Cerebras eyes $3.5 billion raise in Nasdaq IPO and heats up the AI chip race
The race for artificial intelligence chips is seriously heating up, and Cerebras just showed up ready to claim a major spot in the competition.
The American AI chip maker filed updated documents for its Nasdaq IPO, targeting a raise of up to $3.5 billion. The plan is to sell 28 million shares, with an estimated price range between $115 and $125 each, according to the updated prospectus submitted on Monday. If the offering hits the top of that range, the company could reach a market valuation of up to $26.6 billion based on outstanding shares.
That is no small deal.
To put things in perspective, back in February the company had already received a valuation of $23 billion in a Series H investment round that included AMD (Advanced Micro Devices) as one of the investors.
And with a multibillion-dollar contract with OpenAI locked in and revenue numbers growing strong, Cerebras hits the public market with a pitch quite different from most: being a real alternative to Nvidia GPUs in the AI ecosystem.
Let’s break down everything about this move 👇
What Cerebras is and why it matters so much
Cerebras Systems was founded in 2016 with a pretty clear mission: build AI chips that could do what traditional GPUs simply cannot do with the same efficiency. The company bet on a completely unconventional architecture, developing what it calls the WSE, or Wafer-Scale Engine, which is essentially the largest silicon chip ever produced at commercial scale in the world. To give you a sense of the engineering involved, the Cerebras chip occupies an entire silicon wafer, while Nvidia GPUs use only a fraction of that same wafer. That means far more processing cores, far more integrated memory, and in many scenarios, significantly superior performance for training large language models and other AI workloads.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
This technical edge is not just marketing. When you need to train a language model with billions of parameters, latency and data transfer between separate chips become serious bottlenecks. With the Cerebras WSE, a large part of that problem simply goes away, because everything is integrated into a single chip. This accelerates AI model training significantly and reduces the infrastructure complexity required to get the job done. That is exactly why major players in the field, including OpenAI itself, saw value in the company’s technology and signed meaningful partnerships and contracts with it over the past few years.
Beyond that, Cerebras did not just sit around building hardware. The company built an entire software ecosystem around its chips, making integration with popular machine learning frameworks like PyTorch straightforward. This is critical, because it does not matter how powerful the hardware is if developers cannot actually use it in a practical way day to day. With this combination of innovative hardware and solid software support, Cerebras positioned itself as a serious and technically capable alternative in a market that, until recently, was practically a Nvidia monopoly.
The IPO and what is at stake with this capital raise
Going public is never a simple decision, especially in a market as volatile as tech. But Cerebras seems to have timed this move pretty well. The Nasdaq IPO arrives at a moment when global interest in artificial intelligence infrastructure is at an all-time high, with governments, major corporations, and startups all competing for access to chips and computing capacity. By filing the documents for the initial public offering, Cerebras signals to the market that it is ready to scale its operations significantly and needs capital to do it — capital that this IPO raise can provide in a highly efficient way.
The offering numbers are impressive on their own. With 28 million shares being put up for sale in a price range between $115 and $125, the company could raise up to $3.5 billion from this offering alone. On top of that, the company reserved an option to sell an additional 4.2 million shares to underwriters after the IPO, which could generate another $525 million in proceeds at the top of the price range. This money is expected to go primarily toward expanding chip production capacity, investing in R&D for next-generation hardware, and strengthening the engineering and sales teams. It is the kind of capital that can make the difference between being a relevant player and becoming one of the pillars of global AI infrastructure.
Worth noting that this is not the first time Cerebras tried to go public. The company pursued an IPO in 2024 but ultimately withdrew the filing, at a time when its business model was transitioning from direct hardware sales to operating a cloud service built on its own chips. In April of this year, Cerebras filed for the IPO a second time, this time apparently with more confidence in its business fundamentals and investor appetite for companies tied to the AI ecosystem.
An important detail is that Andrew Feldman, co-founder and CEO of Cerebras, is not selling shares in this offering. After the IPO, he will retain 10.3 million shares, which could be worth up to $1.28 billion at the top of the price range, according to the documents filed with the SEC. That sends a strong message to the market: the founder believes in the long-term value of the company and is not looking for a quick exit.
The numbers backing the investor bet
One of the most important factors for any IPO is the company’s ability to demonstrate real revenue growth, and Cerebras delivers exactly that. The company’s fourth-quarter revenue grew approximately 76% year over year, reaching $510 million. Even more impressive, the company posted $87.9 million in net income during the period, something rare for high-growth tech companies that typically prioritize expansion over profitability in their early stages.
These numbers carry even more weight when compared to the broader tech IPO landscape over the past few years. Relatively few tech companies have gone public since central banks around the world raised interest rates in 2022 to combat inflation, making investors far less interested in companies that were not generating profit. But with the rise of generative AI products like OpenAI’s ChatGPT, appetite for companies tied to artificial intelligence came roaring back.
