Cerebras

Cerebras was founded in 2015 by Andrew Feldman, Gary Lauterbach, Michael James, Sean Lie and Jean-Philippe Fricker with the aim of bringing wafer-scale computing to market, at a time when it was unclear whether this was technically possible. The five had previously built the cloud server company SeaMicro, which they sold to AMD for $334 million; Feldman was SeaMicro’s chief executive from November 2007 to February 2012. The company was incorporated in Delaware in April 2016.⁠[1][2][3][3]

In the founders letter in its IPO prospectus, the founders wrote that they made two bets: that general-purpose processors would not be sufficient for AI and purpose-built silicon would be needed, and that a new computer architecture would have to be designed from first principles. Viewing AI as a problem bound by communication latency, and noting that on-chip communication is thousands of times faster than communication across chips, they decided to use an entire wafer for one chip, a technique known as wafer-scale integration.⁠[3][3][3][3][3]

In December 2016, Cerebras, then still in stealth, received an investment from Benchmark of about $25 million. As of the May 2026 prospectus, entities affiliated with Benchmark, Foundation Capital and Eclipse each held more than 5% of the company; OpenAI CEO Sam AltmanSam AltmanSam Altman is an entrepreneur and investor, and the co-founder and chief executive officer of OpenAI. He previously served as president of Y Combinator and founded the location-based social service Loopt. He also helped found Tools for Humanity and Merge Labs, has invested in energy and life sciences ventures, and has driven public discussion of AI governance, employment, and wealth distribution.Open the full entry, Sun Microsystems founder Andy Bechtolsheim and Greg Brockman are among the company’s individual investors.⁠[4][4][3][3][3][1][1][1]

Exhibits made public in Elon Musk’s lawsuit against OpenAI show that OpenAI considered acquiring Cerebras in or around 2017: co-founder Ilya Sutskever listed items such as “Negotiate merger terms with Cerebras” in a July 2017 email and wrote in September that any purchase would likely be done through Tesla, but the deal did not happen. Feldman later said the OpenAI and Cerebras teams had met frequently since 2017 to share research and early work.⁠[5][5][5][5][6]

After designing the giant chip and having it manufactured by TSMC, Cerebras was stuck on “packaging”: attaching the chip to a board, delivering power, managing heating and cooling, and moving data in and out. Feldman recalled that the chip was 58 times larger and used 40 times as much power as anything before it, with no off-the-shelf heat sinks or suppliers; the company was spending about $8 million a month and had burned through nearly $200 million on this single problem, even building its own machine to drive 40 screws simultaneously without cracking the wafer. In July 2019, a packaged chip ran successfully in a computer for the first time.⁠[2][2][2][2][2][2][2]

On August 19, 2019, Cerebras unveiled its first Wafer Scale Engine (WSE): a single chip measuring 46,225 square millimeters with more than 1.2 trillion transistors, 400,000 AI-optimized compute cores and 18 GB of on-chip SRAM, 56.7 times the area of the largest GPU at the time; the first WSE was built on TSMC’s 16-nanometer process. The company then introduced the WSE-based CS-1 system and delivered its first systems in 2020.⁠[7][7][7][8][1][3]

In April 2021, Cerebras unveiled the second-generation WSE-2, built on TSMC’s 7-nanometer process with 2.6 trillion transistors and 850,000 AI cores, to power the CS-2 system; it more than doubled the first generation’s transistor count, core count, memory and bandwidth.⁠[8][8][8]

In November 2021, Cerebras raised $250 million in a Series F round at a valuation of more than $4 billion, bringing total funding to $720 million. The round was led by Alpha Wave Ventures with participation from Abu Dhabi Growth Fund (ADG) and G42, joining existing investors including Altimeter Capital, Benchmark, Coatue and Eclipse Ventures; the company also opened offices in Tokyo and Toronto.⁠[9][9][9][9]

In November 2022, Cerebras unveiled Andromeda, an AI supercomputer built from 16 CS-2 systems with 13.5 million AI cores and more than 1 exaflop of AI compute, which achieved near-linear scaling on large language model training using data parallelism alone. The founders later wrote that during this period the products were extraordinary but the market was not ready, with only a few supercomputing and life-sciences customers seeing their potential.⁠[10][10][10][10][3]

On July 20, 2023, Cerebras and the Abu Dhabi technology holding group G42 announced Condor Galaxy, a planned network of nine interconnected supercomputers with a total of 36 exaFLOPs. The first, CG-1 in Santa Clara, California, linked 64 CS-2 systems for 4 exaFLOPs of training capacity and was offered as a cloud service, the first time Cerebras not only built but also managed and operated a dedicated AI supercomputer. G42 accounted for 83% of Cerebras’s revenue in 2023.⁠[11][11][11][11][12]

