Conference Presentation, Fireside Chat, Interview
OpenAI x Cerebras: Sachin Katti & Andrew Feldman in Conversation | RAISE Summit 2026
- Starting in 2025, models are expected to achieve sufficient capability for widespread adoption, with inference speeds becoming a critical product attribute alongside quality and accuracy.
- A collaboration agreement signed in December 2024 commits "north of $20 billion" in compute capacity over the subsequent several years to deliver high-performance results.
- The Cerebras 5.6 model is projected to run at 750 tokens per second, described as an order of magnitude faster than existing alternatives, creating a "magical" user experience.
- Codex is transitioning to the default company interface for engineers and is expected to expand into legal, go-to-market, and finance functions to handle increasingly complex tasks.
- AI agents are anticipated to perform advanced organizational tasks, such as "human reorgs," while software optimizations aim to alleviate supply chain bottlenecks in memory, GPUs, and CPUs.
- OpenAI model releases are forecast to accelerate to approximately one per month as AI tools increasingly automate AI research itself.
- Enterprise productivity metrics are shifting from token counts to measuring inference speed and the intelligence gain of each model generation.
- Infrastructure requirements are becoming highly heterogeneous, driving demand for increased CPU, GPU, networking, storage, and memory resources.
- Data center capacity is a identified bottleneck, with 200 megawatts expected to be completed by the end of next year across France, Norway, and Finland, though partial delivery may occur this year.
- The company is actively sourcing supply globally and employing creative software strategies to manage constraints, acknowledging the absence of a "silver bullet" solution.
- The global supply chain is expected to experience exponential growth that simultaneously creates new bottlenecks across all sectors.
- Sovereign AI is projected to become a critical national resource as nations seek to reduce dependency on foreign intelligence infrastructure.
- AI capability improvements are expected to continue accelerating for the next 12 months, leading to a focus on ensuring broad access to these capabilities thereafter.
- Uncertainty regarding short-term developments remains high, with 12 months viewed as a significant timeframe and the specific trajectory for the next three months considered unpredictable.