Start-ups seek to challenge Nvidia’s dominance over AI chip market

Start-ups are rushing to take advantage of the growing demand for specialized chips that power artificial intelligence, as the shortage of Nvidia’s latest products presents a once-in-a-generation opportunity. opportunity for new challengers to the dominance of the world’s most valuable semiconductor company.

A big upgrade to Nvidia’s sales forecast, driven by AI, pushed the market capitalization above $1tn in May but demand is expected to outstrip supply for the latest chips until next year.

Among the companies developing alternatives include SambaNova, Graphcore and Tenstorrent which have together raised more than $3bn in the past few years, according to figures compiled by Dealroom.co, which tracks private tech deals.

However, few have made significant inroads against Nvidia, whose A100 and H100 chips have become the go-to for companies like OpenAI and Inflection AI that need to process large volumes of data to run their services. in AI.

Cerebras, a Silicon Valley-based AI chip start-up that has raised $730mn since it was founded in 2016, this week announced that it will build and operate a network of supercomputers for Abu Dhabi-based tech group G42.

The deal could be worth “in excess of $100mn” if certain milestones are met in the coming months, according to Cerebras chief executive Andrew Feldman.

“AI today has an insatiable need for computation,” he said. “If you are David fighting Goliath you will look for cracks . . . [Nvidia’s] The inability to respond to demand is just one crack.”

The deal with G42, a private company that works in several sectors such as healthcare, energy and cloud computing, is one of the largest contracts of its kind for a potential rival to Nvidia.

G42 plans to use some of the new computing resources itself, while also selling any “excess capacity” to other customers through its cloud-computing arm with Cerebras.

“People can’t get the hardware they want, or it’s too expensive,” said Talal Alkaissi, chief executive of G42 Cloud. “The market is hungry for alternatives.”

Over the years, start-ups have variously claimed to outperform certain Nvidia products for particular types of workloads, including training large language models that power chatbots such as of ChatGPT and other “generative AI” systems capable of producing human-like text and realistic imagery.

But AI researchers, and start-ups that have turned their research into commercial products, still prefer Nvidia’s technology, according to traders, investors and sector analysts.

“None of these start-ups are making any significant amount of revenue,” said Jakub Zavrel, whose company Zeta Alpha tracks references to specific processors in AI research papers for tech investor Air Street Capital’s State of AI report.

While Cerebras has seen an uptick in research citations this year, surpassing Graphcore, they count in the dozens versus thousands of researchers citing Nvidia chips, Zavrel said. He predicts that the latest chips from AMD are more likely to take share from Nvidia than any of its private rivals. Intel is also preparing its own attack on Nvidia after acquiring another AI accelerator startup, Israel-based Habana Labs, for $2bn in 2019.

At the same time, many of the cloud computing providers that are buying chips to provide services to the new wave of AI companies and their business customers are also developing their own semiconductors.

Amazon Web Services launched Trainium, its custom chip for machine learning, in 2020, while Google Cloud offered TPUs, or Tensor Processing Units, to customers for five years.

Microsoft, which ended a relationship with Graphcore in 2020 after just one year, is also developing its own custom silicon for AI, further pressing the opportunity for start-ups hoping to get into market through cloud providers.

To win the G42 contract, Cerebras had to go beyond building some of the world’s most powerful processors — already an engineering feat few venture capital investors are willing to finance — by building and -operate the entire infrastructure needed to host them as well. Alkaissi calls it “white glove service” from Cerebras.

Some AI investors argue that chip startups need to go further to match Nvidia’s offering.

“It’s not just a matter of designing the best chips, making the chips and bringing them to market in a way that people want them,” said David Katz, a partner at Radical Ventures, an investor in AI-oriented technology. “Nvidia has invested for a long time in an ecosystem that lives around chips . . . that wins the hearts and minds of engineers working at the bare metal level. That includes software and support, especially is Cuda’s toolkit for programming its chips.

Faced with such a daunting set of tasks, some start-ups have moved away from a head-on competition with Nvidia.

Celestial AI, a Silicon Valley-based startup that raised $100 million in June, is once again focusing on “complementing” rather than competing with Nvidia, according to its chief executive Dave Lazovsky, by developing optical technology for connecting high-performance AI processors. memory required to feed them data.

“Most of these AI startups that are trying to compete with Nvidia don’t have a chance because they’re fighting the wrong battle,” he said. “The bottom line is memory requirements will continue to grow about 100 times faster than compute requirements.”

Fabrizio Del Maffeo, chief executive of Netherlands-based Axelera AI, develops AI chips designed for cars, medical devices and security cameras, rather than the cloud and data centers where the most powerful Nvidia chips are in high demand.

“I’ve always said since day one that it’s crazy to fight a trillion-dollar company with unlimited resources,” Del Maffeo said.

Additional reporting by Richard Waters