15 times; in some of the more specialized workloads, Cerebras are fastMore than 1000 times.

As of December 31, 2025, Cerebras had 708 employees, 426 of whom were in the United States. As of March 31, 2026, the company had 96 issued patents and 50 patents pending worldwide.

Three,Revenue Jumps 76% in 2025, Turning a Profit

In 2022, 2023, 2024, and 2025, Cerebras' full-year revenues will be $24.6 million, $78.7 million, $290 million, and $510 million, respectively; net income will be $178 million, $127 million, $482 million, and $238 million, respectively, and R&D expenses will be $155 million, respectively, US$140 million, US$158 million, and US$243 million, respectively.

▲Changes in Revenue, Net Income, and R&D Expenditures of Cerebras from 2022 to 2025

During the same period, its gross margins were 11.7%, 33.5%, 42.3%, and 39.0%, respectively.

Cerebras' 2024 and 2025 GAAP net income and non-GAAP net income are as follows:

As of December 31, 2025, the company's accumulated losses wereUS$905 million (about RMB 6.141 billion).

Cerebras delivers solutions to its customers in a variety of ways, providing both AI supercomputers and high-speed inference services on the cloud, as well as co-developing AI products with its customers.

By business, in 2025, Cerebras' share of hardware revenues will be about 701 TP4T, and its share of cloud services and other services revenues will be close to 301 TP4T.

As seen in the table below, the company's turnaround to profitability in 2025 is largely dependent on realizing a net profit of approximately $310 million in the second quarter of 2025.

Fourth, the Middle East is a big gold master, OpenAI, AWS are customers

Cerebras' strong growth in revenue and net profit is closely linked to the Middle East.

In 2024, Group 42 Holding Ltd, an Abu Dhabi technology group, together with its subsidiary G42, contributed Cerebras up to85%of revenue.

In 2025, MBZUAI, the UAE University of Artificial Intelligence, contributes 621 TP4T of Cerebras' revenue, and G42 contributes 241 TP4T, with the two combined contributing86%of revenue.

This indicates that over the past two years, Cerebras has exceededeighty percentof its revenues are from the Middle East, which is a core source of contribution to its 2025 turnaround.

The change occurs in 2026.

In January, Cerebras announced a deal with OpenAI worth overUS$20 billion (approximately RMB 136.3 billion)a multi-year agreement to provide OpenAI with 750 megawatts of AI arithmetic, to be deployed in tranches from 2026 through 2028.OpenAI has also agreed to pay CerebrasUS$1 billion (about RMB 6.8 billion)to fund the development of data centers that run their AI products.

In March, Cerebras entered into a multi-year partnership with Amazon Cloud Technologies (AWS) to scale Rapid Reasoning to a larger scale through global distribution.

Two new collaborations solidify investor confidence in Cerebras' growth prospects. If these partnerships are successfully executed, Cerebras will have little to worry about in terms of revenue for the next three years.

Conclusion: AI reasoning market expects the “next NVIDIA” to emerge”

As demand for AI arithmetic continues to skyrocket, AI chip companies have challenged the overlord, NVIDIA.

As of the close of the U.S. stock market today, NVIDIA's total market capitalization was $5.71 trillion (about RMB 38.74 trillion), holding on to the throne of the world's highest market value.

In the field of AI training chips, NVIDIA GPUs occupy the vast majority of the market and are difficult to match.

The AI inference chip market, on the other hand, is a hundred times more competitive, with several companies judging that GPUs are not the best form of hardware and trying to explore innovative chip designs that are more suitable for large-scale and efficient AI inference.

In 2025, Cerebras revenue will be only $510 million (roughly Rs. 3,460 million), a far cry from NVIDIA's $215.9 billion (roughly Rs. 1.46 trillion) in revenue.

But the two endorsements, OpenAI and AWS, have opened the market to a new imagination.

AI infrastructure diversification has finally come to a stage where the U.S. capital markets are willing to bet on it.

Source: Smart Stuff