Modal CEO discusses growth and innovations in AI infrastructure

Modal: Modal is an AI infrastructure platform offering serverless compute, storage, functions, and sandboxes to help developers build and scale AI applications such as inference and agent workloads. It aggregates demand across thousands of customers to provide elastic GPU capacity rather than fixed long-term contracts. In the news, Modal’s CEO discusses the company’s approach to handling bursty AI workloads, the growth of its sandbox product, and strategies for sourcing capacity from multiple providers. NVIDIA: NVIDIA is the dominant supplier of GPUs powering AI training and inference workloads. Modal procures capacity from a wide range of providers that rely on NVIDIA hardware while also evaluating alternatives. The news explores market dynamics around NVIDIA supply, pricing trends, and the feasibility of non-NVIDIA accelerators. OpenAI: OpenAI is a leading AI research lab that purchases substantial GPU capacity for large-scale model training and inference. It is referenced in discussions of concentrated demand and how smaller or alternative clouds can still secure allocations. The news situates OpenAI within broader conversations on capacity allocation and market concentration. Spotify: Spotify is a music streaming platform where Erik Bernhardsson spent seven years building its early music recommendation system using matrix factorization and other compute-heavy techniques. His time there highlighted limitations in existing cloud infrastructure for AI workloads, directly inspiring Modal’s serverless approach. The news recounts this background as context for Modal’s founding thesis. Anthropic: Anthropic is an AI company that integrates Modal sandboxes into its managed agents ecosystem through a bring-your-own-sandbox model. The partnership supports secure execution environments for agent workflows. The news highlights this collaboration alongside broader topics of sandbox security and ecosystem development. Akshat Bubna: Akshat Bubna is co-founder and CTO of Modal, where he has contributed to core infrastructure primitives and early product development. He co-created the initial sandbox prototype after recognizing the need for environments to run code generated by language models. The news references his prior interviews and ongoing role in shaping Modal’s multi-product strategy for agents and inference. Basil Chatha: Basil Chatha is a new cohost on the podcast series, joining from the forward-deployed side of AI engineering. He participates in the interview with Modal’s CEO on the occasion of the company’s Series D. The news introduces him as part of ramping up new show content focused on AI infrastructure topics. Erik Bernhardsson: Erik Bernhardsson is the founder and CEO of Modal, with prior experience building recommendation systems at Spotify and leading engineering at Better.com. He co-founded Modal to address shortcomings in infrastructure for compute-intensive AI and machine learning tasks. The news centers on an extended interview with him covering Modal’s product evolution, GPU economics, and vision for ephemeral, high-performance sandboxes. GPU Contracts: Traditional GPU capacity contracts lock buyers into fixed blocks of chips for multi-year periods that rarely match the variable, bursty patterns of real AI workloads. Sandbox Growth: Demand for isolated execution environments has risen sharply with the rise of coding agents, background agents, and reinforcement learning applications. Capacity Sourcing: Infrastructure providers are increasingly securing multi-year GPU reservations from a diverse set of clouds and neoclouds to meet growing elastic demand.