Modal Just Tripled to $15.75 Billion. The AI Infrastructure Wars Are Officially On.

Five months ago, Modal was a $4.65 billion company with a $355 million Series C and a bold claim: AI needed new plumbing. On Wednesday, the AI infrastructure startup settled the argument with a $750 million Series D led by Accel at a $15.75 billion post-money valuation. Valuation up more than threefold. Usage tripled. Annualized run rate above $1 billion. The market has delivered its verdict.

It is the kind of round that used to define an era. In May, Modal told Reuters its annualized revenue had passed $300 million. Now, per CEO Erik Bernhardsson's Wednesday announcement, the company has crossed $1 billion, growing more than 15 times year over year. Sandboxes, the product that gives AI agents isolated compute environments, is a $300 million run rate business on its own. Few startups have ever grown this fast in a business as capital intensive as renting GPUs by the hour.

What Modal actually sells

Modal sells the layer most AI users never see: the machines that run the models. Its cloud platform lets developers train models and run compute-heavy workloads without managing servers, covering inference, training, sandboxes, and functions. The pitch is serverless AI: you write code, Modal finds the GPUs, and you pay for what you use.

The customer list tells you who has the problem. Coding startup Cognition uses it. Music generator Suno uses it. Substack, the publishing platform, uses it. And Ramp, the fintech that just raised $1.85 billion at a $60 billion valuation, uses it too. The common thread is companies shipping AI features fast, where managing GPU infrastructure would be a distraction from the actual product.

Bernhardsson's framing was blunt: "Just five months ago, when we announced our Series C, I wrote that demand was accelerating. Since then, usage has tripled and we have grown more than 15x year-over-year. We recently crossed $1B in annualized run rate with no sign of slowing down."

The money is no longer asking who builds the best model. It is asking who owns the machines the models run on.

The battlefield is getting crowded

Modal is not alone, and investors know it. Bloomberg has reported that rival Baseten is seeking capital at around a $26 billion valuation, roughly double its June mark. Fireworks said in July that its own annualized revenue had reached $1 billion, a fivefold increase. Fal, which focuses on inference for image and video generation, has also been exploring a new round, per The Information.

Accel, leading the Series D, put the bull case plainly in its Wednesday post: the infrastructure to train, serve, and run AI is the fastest growing market in technology, and the industry is only in the very early stages of building it. As AI spreads from early "AI-natives" to every company, in the words of Accel's Rich Wong and Gonzalo Mocorrea, Modal aims to supply the building blocks for everyone from the smallest startups to the largest companies in the world.

Kleiner Perkins joined the new round, and every existing backer returned: Menlo Ventures, General Catalyst, Redpoint, Lux Capital, and Bain Capital Ventures. When existing investors all re-up at triple the valuation, it usually means the metrics they see privately are better than the public ones.

The uncomfortable question: margins

There is a catch, and it applies to the whole sector. Inference companies run on borrowed hardware: they rent or buy GPU capacity, then resell it with software on top. The gross margins are thinner than classic software, and the underlying compute cost is set by NVIDIA and the cloud giants, not by them. Fireworks, Baseten, Fal, and Modal are all racing to grow while absorbing that squeeze.

The bet is that scale and software differentiation fix the math. Serverless abstractions, better scheduling, and workloads like Sandboxes that bundle compute with developer workflow create pricing power that raw GPU rental does not. Modal's $300 million Sandboxes run rate is the evidence offered: customers are not just buying compute, they are buying into a platform.

Why this round matters more than the last one

Three things make this raise different from the last wave of AI funding. First, it is revenue-backed, not model-backed: a $1 billion run rate at Series D is the kind of number public markets respect. Second, it marks the center of gravity moving down the stack, from the labs training models to the infrastructure serving them. PYMNTS noted back in 2025 that inference would be the next big layer precisely because latency, cost, scale, and energy determine whether an AI service works or fails. That thesis just got $750 million of confirmation.

Third, Modal is going global. The company recently opened a London office, calling London the AI capital of Europe, and hired Hugh Killingbeck-Jones as VP EMEA to build out sales, solutions engineering, partnerships, and marketing teams. European customers bring different regulatory requirements, and a local footprint is how infrastructure companies win them.

The old question in AI was who builds the best model. The new question is who owns the cheapest, fastest, most reliable way to run one. Modal just sold investors its answer: worth $15.75 billion, and climbing.