Cerebras Market Cap: $26B, Revenue by Year and IPO (2026)

Cerebras Systems was worth about $26 billion at the market close on October 22, 2026.

That is the stock market's verdict five months after the company went public. Cerebras priced its Nasdaq IPO at $185 a share on May 13, 2026, a fully diluted valuation of $56.43 billion, Reuters reported. The stock closed its first day at $311.07, which CNBC calculated as a market value of about $95 billion.

By October 22 the shares had fallen to $108.94, a market cap of $25.88 billion based on S&P Global data. That figure uses shares outstanding. On the fully diluted share count used to price the IPO, the same share price works out to roughly $33 billion. Cerebras reports third-quarter results on November 12, 2026, against its own guidance of $214 million to $216 million in core revenue.

The revenue is real and growing. Cerebras reported $510 million of revenue in 2025, up 76%, after $290.3 million in 2024 and $78.7 million in 2023. For 2026 it guides to $880 million to $890 million of "core" revenue, a non-GAAP measure, and its finance chief says it plans to more than triple revenue in 2027.

The short version of how it grew: Cerebras built the largest chip ever made and sold it for years to labs and one Abu Dhabi customer group. Then it realized the chip's biggest advantage was how fast it could answer questions, not how fast it could train models. That turned it into an inference cloud and won it a 750-megawatt contract with OpenAI.

It is also a story about concentration. For most of its history, customers in the United Arab Emirates paid most of the bills. In October 2026 the stock fell further on reports that OpenAI had picked Nvidia for a new speed tier.

Here is how it happened.

Cerebras's growth in one table

Date	Milestone
2015	Andrew Feldman, Gary Lauterbach, Michael James, Sean Lie and Jean-Philippe Fricker found Cerebras; it is incorporated in April 2016
2016	Benchmark leads a $27M Series A
August 2019	Cerebras unveils the Wafer Scale Engine, a 1.2 trillion transistor chip
November 2019	A funding round values Cerebras at $2.4B, per Bloomberg
November 2021	Raises a $250M Series F at more than $4B; $720M raised to date
July 2023	Announces Condor Galaxy with G42, starting with a $100M+ supercomputer
2023	Revenue reaches $78.7M, and G42 accounts for 83%
August 2024	Launches Cerebras Inference
September 2024	Files for an IPO; first-half revenue is $136.4M
March 2025	Announces six new inference data centers
September 2025	Raises a $1.1B Series G at $8.1B
October 2025	Withdraws the IPO filing
January 2026	Announces a 750MW deal with OpenAI
February 2026	Raises a $1B Series H at about $23B
March 2026	Signs a deal to put Cerebras systems inside AWS
April 2026	Files publicly again with 2025 revenue of $510M
May 2026	Lists on Nasdaq as CBRS and closes day one at about $95B
August 2026	Reports Q2 core revenue of $209.9M and $25.4B in contracted future revenue
October 2026	Stock closes at $108.94 on October 22, a market cap of about $25.9B

1. Cerebras bet that one giant chip would beat thousands of small ones

Cerebras was started by people who had already built and sold a hardware company.

Andrew Feldman co-founded server maker SeaMicro and sold it to AMD for more than $300 million in 2012, Bloomberg reported. Cerebras's IPO prospectus shows co-founder Sean Lie followed the same path, from SeaMicro to AMD. Feldman, Lie, Gary Lauterbach, Michael James and Jean-Philippe Fricker started Cerebras in 2015.

Their idea went against 60 years of chip design. Instead of cutting a silicon wafer into hundreds of small chips, Cerebras would use the whole wafer as one processor. The company's own history says that at the time, it was unclear whether this was technically possible.

The main obstacle was defects. Every wafer has flaws, and a bigger chip is more likely to hit one. Cerebras designed the chip with spare building blocks so that flaws could be detected, shut down and routed around, a technique borrowed from memory manufacturing.

In August 2019, Cerebras unveiled the result. The Wafer Scale Engine had 1.2 trillion transistors and 400,000 cores across 46,225 square millimeters of silicon, 56.7 times larger than the biggest GPU. It was manufactured by TSMC.

By then Cerebras had raised more than $100 million from investors including Benchmark, Andy Bechtolsheim and Sam Altman. Altman's early check would matter again seven years later.

2. Cerebras sold whole supercomputers, not chips

A chip that size could not be dropped into a normal server.

