Sachin Katti
Showing 1–3 of 3 transcripts.
- RAISE Summit12 min
Agents are Here. Compute has to Change | Sachin Katti, OpenAI | RAISE Summit 2026
OpenAI has entered an agentic era where its internal roadmap is realized through AI systems handling nearly all internal token generation and conducting autonomous research tasks. This shift has compressed model release cycles from half a year to monthly intervals by enabling 50x growth in research experimentation and expanding compute demands to include synthetic data and reinforcement learning. As non-engineering departments and AI interns drive adoption, the organization is simultaneously rebuilding its infrastructure with heterogeneous hardware and complex networking to support the iterative, tool-heavy workflows required for this new phase of recursive development.
- RAISE Summit22 min
OpenAI x Cerebras: Sachin Katti & Andrew Feldman in Conversation | RAISE Summit 2026
Sachin Katti, Andrew Feldman, Henri Delahaye
In December 2024, OpenAI and Cerebras finalized a historic $20 billion partnership to deploy GPT-5.6 on Cerebras infrastructure, enabling a record-breaking inference speed of 750 tokens per second while expanding data center capacity across Europe to address sovereign AI demands. This collaboration has accelerated OpenAI's internal operations, establishing Codex as the default interface for all employees and driving a monthly release cadence of frontier models through AI-assisted research workflows. Looking forward, the alliance prioritizes enterprise adoption over token volume, anticipating that the rapid integration of AI agents will fundamentally rewire organizational structures and redefine productivity metrics within the next year.
- Stanford Online46 min
Stanford MS&E435 Economics of the AI Supercycle | Spring 2026 | Infrastructure, Capstone Case
Former Intel CTO Kati Katti, now at OpenAI, details the organization's aggressive pursuit of 30 gigawatts of compute capacity to support a future dominated by complex agentic workloads and massive inference demands. This strategy involves overcoming severe supply chain bottlenecks and grid constraints through specialized infrastructure like nuclear power and Cerebras accelerators while prioritizing gigawatt-scale deployment over fragmented edge solutions. Katti predicts the AI value chain will eventually shift from hardware foundations to application-layer outcomes, as the industry races to solve the critical shortage of logic and memory fabrication capacity.