Silicon Valley chip maker Cerebras Systems announced the completion of a US$1.1 billion financing, with the company's valuation reaching US$8.1 billion. Cerebras is unique in the AI chip market, with its developed chip size reaching the size of a meal, far exceeding the size of a traditional chip ticket, and is better than NVIDIA (NVIDIA) products in processing programming and natural language query, or becoming a strong competitor.
This financing was led by Fidelity Management & Research Company and Atreides Management, and new investors include Tiger Global, Valor Equity Partners and 1789 Capital, a partner of the eldest son of US President Trump. Cerebras Executive President Andrew Feldman said the company is preparing for its first public offering (IPO) in the hope of growing semiconductor demand opportunities brought by the AI craze. Cerebras Systems customer list includes Meta, Amazon Web Services and French AI Laboratory Mistral, which also serves several U.S. government departments and medical institutions. NVIDIA currently accounts for about 80% to 90% of the global AI accelerator market, with a market value of $4.4 trillion.
Investors have a strong background. Business version pictures continue to be expandedIn terms of investment, Atreides Management, founded by Gavin Baker, former investment portfolio manager of Fidelity Management & Research Company, jointly invested in the financing. Valor Equity Partners has invested in many Elon Musk-owned businesses, and 1789 Capital quickly expanded after Trump's second son joined, and its management assets grew from $150 million last year to more than $1 billion. Original investors Altimeter, Alphawave and Benchmark also participated in the round of financing. Citigroup and Barclays Capital serve as joint placement agents.
Andrew Feldman pointed out that NVIDIA is using its large capital to make strategic investments, investing $5 billion in Intel in the past month, committing $10 billion in OpenAI, and investing $500 million in the new entrepreneur Nscale in the cloud. Andrew Feldman believes that when corporate technical advantages weaken, they will start using asset debt tables to bundle customers. NVIDIA responded that the company has won the market with performance-based testing results and customer value, emphasizing its AI-based facilities to provide unparalleled performance, versatility and value combinations.
Complete review and clear listing obstacles. Continue research and development investment to infringementCerebras started its first public offering process a year ago, but was delayed due to the review of investments in the Abudabi AI Group G42 by the Council of Foreign Investment (CFIUS). G42 is still Cerebras' largest customer and key partner. Andrew Feldman revealed that the review process has been completed in March this year to clear listing obstacles. Since its establishment, the company has been continuously investing in research and development investment. According to data disclosed last year, Cerebras lost $67 million in the first half of 2024, with revenue of $136 million.
The advantages of crystalline design are obvious. The training time has been greatly shortened.How enterprises use Cerebras technical advantages has become a key consideration. The company's chips use a circular wafer design, making the entire wafer into a single processor and equipped with 400,000 processing cores. Compared with the traditional solution that requires the assembly of multiple small chips, this design eliminates the delay in inter-chip communication and improves computing efficiency. For enterprises that need to deal with large language models, the Cerebras system can shorten training time from weeks to seconds, significantly reducing computing costs and power consumption. Drug manufacturers GlaxoSmithKline and AstraZeneca have used the Cerebras system for drug development, and the training time has been shortened from 2 weeks to 2 days.
Market competition drama, lock high-speed reasoning and Zhongdong marketThe market competition pattern is changing. Although NVIDIA maintains its leading position with CUDA software generation system and graphics processing unit (GPU) performance, the demand for personalized AI chips continues to grow. Technology giants such as Meta, Microsoft and OpenAI have launched special chips to reduce the dependence on a single supplier. Cerebras focuses on excellent inference speed, and it is said that the speed can reach 18 times that of traditional GPU solutions when running the Llama model. The company announced this year that it will establish a partnership with Meta to provide inference services for the newly launched Llama API.
The Central East market became the focus of Cerebras expansion, and the company announced a supply agreement with the Saudian Arab National Petroleum Corporation last year. Andrew Feldman said the company has held some device export permits and is applying for more export permits from the Trump administration. The computing systems built for the G42 are currently located in the United States. The Chinese East transaction requires the U.S. government export permission, and the Trump administration's attitude will affect Cerebras' business development in the region.
Diversification becomes a trend in the market. The chip market continues to expandAI chip market size continues to expand, and NVIDIA predicts that global data center capital expenditure will reach $3 to $4 trillion in 2030. Although NVIDIA remains the leader in mature hardware and software ecosystems, high prices and shortages of supply are driving enterprises to find alternatives. AMD has gradually expanded its market share through the MI300 series chips, and Intel has launched Gaudi processors to target cost-sensitive enterprise customers. Cerebras has a different approach, focusing on large-scale model training and high-speed reasoning scenes to avoid full competition with NVIDIA.
Looking ahead, the AI chip market will show a trend of diversified development. The cost-effectiveness of personalized chips is enhanced with the standardization of AI work, and enterprises can design specialized processors for specific application scenarios. Companies such as Broadcom help customers develop personalized chips, which are better than general-purpose GPUs in terms of performance and cost. If Cerebras maintains technological advantages and successfully goes public, it will provide investors with options for diversified investment in AI-based facilities. The company's breakthrough in reasoning speed and cooperation with the cooperation with major AI experiment rooms is expected to establish differentiated competition advantages in the reasoning service market.
Nvidia challenger Cerebras raises $1.1bn ahead of IPO Extended reading: Cerebras Systems launches the largest chip size, with inference speeds exceeding NVIDIA Blackwell AI chip manufacturer Cerebras: Demand surges in corporate customers DeepSeek