On March 13, 2024, Cerebras unveiled the third-generation WSE-3: built on TSMC’s 5-nanometer process with 4 trillion transistors, 900,000 AI cores, 44 GB of on-chip SRAM and 125 petaflops of peak AI performance, powering the CS-3 system and delivering twice the performance of the WSE-2 at the same power and price. The same day, Cerebras and G42 announced Condor Galaxy 3 in Dallas, Texas, built from 64 CS-3 systems with 8 exaFLOPs, bringing the Condor Galaxy network to 16 exaFLOPs.⁠[13][13][13][14][14][14]

In May 2024, Cerebras and G42 agreed that a G42 affiliate would buy 22,851,296 Series F preferred shares for $335 million ($14.66 per share), subject to regulatory approval; G42 also received an option to buy additional shares at a discount if it purchased $500 million to $5 billion of additional products and services. In the same month, G42 committed to prepay and place orders for about $1.43 billion of high-performance computing systems, installation and support by February 28, 2025. Between July and September 2024, the remaining Series F-1 preferred shares were bought by other investors for $85 million.⁠[3][12][3][12][3]

On July 31, 2024, Cerebras announced that it had confidentially submitted a draft IPO registration statement to the SEC; that month, Cerebras and G42 filed a joint voluntary notice with the Committee on Foreign Investment in the United States (CFIUS) regarding G42’s share purchase. On September 30, the company publicly filed its S-1, planning to list on Nasdaq under the ticker CBRS. The S-1 showed 2023 revenue of $78.7 million and a net loss of $127.2 million, and first-half 2024 revenue of $136.4 million, 87% of it from G42.⁠[15][3][16][16][17][12][12][12][12]

On August 27, 2024, Cerebras launched its Cerebras Inference service, citing speeds of 1,800 tokens per second on Llama 3.1 8B and 450 tokens per second on Llama 3.1 70B, with prices starting at 10 cents per million tokens. Inference then became the company’s main growth driver; in April 2025, MetaMetaMeta (Meta Platforms, Inc.) is a publicly listed American technology company, renamed from Facebook in 2021, that operates social communication, digital advertising, virtual reality, smart glasses and artificial intelligence businesses.Open the full entry adopted Cerebras for fast inference in its new Llama API.⁠[18][18][19]

In March 2025, Cerebras announced six new inference data centers aimed at expanding inference capacity twentyfold; the Oklahoma City and Montreal facilities would be wholly owned and operated by Cerebras, the rest jointly operated with G42, with 85% of capacity in the United States. The same month, the listing was delayed again as the national security review of G42’s investment dragged on.⁠[20][20][20][20][21][21]

Following engagement with CFIUS, Cerebras and G42 agreed in principle in the first quarter of 2025 to remove G42 from the preferred stock purchase agreement (with G42 instead buying non-voting preferred stock if it completed the purchase) and to remove the pricing and volume commitments from the May 2024 agreement. Based on these representations, CFIUS granted the company’s request to withdraw the joint notice on March 27, 2025. Because G42’s purchase was not completed by April 15, 2025, no new purchase agreement was signed; in the third quarter of 2025, the preferred stock agreement was restated to remove G42 and its option, and the May 2024 purchase agreement was terminated entirely.⁠[3][3][3][3][3][3]

On September 30, 2025, Cerebras announced a $1.1 billion Series G round at a post-money valuation of $8.1 billion, led by Fidelity Management & Research Company and Atreides Management with participation from Tiger Global, Valor Equity Partners and 1789 Capital1789 Capital1789 Capital is a venture capital and investment management firm headquartered in Palm Beach, Florida, United States, founded by Omeed Malik, Chris Buskirk, and Rebekah Mercer, with Donald Trump Jr. as a partner since November 2024. The firm has invested in several defense and technology companies that have received US government contracts, drawing inquiries from congressional Democrats, and the firm has denied any conflict of interest.Open the full entry and existing investors Altimeter, Alpha Wave Global and Benchmark. The company said the funds would support chip and system development and expand U.S. manufacturing and data center capacity.⁠[22][22][22][22]

On October 3, 2025, Cerebras asked the SEC under Rule 477 of the Securities Act of 1933 to withdraw the S-1 filed on September 30, 2024, saying it did not intend to conduct that offering at that time; the registration statement had not been declared effective and no securities had been sold. On December 22, 2025, the company confidentially submitted a new draft registration statement.⁠[23][23][23][23][17]

In 2025, Cerebras reported revenue of $510.0 million, up 76% year on year, including $358.4 million from hardware and $151.6 million from cloud and other services; GAAP net income was $237.8 million, compared with a net loss of $481.6 million in 2024 (excluding stock-based compensation and other items, its non-GAAP net loss in 2025 was $75.7 million). Customer concentration was high: Abu Dhabi’s Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) accounted for 62% of 2025 revenue and G42 for 24% (85% in 2024). As of December 31, 2025, the company had 708 employees, 426 of them in the United States and others in countries including Canada and India.⁠[3][3][3][3][3][3][3][3]

