🧠 Food for thought

Implications, context, and why it matters.

Modal’s valuation rests on a custom multi-cloud stack

• Modal did not build on one cloud vendor alone. It wrote its core stack in Rust, which is known for speed and reliability 1.
• That stack includes a custom filesystem for near-instant container starts plus a runtime that uses memory snapshots to cut cold starts for large AI models to under three seconds 2.
• Another piece is its resource solver. The internal tool uses linear programming, a math optimization method. It moves graphics processing unit (GPU) capacity among Amazon Web Services (AWS), Google Cloud Platform (GCP), and Oracle in real time based on price plus availability 3.
• Its sandbox product gives customers such as Quora’s Poe, Quora’s AI chatbot platform, secure isolated environments for untrusted code 2.

Multi-cloud software is becoming a new pressure point for hyperscalers

• Modal’s rise suggests a new software layer. It can treat hyperscale cloud providers as interchangeable suppliers for some workloads, which can loosen ties to any single cloud ecosystem 2.
• By sending workloads across clouds plus regions, Modal can keep value that might otherwise go to AWS, Google, or Oracle through compute usage and related services 4.
• This approach appears in Ramp, a corporate spend management and finance software company. Per Contrary Research’s summary, it saved about 79% against other major large language model (LLM) providers on a workflow with receipt processing-related steps 1.
• Across the industry, this could push hyperscale cloud providers to compete harder on compute prices. It also could make multi-cloud cost control a more common goal for businesses 2.