In a huge room in a one-story building in Santa Clara, Calif., six-and-a-half-foot-tall machines whirred behind white cabinets this week. The machines make up a new supercomputer, which just went live last month.
Launched Thursday by Silicon Valley startup Cerebras, the supercomputer is built with the company’s specialized chips and is designed to power artificial intelligence products. The chips are notable for their size—the size of a dinner plate, or 56 times the size of chips commonly used in artificial intelligence. Each Cerebras chip has the computing power of hundreds of traditional chips.
Cerebras says it has built a supercomputer for artificial intelligence company G42. The G42 said it plans to use supercomputers to create and power artificial intelligence products for the Middle East.
“What we’re showing here is that there’s an opportunity to build a very large, dedicated AI supercomputer,” said Cerebras CEO Andrew Feldman, adding that his startup wants to “show the world that this work can be done faster, it can be done with less energy, and it can be done at a lower cost.”
Driven by the global artificial intelligence boom, demand for computing power and artificial intelligence chips has soared this year. In recent months, tech giants such as Microsoft, Meta and Google, as well as countless start-ups, have launched AI-powered products after the AI-powered ChatGPT chatbot became popular for the eerily human-like prose it can generate.
But building artificial intelligence products often requires massive amounts of computing power and specialized chips, leading to a frenzy for more such technologies. In May, Nvidia, a leading maker of artificial intelligence system chips, said demand for its products, known as graphics processing units, or GPUs, was so strong that its quarterly sales would beat Wall Street expectations by more than 50%. That forecast sent Nvidia’s market capitalization soaring above $1 trillion.
Ronen Dar, founder of Tel Aviv-based start-up Run:AI, said that “for the first time we’re seeing a huge increase in demand for computers” thanks to artificial intelligence technology. Run:AI is a startup that helps companies develop artificial intelligence models. This has “created a huge demand for specialized chips” and companies are “rushing to secure access” to them, he added.
To get enough AI chips, some of the biggest tech companies — including Google, Amazon, Advanced Micro Devices and Intel — have developed their own alternatives. Startups such as Cerebras, Graphcore, Groq and SambaNova have also entered the race, aiming to break into a market dominated by Nvidia.
Chips will play such a pivotal role in artificial intelligence that they could alter the balance of power among tech companies and even nations. For example, the Biden administration has recently considered restricting the sale of artificial intelligence chips to China, and some U.S. officials have said that China’s artificial intelligence capabilities may enhance Beijing’s military and security agencies, thereby posing a national security threat to the United States.
AI supercomputers have been built before, including from Nvidia. But startups rarely create them.
Based in Sunnyvale, California, Cerebras was founded in 2016 by Mr. Feldman and four other engineers with the goal of building hardware that accelerates the development of artificial intelligence. The company has raised $740 million over the years, including investments from artificial intelligence lab OpenAI leader Sam Altman and venture capital firms like Benchmark. Cerebras is valued at $4.1 billion.
Because the chips typically used to power AI are small (often the size of a postage stamp), hundreds or even thousands of chips are needed to process complex AI models. In 2019, Cerebras unveiled what it claimed was the largest computer chip ever built, which Mr Feldman said could train artificial intelligence systems 100 to 1,000 times faster than existing hardware.
Abu Dhabi company G42 will start working with Cerebras in 2021. The company used the Cerebras system to train an Arabic version of ChatGPT in April.
In May, the G42 asked Cerebras to build a network of supercomputers in different parts of the world. Talal Al Kaissi, chief executive of G42 Cloud, a subsidiary of G42, said cutting-edge technology will allow his company to build chatbots and use artificial intelligence to analyze genomic and preventive care data.
But demand for GPUs is so high that it’s hard to get enough of them to build supercomputers. Mr Al Kaissi said Cerebras’ technology was both available and cost-effective. So Cerebras used its chips to build a supercomputer for the G42 in just 10 days, Mr. Feldman said.
“The time scale is dramatically reduced,” Mr Alcaci said.
Cerebras says it plans to build two more supercomputers for the G42 next year — one in Texas and one in North Carolina — followed by six more distributed around the world. The network is called Condor Galaxy.
Chris Manning, a computer scientist at Stanford University whose research focuses on artificial intelligence, said startups may find it difficult to compete with Nvidia, because the people who build AI models are used to using software that runs on Nvidia’s AI chips. .
Other start-ups have also tried to enter the AI chip market, but many have “effectively failed,” Dr. Manning said.
But Mr Feldman said he was hopeful. He said that many artificial intelligence companies do not want to cooperate only with Nvidia, and there is also a large global demand for other powerful chips such as Cerebras.
“We hope this drives AI forward,” he said.