In December 2025, Cerebras signed a master relationship agreement (MRA) with OpenAI under which OpenAI committed to buy 750 megawatts of AI inference compute capacity and related services, to be deployed in tranches from 2026 through 2028, each with a term of three or four years that OpenAI can extend to a maximum of five; OpenAI also has the option to buy an additional 1.25 gigawatts by the end of 2030. The companies announced the multi-year deal publicly on January 14, 2026; Cerebras’s prospectus valued it at more than $20 billion (early reports put it at more than $10 billion). OpenAI’s Sachin Katti said Cerebras added a dedicated low-latency inference solution to its platform.⁠[3][3][3][3][6][3][24][6]

As part of the agreement, Cerebras issued OpenAI a warrant in December 2025 to buy up to 33,445,026 non-voting Class N common shares at $0.00001 per share, with vesting tied to compute purchases; in January 2026, OpenAI provided a working capital loan of about $1.0 billion, repayable in cash or through the delivery of compute and maturing no later than the end of 2032. Some customer agreements, including the MRA with OpenAI, contain exclusivity provisions that restrict Cerebras from providing products and services to certain named competitors of those customers. Sam Altman was already an investor in Cerebras.⁠[3][3][3][3][3][25][25][24]

On February 3, 2026, Cerebras closed a $1 billion Series H round at a post-money valuation of about $23 billion, led by Tiger Global with participation from Benchmark, Fidelity, Atreides, Alpha Wave Global, Altimeter, AMD, CoatueCoatueCoatue is a technology- and innovation-focused investment management firm founded by Philippe Laffont in 1999, operating in both public and private markets. Its current website lists Thomas Laffont as co-founder and chief investment officer of private investments; a Form ADV filed on July 29, 2026, disclosed regulatory assets under management of $92,705,722,582.Open the full entry and 1789 Capital, among others. On February 12, OpenAI’s GPT-5.3-Codex-Spark launched in research preview, generating more than 1,000 tokens per second on the Cerebras Wafer-Scale Engine, the first release of the two companies’ collaboration.⁠[26][26][27][27]

That month, G42 and Cerebras announced at the India AI Impact Summit in New Delhi that they would deploy an 8-exaflop supercomputer in India, with MBZUAI and India’s Centre for Development of Advanced Computing (C-DAC) also taking part.⁠[28][28][28]

On March 13, 2026, Amazon Web Services and Cerebras announced a collaboration to deploy a “disaggregated inference” solution on Amazon Bedrock in AWS data centers, with AWS Trainium servers handling prompt processing (prefill) and Cerebras CS-3 systems handling output generation (decode), connected by AWS’s Elastic Fabric Adapter networking. Cerebras said AWS had signed a binding term sheet to become the first hyperscale cloud provider to deploy Cerebras in its own data centers; the commercial agreement includes an initial multi-year lease of systems, exclusivity provisions and minimum capacity guarantees.⁠[29][29][3][25]

On April 17, 2026, Cerebras publicly filed a new S-1. On May 13, it priced 30 million Class A shares at $185 each, far above the initial range of $115 to $125 per share (later raised to $150 to $160). The shares began trading on the Nasdaq Global Select Market under the ticker CBRS on May 14; the offering closed on May 15, with the underwriters exercising their over-allotment option in full for a total of 34.5 million shares, gross proceeds of about $6.4 billion and net proceeds of about $6.2 billion.⁠[17][30][31][32][33][33][33][25]

On the first day of trading, the stock opened at roughly double the offering price and closed with a market value of about $66 billion. Feldman said that day that demand for inference chips was “not speculative.” After the listing, the company has three classes of common stock: Class A with one vote per share, Class B with 20 votes per share, and non-voting Class N; based on holdings as of March 31, 2026, Class B holders controlled about 99.2% of the voting power after the offering.⁠[31][31][34][34][3][3][3]

On June 23, 2026, Cerebras reported its first quarterly results as a public company: first-quarter GAAP revenue of $193.4 million and core revenue of $191.3 million, up 92% year on year; in April, the company had secured a revolving credit facility of up to $850 million to accelerate data center acquisitions. The next day the shares fell almost 20% after the company guided to a narrower full-year gross margin in its core business; Feldman said investors had misunderstood the guidance, explaining that the company needed to temporarily rent back some of its own systems from a large customer to make capacity available before its own data centers came online.⁠[35][35][36][36][36][36]