So Cerebras decided not to sell chips at all. It sold complete systems, with the power delivery, water cooling and networking needed to run a wafer-sized processor. The prospectus says it shipped its first CS-1 in 2020, its first CS-2 in 2021 and introduced the CS-3 in 2024.

The early buyers were research labs and large enterprises. Researchers at Argonne National Laboratory won the Gordon Bell Special Prize in 2022 for COVID-19 research run on a Cerebras CS-2 cluster. Drug maker GSK used Cerebras systems to train biological language models.

The business was still small. Cerebras says it began generating revenue in 2019 and had limited revenue before 2023. In 2022 it brought in $24.6 million and lost $177.7 million.

Investors kept paying up anyway. A November 2019 round valued Cerebras at $2.4 billion, and in November 2021 it raised a $250 million Series F at more than $4 billion. That round brought total funding to $720 million. Two of the new investors were Abu Dhabi Growth Fund and an Abu Dhabi technology group called G42.

3. Cerebras found one enormous customer in Abu Dhabi

G42 turned out to be far more important as a customer than as an investor.

In July 2023, Cerebras and G42 announced Condor Galaxy, a planned network of nine AI supercomputers. The first, CG-1, linked 64 CS-2 systems in Santa Clara. Feldman told EE Times that CG-1 alone was worth more than $100 million, and the full network was expected to cost more than $900 million.

There was a twist in the structure. G42 owned the machines, but Cerebras built, managed and operated them and offered them as a cloud service. It was the first time Cerebras had run a supercomputer for a customer, and it was early practice for the cloud business it would build next.

The revenue followed quickly:

2023 revenue rose 220% to $78.7 million, and G42 accounted for 83% of it, Reuters reported.
In the first half of 2024, revenue jumped to $136.4 million from $8.7 million a year earlier.
G42 was 87% of that, and it had signed agreements to buy $1.43 billion of systems and services, CNBC reported.

The dependence did not end in 2024. G42 was 85% of 2024 revenue. In 2025, Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) was 62% and G42 was 24%. Cerebras's filings treat the two Abu Dhabi organizations as related parties, so together they made up 86% of 2025 revenue.

That money paid for Cerebras's next move. The prospectus says the company financed its operations largely through preferred stock and customer prepayments, including prepayments from G42 and MBZUAI.

4. Cerebras turned its training machine into the fastest answer machine

For its first eight years, Cerebras pitched faster model training. In August 2024 it found a bigger opening.

Running a trained model, called inference, is limited mostly by how quickly the model's weights can be moved from memory to the processor. A GPU keeps most of those weights in separate memory chips. A Cerebras wafer has enough on-chip memory to keep the weights on the chip, so tokens stream out much faster.

On August 27, 2024, Cerebras launched Cerebras Inference. It served Meta's Llama 3.1 8B at 1,800 tokens per second and Llama 3.1 70B at 450 tokens per second, which it said was 20 times faster than GPU-based hyperscale clouds. Pricing started at 10 cents per million tokens.

Cerebras did not ask buyers to take its word for the speed. Independent benchmarker Artificial Analysis said in the launch release that Cerebras was achieving "the fastest speeds we have ever benchmarked" on those models.

It also made switching easy. The API used the OpenAI Chat Completions format, so developers could try it by swapping an API key, and the launch included 1 million free tokens a day.

The company now argues that speed is sticky. Its prospectus says that once customers adopt fast inference, user expectations rise and it becomes difficult to return to slower inference.

5. Cerebras put its speed inside products people already used

A fast API is only useful if developers find it. Cerebras spent 2025 getting into other companies' products.

In February 2025, Cerebras began powering Mistral's Le Chat "Flash Answers" at more than 1,100 tokens per second.
The same month, Perplexity launched its Sonar search model on Cerebras at 1,200 tokens per second.
In April 2025, Meta's new Llama API offered early access to Llama 4 models powered by Cerebras and Groq.

Cerebras also became available through other platforms. Its prospectus lists AWS Marketplace, Microsoft Marketplace, IBM watsonx, Vercel AI Gateway, OpenRouter and Hugging Face as places developers can reach its inference service.

Serving that demand meant building a cloud. In March 2025, Cerebras announced six new inference data centers that would expand its capacity 20 times, to more than 40 million Llama 70B tokens per second. It said 85% of the capacity would be in the United States.

