Powering AI Infrastructure: Solving the 21st Century Manhattan Project | RAISE Summit 2026

Market Scale and Growth Projections

• Anthropic is estimated to have held 1.5 gigawatts (GW) of capacity by the end of 2025, projected to reach 10 GW by 2027.
• This growth represents a 9 GW addition over two years, equivalent to Google's current capacity, with four companies pursuing similar scaling timelines.
• CoreWeave reported an open order book value of $23 billion as of March 31 (prior year), rising to $99 billion by March 31 of the current year.
• Sean Amara (Mirantis) projects that inference will comprise 75% of all GPU workload by the end of 2027.

Primary Operational Bottlenecks and Challenges

• Execution Speed vs. Complexity: The single biggest challenge is the speed of execution required to integrate complex systems without slowing the entire infrastructure down.
• Data Center Readiness: Over 50% of data center projects in the U.S. are currently canceled due to permitting, contracts, or power delivery issues.
• Supply Chain and Hardware: Delays are frequently caused by the complexity of the supply chain ecosystem, including missing components and long lead times for specialized hardware.
• Power Infrastructure Evolution: Rapid technological shifts require data centers to upgrade from standard 415V/277V power delivery to DC bus bars operating at 400V or 800V.
• Regulatory Fragmentation: Building infrastructure in Europe is significantly more difficult than in the U.S. due to varying national regulations, languages, and municipal requirements.

Strategic Responses and Architectural Decisions

• Modular and Open Standards: Ben Richardson (CoreWeave) and Vik Malyala (Supermicro) emphasize the necessity of building with open APIs and standardized layers to allow interoperability between different hardware vendors (NVIDIA, AMD, Intel, ARM).
• Building Block Approach: Supermicro validates platforms from the initial server design through the rack level, integrating compute, networking (Arista, Cisco, Broadcom), and storage (DDN, VAST, Weka) to ensure time-to-online efficiency.
• Digital Twin Technology: Netris utilizes digital twin simulations to validate GPU cluster architectures and network configurations before hardware arrival, reducing deployment errors and time-to-live.
• Cost and Efficiency Focus: Furious AI (Jun) and Netris highlight the rising unsustainability of energy and CapEx costs, driving the need for full-stack solutions that lower Total Cost of Ownership (TCO) for inference.

Market Segmentation and Customer Demand

• Dominant Customers: Demand is currently driven heavily by "Nine out of Ten" largest model labs and hyperscalers, who are overwhelmingly requesting NVIDIA stack solutions.
• Enterprise Shift: Enterprises are increasingly moving toward private infrastructure to control costs, ensure data sovereignty, and diversify supply chains beyond single-vendor reliance.
• Emerging Segments: Telcos and sovereign cloud providers are the next wave of adopters following the initial success of Neo-clouds.
• European Market Dynamics: Smaller European operators face constraints in power capacity (often limited to 5–10 MW), necessitating retrofits, liquid cooling adoption, and modular data centers rather than massive builds.

Risk Factors and Industry Critique

• False Confidence: The industry culture of announcing projects before execution creates a "bubble" perception, as few have the operational experience to build deep-complexity assets.
• Operational Reliability: Sean Amara compares modern AI infrastructure to Formula One cars requiring specialized talent, whereas traditional data centers were likened to BMWs; failure tolerance is near zero.
• Talent Gap: There is a shortage of operational expertise, with many new entrants lacking the 20 years of collective cloud-building knowledge required to manage reliability at scale.

Forward-Looking Outlook

• Inference Migration: The primary growth vector is shifting from massive training clusters to highly distributed, lightweight inference workloads within enterprise environments.
• CFO Decision Making: Future adoption will be driven by ROI analysis and the development of formal AI policies within enterprises rather than purely technological capability.
• Collaborative Ecosystems: Success requires a "village" approach where vendors (chips, networking, systems, software) act as a collective unit to solve failures and maintain uptime, rather than operating in silos.