Conference Presentation, Product Demonstration
Scaling Global AI Infrastructure | Kevin Cochrane, Vultr | RAISE Summit 2026
- The era is characterized as the beginning of a long-term journey toward quantum computing, with the current phase transitioning out of the "trough of disillusionment" into an "agentic AI" epoch focused on deploying new AI-native applications and unlocking real enterprise outcomes.
- Enterprises plan to utilize platform engineering to templatize blueprints for rebuilding customer experiences around AI-native architectures, leveraging tools like "OpenClaw" to pre-build pipelines, services, and tuned models to accelerate iteration and development cycles.
- Financial projections suggest that integrating AI at the core of rethought cloud-native applications can double run-rate revenue, with hypothetical examples indicating an additional $75 in revenue generated per $200 room night.
- Future operational models anticipate global, seamless experiences that follow users across locations such as San Francisco, Paris, and Amsterdam, driven by agentic AI capabilities.
- Infrastructure requirements include the separation of inference workloads into distinct clusters optimized for power efficiency and price-to-performance ratios, distinct from training clusters, alongside dedicated CPUs for data pipelines and GPU orchestration.
- Compute strategy is shifting from centralized training to decentralized, distributed stacks in specific regions including the European Union, San Francisco, Paris, and Mumbai, with provisioning moving toward on-demand, API-driven models that eliminate deployment delays.
- A "complete AI factory" is launching today through collaboration with SUSE, NVIDIA, and Dell/NetApp to orchestrate CPU and GPU workloads, delivered on AMD GPU clusters with optional Vast storage on Super Micro hardware, featuring support for sovereign cloud operations.
- The new infrastructure stack is designed to automatically scale within traditional CI/CD pipelines to handle increasing workloads while ensuring security and compliance for global operations.
- Model efficiency is highlighted by the Voltron Retriever, which boasts an eight to 12 times smaller footprint than other retrievers, and the "Flash model," which is optimized to run on edge devices.
- Uncertainty remains regarding the specific evolution of the journey, as the speaker notes that the outcome depends on how the "adventure unfolds."