Cerebras Stock Slides on Report OpenAI Tapped Nvidia for 'Ultrafast' AI Tier

Shares of AI chipmaker Cerebras Systems tumbled on Monday following industry speculation that OpenAI chose hardware from rival Nvidia to power its newly announced high-speed AI service, dealing a blow to investor confidence.

What Happened

The sell-off, which saw Cerebras stock fall by as much as 7%, was triggered by a report from semiconductor research firm SemiAnalysis. In a post on X (formerly Twitter), the firm stated that OpenAI's new "GPT-6.1 Sol Ultrafast" model tier runs on traditional Nvidia GPUs at a "low batch size," rather than on Cerebras's specialized chips.

OpenAI's "Ultrafast" tier is designed for real-time applications, promising token generation speeds up to eight times faster than standard deployments. Cerebras has not yet issued a public statement confirming or denying its involvement in the project.

Why It Matters for Investors

The report directly challenges a core component of Cerebras's value proposition. The company's technology is specifically architected to excel at high-speed, low-latency AI inference, a task that often requires processing requests individually or in small groups—a technique known as a "low batch size."

Ad

• Cerebras's Advantage: The company's massive Wafer-Scale Engine (WSE) is a single chip designed to hold entire neural networks on its silicon, minimizing the data movement that can create latency. This is considered ideal for low-batch-size workloads.
• Nvidia's Position: Traditional GPUs typically achieve maximum cost-effectiveness by processing large batches of user queries simultaneously. If OpenAI has successfully optimized Nvidia's hardware to run a frontier model efficiently at a low batch size, it could reduce the need for specialized alternatives.

Competitive Landscape

For an AI hardware challenger like Cerebras, securing a contract to power a flagship service from a top-tier lab like OpenAI is a critical milestone. The speculation raises concerns about Cerebras's ability to penetrate the highest end of the market, which remains dominated by Nvidia and its deeply entrenched CUDA software ecosystem.

If market leader Nvidia can adapt its existing infrastructure to meet the ultra-low latency demands of next-generation AI, investors fear that specialized hardware providers may face a more difficult path to securing major data center contracts.