Key Takeaways

  • Modal secured a $15.75 billion Series D valuation after growing from under $100 million to a $1 billion revenue run rate in five months.
  • Jason Lemkin projects Modal will hit $1.4 billion by year-end and $5 billion next year, making the price tag cheap even if gross margins settle below 40%.
  • The top four or five open-weight inference providers generate $5 billion to $6 billion in combined run rate, which Rory O'Driscoll calculates is less than 5% of the $100 billion-plus earned by OpenAI and Anthropic.
  • Neoclouds like CoreWeave absorb heavy capital expenditures and debt financing, creating pressure for them to build native inference layers and recapture developer margins.

The Disagreement

Lemkin views Modal's $15.75 billion valuation as an obvious bargain because its software layer captures developer loyalty faster than traditional enterprise benchmarks can track. As Lemkin put it, “Modal was, you know, by all intents, it was cheap. Yeah. I mean, 3 billion to 1 billion in 5 months. At the end of the year we end up at 1.4 four billion next year five billion even if the margins never hit the 40s... we make money on this deal like there's just no question.” Stebbings backed up the raw velocity, remarking, “I'm constantly shocked by like the infinite expanding size of the inference market.”

O'Driscoll does not dispute the developer traction, noting that Modal sat “under 100 million like mid late last year” before this surge. His skepticism centers on the suppliers sitting directly beneath them. Neoclouds like CoreWeave borrow billions of dollars to buy silicon, run physical facilities, and service debt. O'Driscoll pointed out that these infrastructure operators will not stay quiet while third-party platforms resell their capacity for higher multiples: “Open weight going back to Jason's comments and CoreWeave all the neoclouds are saying oh my god let me get this straight we do all the building work we buy all the compute we borrow all the capital and then these bozos come in buy from us sell it to developers and make a [ __ ] ton why don't we do that right.”

Who Wins the Inference Margin War

Lemkin is right about short-term developer capture, but O'Driscoll identifies the structural trap waiting in the supply chain. Neoclouds struggle to build good developer tooling because their organizational DNA revolves around data center operations and power procurement. Modal won developer workflows by delivering zero-configuration serverless execution, fast container starts, and clean Python SDKs. Engineers do not want to manage raw clusters on CoreWeave when they can deploy a function on Modal in three lines of code.

Yet developer love cannot protect gross margins forever. Top inference platforms currently pool roughly $5 billion to $6 billion in annual run rate. Compared to OpenAI and Anthropic generating north of $100 billion to $110 billion, independent inference vendors represent less than 5% of frontier model spending. If proprietary foundation models maintain that revenue lead, inference platforms must fight over open-weight workloads where price sensitivity is extreme. When GPU supply expands and bare-metal pricing stabilizes, neoclouds will either build usable developer interfaces or acquire them. If neoclouds offer competitive software wrappers directly on their own iron, pure software inference platforms face immediate price compression.

What to Do With This

Audit your production inference bills across external model APIs and serverless container platforms tomorrow morning. Calculate your effective cost per one million tokens across both environments. If your team pays a software markup on hosted GPUs purely to avoid cluster management, price out reserved instances on a bare-metal provider before your inference volume triples.