Cerebra’s $10 billion deal with OpenAI positions the startup and its wafer-scale engines as a challenger to Nvidia in the AI chip market, while helping OpenAI try to accelerate the performance of its larger AI models.
The multiyear deal revealed on January 14 requires Cerebra to deliver 750 MW of wafer-scale systems to OpenAI by the end of this year. Will use OpenAI Wafer-scale engine To provide near-real-time responses for tasks such as coding, inference, image generation, and complex reasoning. The Cerebras version of the AI chip is larger and, the vendor says, faster than the Nvidia GPU.
compromises cerebrusWhich has struggled to expand its customer base beyond G42, the Abu Dhabi-based AI and tech holding company since it was founded in 2015, is gaining a foothold in the AI chip market as one of several chipmakers trying to rival Nvidia. It also addresses a key challenge for enterprises that have found that large-scale generative AI models are often too slow or expensive to use in real-time.
a platform for cerebrus
The deal gives Cerebra the opportunity to prove the capabilities of its wafer-scale engine.
While some have previously viewed the AI vendor as a somewhat experimental company focused on science applications, this partnership is “the final stamp of legitimacy,” said Mike Lyon, an analyst at Omdia, a division of Informa TechTarget.
“This turns them overnight from a niche option to a serious contender that every other AI lab now has to pay attention to,” Lyons said.
Potential benefits for enterprises and OpenAI
Competition from chip makers including Cerebras, as well as longtime semiconductor companies Broadcom and AMD, could benefit enterprises by potentially driving down AI service prices over time.
“Overall, an enterprise customer has more choices when it comes to how they get their AI stuff that they want,” said David Nicholson, an analyst at Futurum Group.
The deal also tackles a paramount market issue: AI model speed.
“The industry is struggling with a tough tradeoff at the moment, where smart models are becoming very heavy and slow to run,” Lyons said. “It appears that this partnership is trying to solve that specific friction point. By focusing on estimate speedIt seems they are trying to ensure that future AI agents can handle complex tasks without the interruptions that currently frustrate users.
For OpenAI, which has seen competition from Google and others intensify amid questions about OpenAI’s long-term financial viability, the deal with Cerebrus is a win, especially after Google’s head agreement with apple Earlier this week Google partnered with the Gemini Foundation model to power Apple’s AI initiative. The deal shows that OpenAI is expanding its infrastructure for a future when AI works for users, not just interacts with them, Lyons said.
“There’s a deeper story here about the shift from chatbots to real agents,” he said. “When you’re just chatting, a delay of a few seconds is fine. But when you have an AI agent that’s trying to solve a complex problem that requires twenty steps of ‘thinking’ in the background – that requires a level of speed that standard hardware struggles to deliver efficiently.”
Chipping away at nvidia
If Cerebras proves that its hardware can accomplish this, it could chip away at Nvidia’s market share.
“In my opinion, Cerebra is the biggest threat to Nvidia from a hardware company perspective,” Nicholson said. “This negatively impacts Nvidia in the long term, if OpenAI proves that a data center built on Cerberus technology is superior.”
Nicholson said some observers see Nvidia at a relative disadvantage because it lacks a wafer-scale product. Nvidia’s approach involves producing many different chips by cutting a wafer and assembling it, which can result in mechanical and electrical errors due to discarded, defective chips. In contrast, Cerebras keeps the entire wafer intact as a single, large chip, connecting only the functioning cells. This approach can reduce system complexity and potential error points.
“On the surface, it’s so obvious that what Cerebras does is technically superior,” Nicholson said. “But Nvidia is able to get away with its inferior technology because it has the full stack and it has the industry connections and momentum.”
However, Nicholson said, Cerebras faces the challenge of delivering at scale almost immediately.
Furthermore, integration is a challenge. Enterprises interested in Cerebra may find its technology too complex to integrate with their systems and feel they need more talent to make it work better. This may make it less attractive to non-hyperscalers.
“It’s much easier to go with Nvidia’s off-the-shelf solution,” Nicholson said.
Meanwhile, Nvidia last month signed a $20 billion licensing deal with Grok, another AI chip startup. And Nvidia is in talks with OpenAI to sell Nvidia chips, a generative AI vendor representing 10 gigawatts.
Cerebras says it has also made deals with IBM and Meta. There was already a relationship between OpenAI and Cerebras, with OpenAI at one time exploring the possibility of acquiring the chip maker, The Wall Street Journal reported. OpenAI CEO Sam Altman is a private investor in Cerebrus.
