Panel
RAISE Summit 2025: AI Infrastructure. Compute is the New Oil
RAISE SummitRoman Chermin, Youssef El Manssouri, Shaun O'Meara, Stephen Balaban, Chase Lochmiller, Dylan Patel
- The 21st century economy is projected to be driven by AI, with growth historically rooted in energy.
- Nebius plans to build a full stack from the ground up, including data center construction, hardware, and software for customers ranging from large labs to startups.
- Sisters Cloud targets the construction of hundreds of megawatts to one gigawatt of AI factories in France and Europe.
- Mirantis software aims to dynamically fracture large clusters, such as an 8,000 GPU unit, into sub-clusters allocated as frequently as every 15 minutes.
- Crusoe completed its first two 100-megawatt data center buildings in just over 300 days, with six subsequent buildings scheduled to come online in just over 200 days.
- Collective new AI infrastructure capacity in the U.S. is expected to rise by 10 gigawatts over the next one and a half to two years.
- The EU currently holds less than 5% of global compute power, while hyperscalers control more than 60%.
- Future revenue generation from all global floating point operations (flops) in 10 years is predicted to largely accrue to those who control the models.
- Nebius anticipates that inference platforms will be the first software layer aggregated from infrastructure as the industry shifts from prototypes to production workloads.
- The industry is expected to move from serving large labs and hyperscalers toward a primary focus on enterprises, where the main long-term competition resides.
- Software development cycles are predicted to shorten significantly, potentially replacing current complex software models with direct prompting to generate services or products.
- Long-term business value is seen in enterprise sectors where companies currently pay two to three times the price found on Amazon due to market confusion.
- The market structure is expected to become more oligopolistic in certain areas due to economies of scale, though not necessarily resulting in pure winner-take-all dynamics.
- A "risk of existence" is identified for many ecosystem players if hyperscalers aggressively expand their capabilities and acquisition strategies.
- Competition with hyperscalers is characterized as a very difficult task involving large ecosystems and top-down strategies.
- High switching costs for enterprises moving from hyperscalers to alternative clouds make price competition challenging.
- Large tech companies' decision-making architectures are designed to be slow, causing delays that compound when bringing infrastructure to market.
- The infrastructure landscape must adapt to the high power density required by AI accelerators, acknowledging that AI consumes significant energy.
- Success in the enterprise space requires moving away from selling raw infrastructure toward API-friendly, developer-focused products and partnerships.
- While infrastructure-as-a-service currently generates significant revenue due to insatiable demand, the long-term business model shifts toward platform and purpose-driven software.