Modal Labs Tripled Its Valuation to $15.75B in Four Months
Modal Labs Tripled Its Valuation to $15.75B in Four Months
Four months ago, Modal Labs was worth $4.65 billion. A fresh round is now set to price it at $15.75 billion — fourteen times what it fetched in September 2025. Around 150 people work there; annualized revenue is north of $300 million; and outside AI engineering circles, almost nobody has heard of it. There is no cleaner snapshot of the mood in venture capital right now.
Modal rents out computing power for running AI models. The New York startup is closing in on a $750 million round led by Accel that would value it at $15.75 billion post-money, a person familiar with the deal told TechCrunch this week. Modal wouldn't comment — though in this market, nobody turns down three-quarters of a billion dollars without a word, and silence has its own grammar. Axios and Bloomberg had already confirmed the fundraising was underway; the size of the round itself was new.
Sit with that number for a second. $750 million — for a company that raised $355 million in May at a $4.65 billion valuation, itself a jump from a $1.1 billion Series B in September 2025. In thirteen months, the price of this company has gone up roughly fourteen times. The revenue, by the company's own telling, is over $300 million annualized — real money, and also roughly one-fiftieth of the new valuation.
What Modal actually does
Modal makes its living from what the industry calls inference — running an AI model that has already been trained so it can generate answers, text, code or images. Developers use Modal's platform to train models and run other compute-heavy workloads without managing their own servers. The pitch, from the beginning, has been radical simplicity for developers drowning in infrastructure complexity.
The company's founder is the reason it exists in its current form. CEO Erik Bernhardsson spent years building data teams at Spotify — he built Annoy, the open-source nearest-neighbor library, and Luigi, the workflow engine — and later served as CTO at Better.com. CTO Akshat Bubna studied at MIT and was an early engineer at Scale AI. In 2021, Bernhardsson wrote a blog post called "Software infrastructure 2.0: a wishlist" that reads, in retrospect, like Modal's founding document: "The word cluster is an anachronism to an end-user in the cloud! I'm already running things in the cloud where there's elastic resources available at any time. Why do I have to think about the underlying pool of resources? Just maintain it for me."
That frustration became the product. On a dev-tools podcast last year, host Jack Bridger laid out the pain — learning EC2, paying $300 to $400 a month for GPUs that sit idle — and Bernhardsson, laughing, replied: "I agree, but I'm biased."
The inference bet turned out to be the right one. As Bernhardsson told The New Stack, Modal "found a strong market fit with AI inference applications" almost by accident — customers started showing up with generative-AI workloads, and the company's architecture, built for spinning containers up and down in seconds, turned out to be exactly what inference needed. "The big challenge is just, like, scale and stability and performance," he explained. "When you're running thousands and thousands of GPUs at scale and handling 10,000 requests a second, obviously you're building a very different product."
The customer list tells the story: Cognition, the coding startup; Suno, the music generator; Ramp, the fintech; Substack, the publishing platform. All of them need enormous, spiky compute. None of them wants to run a data center.
Why investors are paying 50 times revenue
Do the math that matters to Accel. At $15.75 billion on a bit over $300 million of annualized revenue, they're paying more than 50 times revenue for a company whose core input — GPU compute — is a commodity it doesn't own. That multiple only makes sense if you believe two things: that inference demand is going to explode for years, and that Modal will capture enough of it at good enough margins to justify the price.
The first belief is the easy one. Demand for inference has surged as businesses deploy open-source models and need outside infrastructure to handle heavy computing loads. Modal is not alone in the trade. Bloomberg has reported that Baseten, another inference startup, is nearing a new round at a $26 billion valuation — roughly double what it was worth in June. The Information has reported that Fireworks and Fal, which provides inference infrastructure for video and image generation, have also been talking to investors about new rounds at sharply higher valuations. Fireworks said in July that its annualized revenue had hit $1 billion, a fivefold increase in a year. One source told reporters that several inference startups could reach that milestone by year end.
It's a land grab, and the VCs are acting like the land is running out. AI coding startup Factory tripled its valuation to $5 billion on a $200 million raise. CoreWeave, the neocloud that rents out GPU clusters, has become the sector's bellwether. When the entire category is repricing every four months, waiting for a "reasonable" entry point means never getting in.
