news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Leidi.com Lotte August 3

According to foreign media reports, Nvidia's competitor, artificial intelligence chip manufacturer Cerebras Systems has secretly applied for an initial public offering (IPO) in the United States, and will be listed as early as October 2024, preparing to challenge industry giant Nvidia.

Cerebras Systems is an AI chip manufacturer based in Silicon Valley, USA. It was founded in 2016 and was valued at US$2.4 billion in its Series E financing in 2019.

Existing investors in Cerebras Systems include Abu Dhabi Growth Fund and Coatue Management

In 2021, Cerebras Systems once again received US$250 million in Series F financing, led by Edge Capital. Other investors included Altimeter Capital, Benchmark Capital, Coatue Management, Eclipse Ventures, Moore Strategic Ventures and VY Capital.

After this round of financing, Cerebras Systems' total financing amount has reached US$750 million, and its post-investment valuation has exceeded US$4 billion. OpenAI CEO Sam Altman is also an important investor in Cerebras Systems.

In 2019, Cerebras Systems released the first generation of WSE (Wafer Scale Engine) chips. The WSE 3 chip developed by Cerebras is cut from a whole wafer, is larger than a plate, and needs to be held up by people with both hands.

Typically, chipmakers cut 12-inch silicon into thin slices. After cutting, the wafer will be further cut into hundreds of independent small chips, and then transistors will be integrated on these small chips. The WSE-1 is directly "grown" on a whole wafer and is the largest AI chip.

In March 2024, Cerebras Systems launched Wafer Scale Engine 3. The 5nm, 4 trillion transistor WSE-3 is designed for training the industry's largest AI models, powering the Cerebras CS-3 AI supercomputer, delivering 125 petaflops of peak AI performance through 900,000 AI-optimized computing cores.

Officials said that CS-3 has an ultra-large memory system of up to 1.2 PB, designed to train next-generation cutting-edge models that are 10 times larger than GPT-4 and Gemini. A 24 trillion parameter model can be stored in a single logical memory space without partitioning or reconstruction, which simplifies the training workflow and improves developer productivity.

Training a model with one trillion parameters on a CS-3 is just as easy as training a model with one billion parameters on a GPU.

Cerebras aims to grab market share from Nvidia, claiming its chips are not only more powerful than Nvidia's graphics processing units, but also more cost-effective.

Cerebras Systems' IPO could also be a major test of interest in artificial intelligence-related stocks.

Over the past year, Nvidia's market value has continued to rise, once exceeding $3 trillion, becoming the world's most valuable company. As of today's close, Nvidia's stock price was $10.727 billion, and its market value is still as high as $2.6387 trillion.

However, a growing number of investors are concerned that AI technology is being “overhyped.” Some investors say that many of the assumed uses of AI “will never be cost-effective, will never really work, will simply consume too much energy, or will prove to be unreliable.”

This has also exacerbated the volatility of Nvidia's stock, and more and more investors are turning to value stocks.

LeiDi was founded by media person Lei Jianping. Please indicate the source if reprinting.