OpenAI x Cerebras: Sachin Katti & Andrew Feldman in Conversation | RAISE Summit 2026

Strategic Partnership & Deal Terms

• OpenAI and Cerebras formalized a collaboration in December 2024 valued at over $20 billion for multi-year compute delivery.
• The partnership addresses the critical need for low-latency inference as AI models transition from high-quality outputs to interactive, real-time workflows.
• The deal is described by Cerebras leadership as one of the largest in Silicon Valley history, specifically targeting the portion of workloads where speed is a primary determinant of value.

Performance Capabilities

• The collaboration enables the deployment of the GPT-5.6 model (referred to as "5.6" in the transcript) on Cerebras infrastructure.
• GPT-5.6 is projected to run at 750 tokens per second, a performance level described as an order of magnitude faster than competing frontier models.
• This speed is positioned as a defining factor for user experience, analogous to the shift from search quality to latency optimization that drove Google's 2010s growth.

Internal Adoption & Productivity Metrics

• Codex has become the default interface for OpenAI employees, replacing traditional browsers for internal tasks ranging from engineering to legal and HR functions.
• Internal usage includes complex tasks such as HR-driven human resource reorganizations, demonstrating the capability of AI agents to manage organizational structures.
• OpenAI's primary productivity metric is the release cadence of frontier models, which has accelerated to one new model per month due to AI-assisted research workflows.
• Leadership advises enterprises to prioritize business output metrics over token usage counts, noting that token volume is merely a means to achieve efficiency, not a proxy for AI-native maturity.

Global Infrastructure Expansion

• Cerebras announced a $200 million investment (implied by "billions of dollars" context and scale, specifically cited as building 200 megawatts of capacity) to support OpenAI in Europe.
• New data center capacity is being deployed in Lyon, France; Norway; and Finland, with delivery targets of late 2024 for a portion and completion by the end of 2025.
• The expansion is driven by "sovereign AI" trends, recognizing intelligence infrastructure as a critical national resource that nations cannot afford to import-dependently.
• The strategy aims to ensure parity in infrastructure availability between the U.S., Europe, and Asia to prevent geographic bottlenecks in innovation.

Enterprise Transformation & Organizational Shifts

• The widespread adoption of agents is expected to fundamentally rewire enterprise structures, challenging traditional team sizes, reporting lines, and role definitions.
• Organizations are encouraged to experiment with "AI workers" and agents reporting to specific roles to determine optimal integration before standardizing processes.
• The consensus is that enterprise AI integration is a learning curve requiring iteration, where early failures are expected before achieving a competitive advantage through optimized agent swarms.

Forward-Looking Outlook

• The pace of AI capability improvement is accelerating, with leaders stating that the rate of change in the next 12 months will be "astonishing."
• The primary focus for the coming year is shifting from model development to enterprise and consumer adoption, ensuring capabilities translate to real-world daily workflows.
• Infrastructure remains a universal bottleneck, necessitating heterogeneous investments in CPUs, GPUs, networking, and storage while driving software optimizations for efficiency.
• The market is moving from a phase of "time to market" for new models to a phase of "efficiency" and scaling for existing intelligence.