The cloud business started to show up in the numbers. Revenue from cloud and other services grew from $78.3 million in 2024 to $151.6 million in 2025. By mid-2026, Cerebras had added cloud contracts with AI coding companies Cognition and Lovable and listed Block, Figma, AlphaSense, GSK and CrowdStrike as customers.

6. Cerebras restructured the G42 deal and kept raising privately

Cerebras first filed to go public on September 30, 2024. The filing made the G42 problem public.

G42 had agreed to buy a $335 million stake in Cerebras, subject to regulatory approval. It also held an option to buy more shares at a discount if it spent between $500 million and $5 billion more on Cerebras products. Because G42 is a foreign investor, the deal needed review by the Committee on Foreign Investment in the United States (CFIUS).

The review stalled the IPO. CNBC summed up the filing with an investor's remark that it had "too much hair."

Cerebras fixed it by shrinking G42's role. In early 2025, the two companies agreed to remove G42 from the stock purchase, offer it non-voting shares instead, and drop product pricing and volume commitments. CFIUS allowed the review to be withdrawn on March 27, 2025. G42 never completed the purchase, and its option was terminated.

Meanwhile, Cerebras raised private money. On September 30, 2025, it closed a $1.1 billion Series G at an $8.1 billion valuation, led by Fidelity and Atreides Management with Tiger Global, Valor Equity Partners and 1789 Capital. "When you have enormous opportunities," Feldman told CNBC, it is good practice "not to let them fall by the wayside for lack of capital."

Three days later, Cerebras withdrew its IPO registration. A spokesperson told CNBC the prospectus was out of date and the company still hoped to go public as soon as possible.

The G42 unwind also shaped Cerebras's 2025 accounting. When G42's purchase lapsed, a liability tied to it was cancelled, producing a one-time gain of $363.3 million. That gain is why Cerebras reported 2025 net income of $237.8 million. Excluding it and other items, the company had a non-GAAP net loss of $75.7 million.

7. Cerebras made OpenAI's model fast, and OpenAI signed up

Cerebras's biggest deal grew out of an engineering favor.

In 2025, OpenAI worked with Cerebras to make its open-weight gpt-oss models run well on Cerebras hardware. Feldman told CNBC that collaboration led to technical conversations and a term sheet just before Thanksgiving. OpenAI CEO Sam Altman had been one of Cerebras's early backers.

The contract was signed on December 24, 2025. Under the master relationship agreement filed with the SEC, OpenAI committed to buy 250 megawatts of Cerebras capacity by the end of 2026, 500 megawatts by the end of 2027 and 750 megawatts by the end of 2028.

OpenAI announced the partnership on January 14, 2026. "Cerebras adds a dedicated low-latency inference solution to our platform," OpenAI's Sachin Katti wrote.

The value was reported differently over time. At announcement, people familiar with the deal said it was worth more than $10 billion, Reuters reported. Cerebras's later SEC filings describe it as a multi-year deal valued at more than $20 billion.

The deal went beyond buying compute:

OpenAI lent Cerebras a $1 billion working capital loan at 6% interest to help build the capacity.
OpenAI received a warrant for up to 33.4 million Cerebras shares at a nominal price, subject to vesting conditions.
The contract includes exclusivity terms that stop Cerebras from selling certain products to named OpenAI competitors.

Delivery started on January 23, 2026. On February 12, OpenAI released Codex-Spark, a real-time coding model running on Cerebras. In August, Cerebras began powering an Ultrafast mode for GPT-5.6 Sol at up to 750 tokens per second, in limited preview.

The contract transformed Cerebras's backlog. At the end of 2025 it had $24.6 billion in remaining performance obligations, revenue under contract but not yet earned, with a significant amount from OpenAI. That is nearly 50 times its 2025 revenue.

8. Cerebras turned cloud giants and neoclouds into distributors

The OpenAI deal changed how investors priced Cerebras. In February 2026, it raised a $1 billion Series H at about $23 billion, nearly triple the valuation of four months earlier. Tiger Global led, AMD joined, and Benchmark put in at least $225 million through two funds created for the deal, TechCrunch reported.

A month later, Amazon became a partner. AWS agreed to deploy Cerebras CS-3 systems in its own data centers and sell access through Amazon Bedrock. The design splits inference in two. Amazon's Trainium chips handle the first step of reading a prompt, and Cerebras handles generating the answer, which is where its memory bandwidth matters most.

