What Is Cerebras? Understanding the AI Chip Company
Cerebras (CBRS) is a U.S. technology company focused on AI chips and AI computing, headquartered in Sunnyvale, California. The company develops AI processors, AI supercomputers, and supporting software platforms for both AI model training and inference. While NVIDIA has built its AI computing ecosystem primarily around GPUs, Cerebras has taken a fundamentally different and more aggressive approach: instead of distributing AI workloads across large numbers of smaller chips, the company aims to turn an entire silicon wafer into a single massive processor. This wafer-scale computing architecture has become the defining technology behind Cerebras.
Cerebras was founded by Andrew Feldman, who also serves as the company's CEO. Feldman was previously a serial entrepreneur and co-founded SeaMicro with Gary Lauterbach, a server chip company focused on low-power, high-density server technology. SeaMicro was acquired by AMD in 2012. This experience gave Feldman deep exposure to data center computing, power consumption, and chip architecture, and led him to consider whether traditional computing architectures could keep up with the rapidly growing size of AI models. In 2015, Feldman and several key engineers from his SeaMicro days founded Cerebras and pursued what was then a highly unconventional idea: instead of continuously adding more GPUs, why not put significantly more computing power into a single chip?
As a result, Cerebras did not attempt to make incremental improvements to the traditional GPU architecture. Instead, it set out to build a massive processor designed specifically for AI from the ground up, eventually developing a product strategy centered on wafer-scale computing. In simple terms, NVIDIA's approach is to connect large numbers of GPUs to increase AI computing capacity, while Cerebras seeks to concentrate more computing resources on a single wafer and reduce the amount of data that needs to move between chips. This fundamentally different architecture is at the heart of Cerebras' strategy to challenge NVIDIA and other established AI chip companies.
Data Source: Cerebras S-1
What Are Cerebras' Core Competitive Advantages in AI Computing?
Cerebras' biggest competitive advantage is its fundamentally different AI computing architecture. As generative AI models have expanded from billions of parameters to trillions of parameters, the computing resources required for AI training have increased by approximately 40,000 times over the past five years. Traditional GPU-based systems typically distribute large models across hundreds or even thousands of GPUs and rely on high-speed networking to continuously exchange data. This can increase power consumption and latency while forcing AI developers to manage complex distributed computing systems. Cerebras takes a different approach by concentrating more computing, memory, and bandwidth within a single massive chip, reducing the need for data transfers between chips.
At the center of this architecture is the Cerebras Wafer-Scale Engine (WSE). The third-generation WSE-3, for example, is approximately 57 times larger than NVIDIA's H100 and features around 900,000 computing cores, 44GB of on-chip SRAM, and 21 PB/s of memory bandwidth, representing approximately 52 times, 880 times, and 7,000 times those of the H100, respectively. A much larger chip allows more data to remain directly on the processor, reducing the latency and energy consumption associated with accessing external memory. This is a key part of Cerebras' strategy to deliver faster AI training and inference compared with traditional GPU architectures.
Cerebras also aims to simplify distributed computing for AI training. Because a single WSE provides substantial computing capacity and on-chip memory, it can accommodate large AI models without requiring them to be distributed across a large number of processors in the same way as GPU clusters. When additional training capacity is required, more WSE systems can be added, with different chips processing different training data without requiring a complete redesign of model partitioning and inter-chip communication. Cerebras says that some customers using its systems have achieved more than 10x faster training completion times compared with leading 8-GPU systems of the same generation.
Data Source: Cerebras S-1
Cerebras Financial Analysis: Revenue Growth, Business Model, and Growth Outlook
In the second quarter of 2026, Cerebras continued to deliver rapid revenue growth. Core revenue reached $209 million, up 103% year over year. Core cloud and other services revenue reached $127.7 million, representing a substantial 287% year-over-year increase and becoming the company's most important growth engine. Core hardware revenue was $82.10 million, up 17% year over year. During the same period, core gross margin was 40.6%, approximately 9.4 percentage points higher than the same period last year. The company expects gross margin to temporarily bottom in the third quarter and improve in the fourth quarter as its own systems and data centers come online. Its long-term target is to increase core gross margin to above 60%.
Looking ahead, Cerebras is entering a period of rapidly expanding demand for AI computing. The company has described 2026 as a "foundational year" focused on preparing for large-scale deployments in the future. As of June 30, 2026, the company's remaining performance obligations (RPO) had reached $25.4 billion, while full-year 2026 core revenue is expected to reach $880 million to $890 million. In addition to existing major customers such as OpenAI, Cerebras is expanding its presence among hyperscale cloud service providers such as AWS, as well as in programming, security, and enterprise AI applications. The company expects to offer services through AWS Bedrock in the first quarter of 2027, further expanding its cloud distribution channels for AI computing.
To prepare for rapidly increasing demand for AI computing, Cerebras is also significantly expanding its data center footprint and hardware capacity. The company has secured more than 600 MW of data center capacity and is continuing to build its presence in markets including the United States, France, Finland, and Canada. On the manufacturing side, Cerebras is expanding production through partners such as Flex and Sanmina. On the technology front, the company plans to launch its fourth-generation CS-4 system and expects to introduce CS-5 in the second half of 2027, with the goal of continuously improving system performance over the next several years. Overall, Cerebras' growth strategy is gradually evolving beyond simply selling AI chips toward a model combining AI hardware, cloud computing services, and long-term orders from major customers. If AI inference demand continues to grow rapidly, the company could have significant room to expand revenue in the future.
Data Source: Cerebras
Cerebras Stock Performance: How Has CBRS Stock Performed?