The ugly math nobody puts in the press release
Here's the part of the story the funding announcements skip. Inference companies operate on thin margins, because the cost of acquiring or leasing compute is brutally high and largely outside their control. Revenue can triple while profits barely move. The sector's dirty secret is that much of the "growth" is pass-through: Modal rents GPUs from cloud providers, marks up the convenience, and hopes the spread covers the operation.
We checked the timeline ourselves, and it's worth stating plainly: Modal went from $1.1 billion (September 2025) to $4.65 billion (May 2026) to $15.75 billion (this week). That is not a valuation curve; that is a vertical line. The company's revenue grew too, but nothing like that fast. Somewhere in the gap between those two curves sits either the defining infrastructure business of the AI era or the defining overpay of the cycle, and there is genuinely no way to know which one from the outside.
There was also a July incident that should have given investors pause, and apparently didn't. A customer's misconfigured, unauthenticated endpoint allowed external code execution, and customer data was accessed. Bubna said the issue stemmed from the customer's own publishing mistake and emphasized that Modal's platform itself was not compromised. Fair enough — but for a company selling developers the promise "you don't have to think about infrastructure," a security incident rooted in exactly that abstraction is, at minimum, an awkward look.
Modal Labs $750M: the questions everyone asks
What are the terms of Modal's new round? $750 million led by Accel, valuing Modal at $15.75 billion post-money — up from $4.65 billion in May (on a $355 million raise) and about $1.1 billion at its Series B in September 2025, roughly thirteen months earlier.
What is serverless AI inference? Modal rents out computing power for running already-trained AI models — inference — plus training and other compute-heavy workloads, without customers managing their own servers; its platform spins containers up and down in seconds, which turned out to be exactly what generative-AI workloads needed.
How does the ~50x revenue multiple work? At $15.75 billion on a bit over $300 million of annualized revenue, Accel is paying more than 50 times revenue — a bet that inference demand explodes for years and that Modal captures it at good enough margins, even though its core input, GPU compute, is a commodity it doesn't own.
What was the July incident? A customer's misconfigured, unauthenticated endpoint allowed external code execution, and customer data was accessed. CTO Akshat Bubna said the issue stemmed from the customer's own publishing mistake and emphasized that Modal's platform itself was not compromised.
Our read: this is the trade of the cycle, and it might work
Let's give the bull case its due, because it's stronger than the skeptics admit. Inference is where this whole AI wave either works or breaks. Training gets the glory; inference is the part that actually bills — tokens burned, invoices sent, revenue recurring. Deploying an open-weight model means someone has to run it. The hyperscalers will take the whales; the long tail — thousands of companies — needs a Modal, a Baseten, or a Fireworks. The market is real, it's growing obscenely fast, and the winners will be enormous.
But $15.75 billion is a colossal number for a 150-person company renting out other people's GPUs, and the history of infrastructure land grabs says the margins compress exactly when the growth looks best. Compute gets cheaper, competition gets fiercer, and the 50-times-revenue multiple has to come down — either because the revenue catches up, which requires flawless execution for years, or because the valuation doesn't, which is the polite way of describing a bubble.
Accel is betting it's the former. The May investors — Redpoint, General Catalyst — are sitting on a 3x markup in four months and feeling very smart. And Erik Bernhardsson, the Spotify engineer who just wanted someone to maintain the cluster for him, is about to run a company worth more than most airlines. Whether that sentence still sounds reasonable in eighteen months will tell you everything about where this cycle ends.
Sources: TechCrunch's September 30 reporting on the $750M Accel-led round (sourced to a person familiar with the deal); Axios and Bloomberg on the fundraising; Reuters' May 2026 reporting on the $355M round at $4.65B and $300M+ annualized revenue; The New Stack's interview with Erik Bernhardsson; the Latent Space and Scaling DevTools podcast transcripts; The Information via trade-press coverage on Fireworks and Fal.
Related on TechPulse USA: AI coding startup Factory tripled its valuation to $5 billion on a $200 million raise — the funding-mania playbook, same quarter; CoreWeave Forge wants to own the AI loop — the neocloud the whole sector is priced against; and Nvidia's $668M bet on Taiwan's GMI Cloud — the chip giant funding the same inference buildout.
By Daniel Rao, TechPulse USA — covering the companies, models, and money moves shaping artificial intelligence. We read the fine print so you don't have to.
Comments
Post a Comment