On August 12, 2026, Cerebras reported second-quarter GAAP revenue of $180.1 million, up 74% year on year, and core revenue of $209.9 million, up 103%; cloud and other services revenue was $126.0 million, up 281%. As of June 30, the company held $8.6 billion in cash and short-term investments, had $25.4 billion in remaining performance obligations, and had more than 600 megawatts of data center capacity live or under contract for delivery by the end of 2027; it said it planned to more than triple revenue in 2027 and raised its 2026 core revenue outlook to $880 million to $890 million.⁠[37][37][37][37][37][37][37]

On August 13, OpenAI introduced a new Ultrafast service tier in its API, initially for a select group of customers, with Cerebras running GPT-5.6 Sol at up to 750 output tokens per second, about 14 times faster than standard processing. On August 18, Cerebras unveiled the CS-4, the first product of its next-generation Nexus rack-scale platform, built from three new WSE-3T (Wafer Scale Engine 3 Turbo) chips, which the company said is up to twice as fast as the CS-3 and delivers up to 10 times more throughput per watt.⁠[38][38][39][40][40][40]

In late September 2026, a report by research firm SemiAnalysisSemiAnalysisSemiAnalysis is a semiconductor and AI research firm founded by Dylan Patel in 2020. It produces reporting, institutional data models, consulting and hardware testing, covering chip manufacturing, memory, datacenters, inference costs and AI economics.Open the full entry concluded that OpenAI would use Nvidia GPUs rather than Cerebras hardware for the Ultrafast mode of GPT-6.1 Sol, raising market concerns about Cerebras’s role in OpenAI’s inference infrastructure; at the same time, about 19.4 million pre-IPO shares became eligible for sale on September 30. Cerebras shares fell nearly 20% over a week, closing lower for the fifth straight session on October 2 at $166.43.⁠[41][41][41][41][41]

On the evening of October 2, 2026 (a Friday in the United States), OpenAI CEO Sam Altman posted on X: “There is some speculation about our partnership with Cerebras. Cerebras is a close partner, and we have a deep engagement pushing on the frontiers of speed.” On Monday, October 5, Cerebras shares rose about 9.7% to $182.63 in intraday trading, still below the $185 offering price.⁠[42][42][42][41][43][43]

Cerebras’s core product is the Wafer-Scale Engine: a single processor made from an entire silicon wafer, concentrating large numbers of compute cores and on-chip memory on one piece of silicon to reduce the latency of communication between chips. The company says its WSE is 58 times larger than Nvidia’s B200 chip and, for many workloads, delivers inference up to 15 times faster than leading GPU-based solutions. Because the architecture does not use HBM memory, CoWoS packaging or 3-nanometer fabrication, the company says it avoids these supply-constrained components; its wafers are made by TSMC and its systems by contract manufacturers including Flex, Sanmina and Rocket EMS.⁠[3][3][37][37][37]

The company sells on-premises hardware and also rents compute by the month, year or token through its own Cerebras cloud and partner clouds; the founders say Cerebras was among the first semiconductor companies to have a cloud business. Customers include OpenAI, AWS, G42, MBZUAI, Block, Figma, GSK, Lovable and Cognition; its main hardware competitors are Nvidia, AMD and Intel, and in cloud services it competes with AWS, Microsoft Azure, Google Cloud, Oracle and CoreWeave, among others.⁠[3][3][3][3][44][37][3][3]

As of the May 2026 prospectus, Andrew Feldman served as chief executive officer, president and director, Robert Komin as chief financial officer, co-founder Sean Lie as chief technology officer and Dhiraj Mallick as chief operating officer; the company is headquartered at 1237 E. Arques Avenue in Sunnyvale, California.⁠[3][3]

Cerebras: Official website: https://www.cerebras.ai/ (opens in a new window).⁠[3]

Cerebras Investor Relations: Official website: https://investors.cerebras.ai/ (opens in a new window).⁠[37]

X: Public account: https://x.com/Cerebras (opens in a new window)⁠[45]

LinkedInLinkedInLinkedIn is a social networking platform for professionals. It was founded in 2002 by Reid Hoffman and others, launched in 2003, and listed on the New York Stock Exchange in 2011; after Microsoft acquired it for approximately $26.2 billion in 2016, it became a wholly owned Microsoft subsidiary. Its revenue comes from membership subscriptions, advertising, and recruitment solutions.Open the full entry: Public profile: https://www.linkedin.com/company/cerebras-systems/ (opens in a new window)⁠[45]

YouTubeYouTubeYouTube is Google’s video-sharing and creator platform, founded in 2005 by Steve Chen, Chad Hurley, and Jawed Karim and acquired by Google in 2006. Built around video uploading, viewing, and creator earnings, the platform has developed Shorts, YouTube Music, Premium, YouTube TV, and products for children, and is headquartered in San Bruno, California, United States.Open the full entry: Public channel: https://www.youtube.com/@Cerebras (opens in a new window)⁠[45]