Fully Connected 2026: East Coast Recap | CoreWeave
Summary
This recap details CoreWeave's strategy for scaling full-stack AI infrastructure, focusing on the 'AI loop' (Run, Observe, Curate, Improve, Evaluate). Key announcements include CoreWeave Forge, a unified platform, and the introduction of the Vera CPU from NVIDIA. The Vera CPU is designed specifically for the agentic era, featuring Olympus cores, high memory bandwidth, and a scalable coherency fabric, with customers reporting significant performance gains in agent orchestration and data analytics.
Key takeaways
-
The AI Loop: A Systematic Approach to AI Development
10:42
CoreWeave introduced a systematic AI loop (Run, Observe, Curate, Improve, Evaluate) to address scattered signals and 'glue work' in complex AI pipelines, moving away from manual human-in-the-loop processes [6:42].
-
CoreWeave Forge as a Unified Platform
13:56
Forge integrates various stages (inference, sandboxes, observation, curation, improvement) into a single pane of glass, providing observability down to the underlying infrastructure, which is critical for reliable performance [8:36].
-
Vera CPU for the Agentic Era
The second-generation Vera CPU is optimized for the demands of agentic workloads, requiring fast cores, memory, and consistent latency to keep the GPU utilized. It utilizes Olympus cores and a scalable coherency fabric [26:10].
More digests on AI Infrastructure
Technical details
-
CoreWeave AI Tools
1849s
Core Agent Lens helps group failure domains and provides observation skills within deployed agents. ARIA (AI research iteration agent) allows users to ask questions and receive instant feedback on infrastructure and configuration, and can provide concrete suggestions and code examples [18:49].
-
Vera CPU Architecture
The Vera CPU features Olympus cores (an ARM-compatible, proprietary Nvidia design core) and Spatial Multi-threading (SMT), which partitions cores completely for better isolation. It boasts 1.2 TB/s of LPDDR5X memory bandwidth and a 3.4 TB/s bisectional bandwidth scalable coherency fabric [32:53].
-
Performance Benchmarks
Customer tests demonstrated significant performance improvements: one customer reported 1.5x faster job completion with a proxy Python workload, another saw 1.8x better agent orchestration, and one sandbox team reported 3x faster sandbox speed compared to x86 systems [36:53].
Mentioned resources
- CoreWeave Forge
- Core Agent Lens
- ARIA
- Vera CPU
- Olympus cores
Channel & topics
Watch on YouTube · Back to latest
This independent, AI-assisted summary is provided for commentary and informational purposes. It may contain errors or omit important context. Please watch the original video for the creator's complete presentation. Video, thumbnail, and related copyrights belong to their respective owners.