Artificial intelligence chip upstart Cerebras Systems (CBRS.US) announced on Thursday that it is powering a new "Ultrafast" mode for OpenAI's GPT-5.6 Sol model. This new processing mode can deliver speeds up to 14 times faster than OpenAI's standard configuration and is currently being rolled out to a select group of clients requiring ultra-low latency.
On its website, Cerebras stated: "The Ultrafast mode is initially available only to a small group of customers, with access set to expand gradually. Cerebras provides the computing power for the GPT-5.6 Sol Ultrafast mode, achieving speeds of up to 750 output tokens per second without any quality trade-offs. This allows Sol Ultrafast to accelerate the most time-sensitive and mission-critical workloads."
In a separate post, OpenAI commented: "Early usage is helping us understand where order-of-magnitude speed improvements create the most value, and how the product changes when the model can keep pace with the user. As compute capacity increases, we will use these insights to guide deployment."
OpenAI also noted that it has identified several "encouraging" use cases for the Ultrafast mode, including: event response and reliability, financial research and security, customer support and voice, commercial applications, and real-time research and experimentation. The GPT-5.6 Sol Ultrafast mode is currently available as a limited preview and will be gradually expanded. Early adopters of GPT-5.6 Sol Ultrafast include quantitative trading firm Jane Street, AI lead generation company Podium, and financial firms Basis and Rogo.
Cerebras shares fell 11.85% in Thursday trading on the US stock market, after dropping nearly 15% during the session. Company data shows Q2 revenue rose 74% year-over-year to $180.1 million, but this figure fell short of the analyst average estimate of $194 million. The company reported a loss per share of $2.98, significantly worse than the analyst consensus estimate of a $0.17 loss per share.
By segment, Q2 hardware revenue decreased 23% year-over-year to $54.1 million, suggesting the startup with an innovative chip architecture still faces challenges in commercial deployment. However, cloud and other services revenue reached a record high, surging 281% year-over-year to $126 million. Since its IPO in May, Cerebras' stock has risen nearly 29%.
Cerebras' core technological selling point is its unique high-end processor design philosophy — processing a silicon wafer typically used for multiple components into a single, massive chip. The company has built an extensive network of data centers and offers compute power leasing services, initially to validate its technology's feasibility. The current surge in AI compute demand has turned this leasing service into a key growth engine. While Cerebras positions itself as a challenger to Nvidia in the AI chip space, its largest revenue source is now cloud computing services.
Cerebras CEO Andrew Feldman stated: "The hardware business will be volatile due to the timing of order delivery and revenue recognition. That's inherent to the industry." He added that some customers are not yet ready with the data center space needed to accommodate new computing systems, but emphasized that the company remains committed to selling both hardware systems and data center services.
Looking ahead, Cerebras forecasts Q3 revenue of approximately $215 million, above the analyst average estimate of $212 million. Core gross margin is expected to be between 38% and 40%, also exceeding the analyst average estimate of 36%. The company raised its full-year revenue guidance to between $880 million and $890 million, up from its previous range of $855 million to $865 million, compared to the analyst consensus of $867.6 million. The company also expects full-year core gross margin to be between 41% and 43%.