Interview, Fireside Chat, Conference Presentation
Mike Mattacola, Coreweave ｜ Raise Summit 2025
- The RAISE Summit 2025 took place in Paris as a global gathering focusing on AI infrastructure expansion, sovereignty, and energy constraints.
- CoreWeave received its first Dell AI factory rack containing NVIDIA H100 hardware and is currently stress-testing it in a lab environment.
- CoreWeave and NVIDIA's partnership enabled CoreWeave to launch specific GPU generations ahead of major competitors:
- H100s entered the market approximately nine months before larger players.
- H200s and GB200s were also released first by CoreWeave.
- The GB300 model was announced for market release "as of last week."
- To manage GB200/GB300 deployment, CoreWeave, Dell, and NVIDIA executed a joint, intensive troubleshooting process to move from hundreds of thousands to millions of potential failure modes.
- CoreWeave's competitive advantage lies in fine-tuned processes for rapid infrastructure onboarding, testing, and customer deployment rather than just hardware acquisition.
- NVIDIA's operational model supports high-speed innovation through a flat organizational structure with leaders managing small teams (10–20 people), minimizing meetings, and enabling same-day decision-making.
- CoreWeave mirrors NVIDIA's agility while acknowledging the critical need for supply chain planning that extends two to three years in advance for power acquisition, site approval, and data center construction.
- A significant supply chain trend involves consolidation of diverse components into single, high-density racks (e.g., GB200), moving away from traditional disparate hardware procurement.
- Infrastructure trends indicate a shift from general-purpose IT to specialized AI engineering, where data proximity to processing (via HBM) and advanced networking fabrics (NVLink, KVCache) are primary drivers.
- Intel's market dominance is in decline, with x86 market share reported to have dropped from 73% to single digits (the "teens") within a single year.
- CoreWeave views the current era as a return to "speeds and feeds," contrasting with a previous "solutions" phase, driven by the physical requirements of modern AI data centers.