Cerebras pays for some of these partnerships in equity. AWS gets a warrant to buy up to 2.7 million shares at $100, vesting only if it buys well beyond the initial lease. Cerebras now expects the AWS version of this split system on Bedrock in the first quarter of 2027, and a similar setup with AMD in production in the fourth quarter of 2026.

The same idea is reaching smaller clouds. In September 2026:

Gimlet Labs agreed to take Cerebras hardware able to draw roughly 100 megawatts, Reuters reported.
General Compute, a cloud built for non-Nvidia chips, signed a multi-year deal to offer Cerebras inference for coding agents.
Cerebras announced a 165-megawatt data center in Mikkeli, Finland, backed by seven-year capacity contracts.

The hardware kept moving too. In August 2026, Cerebras launched the CS-4, a rack built from three new wafer chips that it says is up to twice as fast as the CS-3. It now has more than 600 megawatts of data center capacity live or under contract for delivery by the end of 2027.

9. Cerebras went public at the height of the AI-chip trade

Cerebras confidentially submitted a new IPO filing on December 22, 2025. It filed publicly on April 17, 2026, Bloomberg reported. This time the story was OpenAI, AWS and $510 million of revenue, not a single UAE customer.

Demand was intense. Cerebras sold 30 million shares at $185, above its indicated range, after orders exceeded supply by more than 20 times, Reuters reported. The $5.55 billion raise was the largest IPO for a U.S. tech company since Uber in 2019, according to CNBC.

The stock opened at $350 on May 14, touched $386 and closed at $311.07. With the underwriters' extra shares, Cerebras says it raised $6.4 billion in gross proceeds, and it added an $850 million credit line.

Control stayed with insiders. Class B shares carry 20 votes each, and after the IPO they held about 99.2% of the voting power. Feldman and Lie both became billionaires through the IPO, CNBC reported.

How fast did Cerebras grow?

Cerebras's revenue figures below come from its SEC filings. Years run January to December.

Period	Revenue	Change
2022	$24.6M
2023	$78.7M	+220%
2024	$290.3M	about 3.7 times
2025	$510.0M	+76%
Q1 2026	$193.4M	+94%
Q2 2026	$180.1M (core: $209.9M)	+74% (core: +103%)
2026 outlook	$880M to $890M core

Two things make the 2026 numbers harder to read.

First, Cerebras now reports two revenue figures. GAAP revenue is reduced by the cost of the warrants it gave OpenAI and AWS, and it includes data center costs passed through to a customer. "Core" revenue strips both out. After the Q2 report, LSEG said it would use the $210 million core figure to measure growth from then on.

Second, the mix is shifting from hardware to cloud. In Q2 2026, cloud and other services revenue rose 281% to $126 million, while hardware revenue fell 23%, mostly because of $28 million of warrant amortization, according to the 10-Q.

Profitability is still ahead. Core gross margin was 41% in Q2 2026. Cerebras's GAAP net loss for the quarter was $450.5 million, mostly tied to stock compensation that vested at the IPO, CNBC reported.

The forward-looking number is the backlog. Remaining performance obligations reached $25.4 billion at the end of June 2026. Cerebras expected to recognize about 15% of its year-end 2025 backlog by the end of 2027 and 43% in the two years after that.

Cerebras's growth loop

Cerebras's growth works roughly like this:

Wafer-scale chips keep model weights on the chip, which makes answers faster.
Speed is shown in public benchmarks and inside well-known products.
Developers and AI companies build features that depend on that speed, such as real-time coding.
Large buyers sign multi-year capacity contracts, often with prepayments or loans.
Those contracts finance more data centers and manufacturing, which brings in the next large buyer.

The loop's weak point is step 4. Each big contract funds the next build-out, but each one also concentrates more revenue in fewer customers.

What Cerebras did differently

It changed what it was selling

Cerebras did not abandon its chip. It changed the pitch from faster training to faster answers, at the moment reasoning models and coding agents made speed valuable.

It let customers fund the build-out

G42 and MBZUAI prepayments, and later OpenAI's $1 billion loan, paid for capacity before the revenue arrived. Cerebras says this access to capital reduced its working capital needs.

It cooperated with bigger rivals

Rather than replace GPUs everywhere, Cerebras positioned itself as the fast half of a combined system with AWS Trainium, AMD and GPU clouds.