A good example is CoreWeave, a Cerebras competitor that operates by renting Nvidia GPUs as a cloud service. CoreWeave, despite being unprofitable, raised $1.5 billion in its IPO last year. If a company that still does not generate profit managed to pull in that kind of capital, the expectation is that Cerebras, with its growing revenue and positive bottom line, could attract even greater interest from institutional investors. 🚀
The multibillion-dollar contract with OpenAI
Perhaps the biggest card Cerebras is holding as it heads to market is the contract signed with OpenAI back in January. The company committed to providing up to 750 megawatts of AI computing power to OpenAI through 2028, in a deal valued at more than $20 billion. That is the kind of agreement that transforms a niche company into a first-tier strategic player in the industry.
Having OpenAI as a marquee customer is a technical and commercial validation that is hard to top. OpenAI is arguably the most demanding company in the world when it comes to AI infrastructure, running some of the largest language models on the planet. If Cerebras technology is good enough to meet OpenAI’s demands at scale, that sends a clear signal to the rest of the market that these chips actually work — in real-world production conditions and under intense pressure.
OpenAI has even revealed the first AI model running on Cerebras chips, reinforcing that this partnership goes well beyond a deal on paper. It is a deep technical integration involving software adaptation, model optimization, and engineering collaboration between the two companies.
Cerebras vs. Nvidia: the fight for AI chips
Saying that Nvidia dominates the AI chip market is almost an understatement. The California-based company controls an estimated 70% to 80% of the GPU market for AI data centers, and its H100 and H200 chips have become practically synonymous with high-performance AI infrastructure. Any company looking to compete in this space needs to bring something very different and very good to convince customers to switch or split their budgets. Cerebras did exactly that, betting on a radically different architecture that delivers specific advantages in certain types of workloads, especially in training large natural language models.
The big competitive advantage for Cerebras lies in its ability to process models with massive parameter counts more efficiently than a cluster of networked GPUs. When you use multiple GPUs together, a significant chunk of processing time gets eaten up by communication between them, shuttling data back and forth. With the WSE, that communication happens within the chip itself, which eliminates a large portion of that bottleneck. In benchmarks released by the company, the WSE-3, the latest version of the chip, showed highly competitive performance-per-watt results compared to Nvidia solutions, especially in inference tasks and large-scale language model training.
Of course, Nvidia is not standing still. The company continues launching new hardware generations at a rapid pace and building a software ecosystem — CUDA — that is extremely difficult for any competitor to replicate, precisely because it has been built over more than a decade and is now the foundation on which most AI frameworks were developed. But Cerebras does not need to beat Nvidia in every segment to be profitable and successful. It just needs to carve out a slice of the market that aligns well with its strengths, and with the explosive growth of the artificial intelligence sector, even a small slice of this market represents an extraordinarily large volume of business. That is where the IPO bet starts to make complete sense. 💡
The state of the tech IPO market
The market for initial public offerings in tech went through a significant dry spell over the past few years. Since central banks raised interest rates in 2022 to fight inflation, investor enthusiasm for tech companies that were not yet profitable dropped considerably. This more restrictive environment pushed many companies to delay their plans for going public, preferring to stay private until conditions improved.
Generative AI completely changed that dynamic. The runaway success of ChatGPT and other products built on large language models created a wave of enthusiasm that reignited investor interest in any company positioned to benefit from this trend. Suddenly, companies like CoreWeave, which rents Nvidia GPUs as a cloud service, were able to pull off billion-dollar IPOs even without showing a profit. Cerebras, which combines revenue growth with actual profitability, enters this landscape in an even more favorable position to attract capital.
The fact that Cerebras tried to go public in 2024 and then pulled back also speaks to a certain strategic maturity. Instead of forcing a listing under unfavorable conditions, the company waited until its business model was more solidified — including the transition to operating a cloud service built on its own chips — and until it had heavyweight contracts, like the OpenAI deal, to present to investors. That patience could be rewarded with a much warmer reception from the market on this second attempt.
What is next for Cerebras
With the capital raised from the IPO, the expectation is that Cerebras will significantly accelerate development of new products and expand its global footprint. The company already has established relationships with manufacturing partners, including TSMC, the world’s largest cutting-edge chip manufacturer, and is expected to keep investing in the next generation of the WSE to maintain its technological lead over competitors. Beyond that, the funds could be used to expand the company’s cloud solution portfolio, making access to Cerebras chips more affordable for companies that cannot purchase the hardware outright — a move that could unlock entirely new markets for the company.
AMD’s participation as an investor in the last round before the IPO is also an interesting signal. AMD and Nvidia are direct competitors in the GPU market, so AMD’s interest in Cerebras could mean different things: it could be a strategic bet on complementary technology, a play to gain access to relevant intellectual property, or simply an investment decision based on return potential. Regardless of the motivation, the fact that a major semiconductor company put money into Cerebras adds another layer of credibility to the company at a time when it is looking to win over public market investors.
The overall outlook for the AI chip sector points to accelerated growth over the coming years. Analysts project that the global AI chip market could surpass $300 billion by the end of this decade, driven by growing adoption of generative language models, computer vision systems, industrial automation, and a host of other applications that demand intense computing power. For Cerebras, being positioned as a solid alternative in this expanding market — with proprietary technology, validated customers, and now public market capital — is a combination that puts the company in a pretty compelling position for the next chapters of this story. 😎
