Infrastructure as Destiny: The Compute-Capital-Cloud Trinity | RAISE Summit 2026

Market Dynamics and Capital Investment

• Demand vs. Supply Gap: Demand for AI compute has outpaced supply, with cloud providers reporting no issues with "offtake" (customer subscriptions) for the first time in years, shifting from speculative investment to immediate, high-demand deployment.
• Capital Availability: Capital is abundant, driven by both investment-grade and non-investment-grade debt, yet the primary constraint remains physical infrastructure availability rather than funding.
• Neocloud Expansion: Neoclouds have shifted from long planning cycles (hyperscalers) to 2–3 month cycles, leading to rapid, leveraged build-outs that strain supply chain lead times, which are now running up to two years.
• Storage Prioritization: Memory and storage have moved from secondary considerations to primary planning criteria, with customers signing multi-year agreements (up to 4 years) valued in the tens of billions of dollars annually.
• Capacity Underwriting: Infrastructure capacity is currently underwritten through 2028, a historical high that allows vendors to plan long-term despite the volatility of the sector.

Operational Efficiency and Technology

• Non-Human Traffic Surge: Cloudflare reports that over 50% of its network traffic is now non-human (agentic), driving exponential growth in compute and networking requirements.
• Utilization Optimization: Cloudflare maintains 15% or less CapEx as a percentage of revenue despite massive growth by utilizing a 330-location, 125-country network for optimized traffic routing and high resource utilization.
• Isolate Architecture: Cloudflare advocates for "isolates" over containers, claiming 100x efficiency gains for agentic workloads to avoid the necessity of deploying billions of CPUs for a projected billion-agent ecosystem.
• Data Compression: Vast Data utilizes proprietary compression algorithms to double or triple effective storage capacity, currently accounting for 10–20% of global enterprise flash deployments and alleviating supply constraints.
• Fault Tolerance: Clockwork Systems focuses on reducing underutilization caused by infrastructure failures, aiming to boost cluster utilization from 30–40% to double that by masking hardware failures during training and inference.

Supply Chain and Infrastructure Challenges

• Interconnect Bottlenecks: Copper and optical interconnects are identified as primary barriers to growth, with lead times extending into 2028; neoclouds struggle to provide the visibility suppliers need to forecast demand.
• Hardware Failure Rates: GPU out-of-service rates in some neocloud clusters have reached 20%, resulting in significant revenue loss (e.g., $1,000/week per GPU idle) and necessitating vendor liability sharing for uptime penalties.
• Operational Rigor: Unlike hyperscalers with deep engineering benches, neoclouds require vendors to assume liability and perform "deep partnerships," including testing and troubleshooting complex fiber and optical fabrics that lack pre-deployment testing.
• Legacy Hardware Revival: Vast Data assists customers in repurposing existing hardware (disassembling arrays, stripping servers) to extend lifespan, with demand for "2025 economics" on older hardware.

Vendor Relationships and SLAs

• Shift to Partnership: The distinction between vendor and customer is blurring, with vendors increasingly assuming liability for uptime penalties and engaging in deep technical co-engineering to ensure resiliency.
• SLA Implementation: Customers are increasingly using Service Level Agreements (SLAs) to formalize reliability standards, moving from litigious post-incident reviews to pre-deployment joint testing in labs (e.g., NVIDIA GPU labs).
• Resilience Investment: A single 2,000-GPU deployment improved operational efficiency by a factor of four against link failures, saving approximately 1,000 GPU hours daily that would otherwise be wasted on retraining.
• Technical vs. Legal: An inverse correlation exists where more technically sophisticated customers tend to be less litigious, focusing on implementation and management rather than contract disputes.

Case Studies and Incidents

• Data Recovery Emergency: Solidigm engineers successfully recovered 10 petabytes of deleted data for a largest neocloud/AI lab customer in two weeks, preventing a 6–12 month delay in model deployment.
• Infrastructure Retrofit: Credo intervened in a catastrophic data center build-out (where drywall installation interfered with optics), spending six months diagnosing failures to prevent a full interconnect rip-out.
• Agentic Security: Cloudflare assisted customers like PeopleLink in distinguishing between legitimate AI traffic and malicious agents stealing customers, shifting focus from simple CDN delivery to security and traffic classification.

Future Outlook (3–5 Years)

• Infrastructure Maturity: Jeff Denworth (Vast) and Greg Mattson (Solidigm) predict a multi-decade expansion cycle, with infrastructure scaling continuing to outpace the 20–25 year timeline of the traditional internet.
• Market Consolidation: Suresh Vasudevan (Clockwork) foresees consolidation among the hundreds of current neocloud providers, shifting toward a more stable, utility-like cloud marketplace.
• Ubiquitous Inference: AI inference will become embedded in every application, ceasing to be a distinct service layer and becoming a standard cloud utility.
• New Business Models: Stephanie Cohen (Cloudflare) predicts the emergence of a machine-to-machine commerce model for the "agentic internet," where every transaction is a commercial opportunity.
• Distributed Architectures: Greg Mattson (Solidigm) highlights the exploration of non-traditional data center locations, including space-based and underwater facilities, driven by the need for distributed systems.
• Interconnect Evolution: Don Barnetson (Credo) notes that interconnect growth is outpacing GPU growth (6x larger than two years ago), requiring major technology upgrades every two years to maintain bandwidth and scalability.