Cerebras (CBRS) officially went public on the Nasdaq on May 14, 2026. The IPO was priced at $185 per share, with an initial offering of 30 million shares and approximately $5.55 billion in proceeds. Underwriters subsequently exercised the overallotment option in full, bringing total proceeds to approximately $6.38 billion. Market sentiment was extremely strong on the first trading day, with CBRS briefly gaining more than 100% from its IPO price before closing at $311.07, approximately 68% above the $185 offering price. The move reflected strong investor expectations for AI chips, AI inference demand, and Cerebras' potential to challenge NVIDIA's position in the market.
After the sharp first-day rally, however, Cerebras stock quickly entered a correction. Because the IPO valuation had already incorporated very high expectations for AI-driven growth, investors began reassessing whether the company's valuation could ultimately be supported by actual revenue and profitability. At the $185 IPO price, Cerebras had a fully diluted valuation of approximately $56.4 billion, compared with full-year 2025 revenue of approximately $510 million. This meant that the company was trading at a valuation multiple far above those of traditional semiconductor companies. As the initial IPO enthusiasm faded, early investors took profits, while the market also began focusing on the high capital requirements of AI infrastructure and Cerebras' profitability during its rapid expansion phase. CBRS subsequently fell to around $160 before fluctuating around $200.
Cerebras' second-quarter earnings report provided another important test for the stock. Although core revenue reached $210 million, up approximately 103% year over year, and core cloud revenue surged 287% to $127.7 million, while the company also raised its full-year core revenue and gross margin guidance, CBRS stock still fell 12.69% that day. The key issue was not weak operating performance, but investor concerns about the costs behind the company's rapid growth. Core gross margin declined from 46.5% in the first quarter to 40.6% in the second quarter, primarily due to higher costs associated with leasing third-party data center capacity. In addition, the company's second-quarter GAAP net loss widened to approximately $451 million. With the stock already carrying a high valuation, the market became more cautious about Cerebras' near-term profitability.
Data Source: Bitunix
What Factors Affect Cerebras Stock Price?
Cerebras stock is closely tied to AI infrastructure investment, AI model inference demand, and the company's ability to improve profitability. Because Cerebras uses a wafer-scale computing architecture that differs significantly from NVIDIA's GPU cluster approach, investors are not only watching orders and revenue growth but also evaluating whether the company's technology can establish a differentiated position in the AI inference market and whether its rapid growth can ultimately translate into sustainable cash flow.
Can Wafer-Scale Computing Establish an Advantage in AI Inference?
Cerebras' key differentiation is its Wafer-Scale Engine (WSE) architecture, which integrates large numbers of computing cores, memory, and high-speed interconnects onto a single wafer. This differs significantly from NVIDIA's approach of building computing clusters from large numbers of GPUs. The architecture has the potential to reduce data-transfer latency between chips and provide advantages in low-latency, high-speed AI inference workloads. However, Cerebras' complete systems are also relatively expensive. As a result, the market is not simply focused on raw computing speed. Investors are increasingly looking at whether Cerebras can deliver advantages in cost per dollar, performance per watt, and inference cost per token compared with GPU clusters. As NVIDIA continues to introduce new-generation products such as Blackwell and AI infrastructure gradually shifts from training toward inference, Cerebras' ability to maintain a relative advantage in inference efficiency and cost-effectiveness through continued technological development will directly influence its market share and valuation.
Capital Spending by OpenAI, AWS, and G42
Large AI companies and cloud service providers such as OpenAI, AWS, and G42 are expected to be important sources of Cerebras' future revenue growth. The market is not only watching whether existing customer orders can be successfully deployed, but also whether these companies will continue increasing their AI data center and computing capital expenditures and whether more AI model developers will adopt Cerebras chips for training or inference. As of the second quarter of 2026, Cerebras' remaining performance obligations had reached $25.4 billion, while its data center capacity had expanded to more than 600 MW, indicating rapidly increasing demand from major customers. If more AI model developers join Cerebras' customer base, the company could reduce its dependence on a small number of major customers while further demonstrating the commercial viability of its wafer-scale computing technology.
Profitability and the Timeline to Positive Cash Flow
Cerebras is still in a period of rapid expansion, so the market is not only asking how much revenue the company can generate, but also when it can become consistently profitable. The company needs to invest heavily in data centers, computing capacity, and chip development. During the first half of 2026, Cerebras still recorded approximately $465 million in GAAP net losses, while operating cash flow remained negative. As a result, investors will continue to monitor gross margin improvement, operating expense control, and the timeline for positive cash flow. If Cerebras can maintain rapid revenue growth while gradually expanding gross margins and eventually generate positive free cash flow, the market may be willing to assign the company a higher valuation.
How to Trade Cerebras (CBRS)
For investors who want exposure to Cerebras stock, the traditional approach is to purchase CBRS shares through a U.S. stock broker. Investors generally need to open a U.S. stock trading account, complete identity verification, and fund the account before purchasing Cerebras shares. However, U.S. stock trading can involve account-opening procedures, cross-border funding, and transaction fees. Some brokers may also have relatively high trading costs, while traditional stock investing generally focuses on long positions and offers less flexibility when responding to short-term market declines.
Another option is to trade Cerebras contracts through cryptocurrency trading platforms such as Bitunix. Compared with directly purchasing shares, contract trading offers more flexible ways to participate in price movements. Investors can take long positions when they expect CBRS to rise or short positions when they expect the stock to decline. Bitunix CBRS contracts offer up to 20x leverage, allowing users to gain exposure to a larger position with less capital and potentially improve capital efficiency. Before trading, users need to register for a Bitunix account and complete KYC verification. The platform uses risk management, asset management, and trading monitoring mechanisms to provide a secure and efficient trading environment. However, leverage amplifies both potential gains and losses, so investors should manage their position size according to their own risk tolerance.
