Cerebras AMD partnership lifts shares as AI latency race intensifies

Cerebras shares climbed about 4% on Thursday after the Cerebras AMD partnership put the AI chipmaker into Advanced Micro Devices’ Helios systems, according to CNBC. The move links Cerebras’ chips to AMD’s broader AI infrastructure push and comes as investors track demand for systems designed to generate AI responses with very low delay.

Cerebras Chief Executive Andrew Feldman said at AMD’s AI conference in San Francisco that Cerebras chips will be installed in AMD Helios AI systems used in Cerebras data centers beginning later this year. He also said buyers of servers will be able to configure AMD systems with Cerebras’ wafer-scale chips.

AMD used the event to outline new chips and its Helios integrated system, according to CNBC. The companies said their combined system would deliver five times more tokens per second per watt than competing systems, a performance measure that compares AI output with energy use.

What is the Cerebras AMD partnership?

The agreement brings Cerebras chips into AMD’s Helios AI systems, with deployment in Cerebras data centers slated to start later this year. It also gives server buyers a way to include Cerebras wafer-scale chips in AMD system configurations, according to Feldman’s comments at the event.

For AI infrastructure customers, latency refers to how quickly a system can begin returning an answer after receiving a prompt. CNBC reported that Cerebras-style chips are designed to produce the first AI responses as fast as possible, while making tradeoffs in flexibility and overall power use.

Feldman framed speed as a demand issue for AI services. “When something’s a necessity, people want to use it, and they want to use it quickly,” he said, according to CNBC.

Why low-latency AI systems are drawing investment

The AMD-Cerebras announcement reflects a wider contest among AI hardware suppliers to improve inference, the process of running AI models to generate outputs after training. Inference systems are increasingly judged by speed, energy efficiency and the ability to handle high volumes of user requests.

Nvidia has also moved in this direction. CNBC reported that the company bought assets from Groq in December for $20 billion to bring low-latency technology into its own systems.

Cerebras has been a volatile public-market name since its May listing. The company went public at $185 a share, climbed as high as $386.34 during its first trading session and later fell below $161 in late June, according to CNBC. After Thursday’s gain, the stock was trading at $219.80.

The AMD agreement follows another large customer announcement for Cerebras earlier this year. In January, Cerebras announced a contract with OpenAI to provide 750 megawatts of computing power through 2028, a deal valued at more than $10 billion, CNBC reported.

The latest move gives AMD another partner in AI systems as it presents Helios to customers, while giving Cerebras a new route for placing its chips inside data center infrastructure. The financial terms of the AMD-Cerebras agreement were not disclosed in the reported announcement.

This story draws on original reporting from CNBC.