What could slow Cerebras down?

OpenAI. OpenAI is now central to Cerebras's backlog, and the market is watching every signal. On September 30, 2026, research firm SemiAnalysis said OpenAI would run the Ultrafast mode of GPT-6.1 Sol on Nvidia GPUs instead of Cerebras, CNBC reported. Altman responded that "Cerebras is a close partner," but the stock fell nearly 20% that week.

Concentration in Abu Dhabi. MBZUAI was still 34% of revenue in Q2 2026 and G42 was 9%. Exports to both require U.S. licenses with strict security conditions, and the prospectus flags the war in the Middle East as a risk.

Delivery. If Cerebras misses capacity deadlines or service levels, OpenAI can terminate part or all of the agreement. Cerebras also depends entirely on TSMC for wafers, with no long-term supply commitment.

Margins. Cerebras's 2026 gross margin outlook is far below Nvidia's, whose margins are in the mid-70% range, Reuters noted. Building and leasing data centers is a lower-margin business than selling chips.

Competition from customers. On October 12, 2026, Microsoft said it would make its Maia chips the primary inference system for Microsoft products and services from 2027. Every large cloud building its own inference chip narrows Cerebras's market.

Share supply. Insider shares are released from the IPO lock-up in stages rather than all at once. Feldman and Lie sold more than $240 million of stock between August 20 and September 25 under trading plans, CNBC reported. On October 19, TD Cowen cut its price target from $275 to $175 and the stock hit a new low of $120.12 during the day. It closed at $108.94 on October 22.

Borrowing Cerebras's benchmark-first launch

Cerebras launched its inference service with an independent benchmark quote ready, so buyers and reporters could repeat a third party's number instead of the company's. Okara's Writer Agent drafts strategy-led articles that make a proof-backed case and can publish them. The GEO Agent then tracks whether ChatGPT, Google AI Overviews and Google AI Mode mention you, for up to 15 prompts.

Frequently asked questions

What is Cerebras's valuation?

Cerebras is a public company, so its valuation is its market cap. The stock closed at $108.94 on October 22, 2026, a market cap of about $25.9 billion based on shares outstanding. It priced its IPO in May 2026 at a $56.43 billion fully diluted valuation and closed its first day at about $95 billion. Its last private valuation was about $23 billion in February 2026.

What is Cerebras's revenue?

Cerebras reported $510 million of revenue in 2025, up 76% from $290.3 million in 2024, according to its SEC filings. In the first half of 2026 it reported $373.5 million of GAAP revenue. It expects $880 million to $890 million of core revenue for 2026 and says it plans to more than triple revenue in 2027.

How much funding has Cerebras raised?

Before its IPO, Cerebras had raised about $2 billion in private funding by September 2025, TechCrunch reported, then added a $1 billion Series H in February 2026. OpenAI also lent it $1 billion. The May 2026 IPO raised $6.4 billion in gross proceeds including the underwriters' extra shares, and Cerebras added an $850 million credit facility.

Who founded Cerebras?

Cerebras was founded in 2015 by Andrew Feldman, Gary Lauterbach, Michael James, Sean Lie and Jean-Philippe Fricker, and incorporated in April 2016. Feldman is CEO and Lie is CTO. Feldman and Lie previously worked together at SeaMicro, which AMD bought in 2012.

Who are Cerebras's biggest customers?

Historically, Abu Dhabi's G42 and MBZUAI. G42 was 85% of 2024 revenue, and in 2025 MBZUAI was 62% and G42 24%. OpenAI signed a 750-megawatt contract in December 2025 that accounts for a significant amount of Cerebras's $25.4 billion backlog. AWS, Cognition, Lovable, Mistral, Perplexity and GSK are also customers or partners.

Why did Cerebras withdraw its first IPO?

Cerebras filed in September 2024, but a national security review of G42's planned investment delayed the listing. After restructuring the G42 deal and raising $1.1 billion privately, Cerebras withdrew the filing in October 2025, saying the prospectus was out of date. It refiled in April 2026 and listed in May.

Why has Cerebras stock fallen?

The shares fell about 65% from their first-day close by October 22, 2026. CNBC and other outlets tied the decline to a report that OpenAI chose Nvidia for a new speed tier, staged lock-up expirations, insider sales, and Microsoft's plan to use its own Maia chips for inference.