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
- Market Paradigm Shift: The AI infrastructure sector is transitioning from the multi-tenant CPU architecture of the 20th century to a GPU-centric model where single servers cost over $3 million (e.g., NVIDIA GB200) and cannot be time-sliced by individual users.
- Power Density Trends: Data center rack densities are scaling from 10–12 kilowatts in legacy CPU environments to 140 kilowatts per rack for current NVIDIA servers, with next-generation targets reaching 800 kilowatts per rack.
- Speed as a Competitive Moat: NeoCloud providers leverage organizational agility to outpace hyperscalers, citing Crusoe's construction record of bringing two 103-megawatt data center buildings online in just over 300 days, with subsequent six buildings scheduled for completion in under 200 days.
- Software Virtualization Capabilities: Providers are developing advanced software to dynamically fracture massive clusters (e.g., 8,000 GPUs) into smaller sub-clusters (e.g., 512 GPUs) with allocation cycles as short as 15 minutes to match startup needs.
- Sovereign Cloud Imperative: European entities face a critical compute deficit, holding less than 5% of global compute power compared to over 60% held by US hyperscalers, driving a strategic push for localized infrastructure.
- Infrastructure Investment Targets: Sorrenti aims to utilize France's 5 gigawatts of energy excess to build gigawatt-scale AI factories, targeting a total capacity to ensure Europe does not rely on foreign hyperscaler solutions.
- Enterprise Adoption Barriers: Current enterprise hesitation stems from the high cost of switching ecosystems, the fragmented nature of the 120+ existing NeoCloud offerings, and the lack of simplified, "experience-first" products compared to hyperscaler convenience.
- Strategic Divergence on Value Chain: A debate exists regarding the optimal long-term value layer, with some experts arguing that infrastructure and energy will remain the primary value drivers as AI makes software a commodity, while others contend that moving up-stack to managed services and inference platforms is necessary for enterprise competitiveness.
- NVIDIA Market Disruption: NVIDIA's acquisition of Lepton and release of open-source inference libraries (e.g., vLLM, SGLang) presents a potential commoditization risk for cloud providers' software layers, though some panelists argue this does not diminish the value of specialized physical infrastructure.
- Long-Term Business Outlook: Projections suggest that while current revenue is driven by Infrastructure as a Service (IaaS) due to insatiable demand (e.g., 10 gigawatts of new capacity planned for the US in 18–24 months), future revenue will increasingly consolidate with model builders and sovereign entities controlling Artificial General Intelligence (AGI).
- Vertical Integration Strategy: Leaders like Nebius and Sorrenti advocate for building the full stack from ground-level data center construction to software, arguing that this integration allows for faster deployment of new chips and optimized cooling architectures.
- Ecosystem Distribution Challenge: Hyperscalers maintain dominance through aggressive top-down acquisition strategies and established ecosystems, creating a "lock-in" effect where CTOs and CIOs perceive the switching cost to NeoClouds as prohibitive regardless of price.