High Performance Compute: Building the Cloud for AI Teams | Erik Bernhardsson | RAISE Summit 2026

Modal Origin and Company Culture

• Eric founds Modal after a two-year pandemic sabbatical, driven by 7 years of infrastructure-heavy experience at Spotify (90% infrastructure, 10% ML).
• The company's core thesis addresses the "infrastructure gap" in AI, aiming to rebuild the cloud from scratch for data, ML, and AI applications.
• Product-market fit took two years to achieve following the initial prototyping phase.
• Early hiring (first 50 employees) relied heavily on trusted networks (friends of friends) to ensure high agency and commercial focus.
• Hiring criteria prioritize intelligence, commercial acumen, product excitement, and the ability to clearly articulate the business value of technical work.
• Eric notes that predicting candidate success remains difficult despite 20 years of interview experience, leading to a focus on structured interviews and communication clarity.

Modal Product Capabilities and Infrastructure

• Modal manages infrastructure for AI application builders, currently operating ~20,000 GPUs with a target of 50,000 by year-end.
• The business model is usage-based, allowing customers to scale to 1,000 GPUs within seconds or minutes without managing capacity or routing.
• Key revenue drivers:
  • Inference: Constitutes ~67% of revenue, focusing on LLM, image, audio, and video generation.
  • Sandboxes: Used for agentic workflows, reinforcement learning, background tasks, and "vibe coding" platforms.
• Modal built its own proprietary file system, container runtime, and scheduler, explicitly avoiding Kubernetes and Docker to optimize developer experience.
• Recent product launch: AutoEndpoints (released two weeks prior to interview), a dedicated LLM inference product designed to replace manual coding for deployment.
• Future roadmap for AutoEndpoints includes "magic endpoints" that dynamically learn from traffic, optimize hyperparameters via agents, and improve performance over time.

Market Trends and Strategic Direction

• The industry is shifting from "frontier API" reliance (e.g., Anthropic) to open-source model fine-tuning and reinforcement learning as applications reach commercial maturity.
• Chinese open-source models (e.g., Qwen, Kimi, GLM5) are now competitive with frontier models, enabling cost-effective, high-performance deployments.
• Future High-Growth Areas:
  • Continuous Integration (CI): Identified as a bottleneck in current workflows (e.g., GitHub Actions) due to the need for rapid iteration with AI agents.
  • Real-time Audio/Video: Eric predicts breakthroughs in speech-to-speech models are imminent, driving investment in low-latency infrastructure and global CDN capabilities for inference.
• GPU market dynamics are characterized by a renewed shortage and rising prices, necessitating partnerships with 20+ cloud providers and plans to build proprietary data centers.

Ecosystem and Geographic Strategy

• Modal maintains offices in New York, San Francisco, and Stockholm, with plans to open a London office.
• Eric attributes Europe's lag in AI scale to cultural factors (lower "megalomaniac" global ambition) and structural issues (tax systems, stock option taxation) rather than a lack of talent.
• He advocates for European founders to "build for the world" rather than targeting the European market specifically.
• Spotify's Daniel Ek is cited as a key influence for maintaining a global vision from day one.