Modal closes $750M at $15.75B as sandbox revenue hits $300M
The infrastructure layer answered the market's doubts about AI adoption with a number, and the hardware layer answered with a sell-off. Also: Anduril's employees just got a $90 billion price tag.
Modal Labs closed a $750 million Series D at a $15.75 billion post-money valuation, and said in the same announcement that it has crossed $1 billion in annualized revenue. Accel led the round; Kleiner Perkins joined along with every existing investor, including Menlo Ventures, General Catalyst, Redpoint, Lux Capital, Amplify Partners and Bain Capital Ventures. The company says usage has tripled since its May Series C — a $355 million round at a $4.65 billion valuation — and that revenue is up more than 15-fold year over year. The $15.75 billion figure lands at the top of what was reported in late September, when the round was still in talks; we covered that stage in Modal and Baseten in talks at $15B and $26B.
The detail worth more than the valuation is Sandboxes. Modal launched the product in 2023 for LLM-generated code that has to run somewhere isolated, and it now stands at a $300 million run rate on its own — roughly a third of the company's revenue. Modal says it has spun up close to three billion sandboxes and that they scale to a million concurrent. Getting there meant throwing out most of the old stack: Modal built its own file system, container runtime, scheduler and networking, on a capacity pool that now spans more than 20 clouds. The customer list reads as a map of the agent economy — background agents at Cognition, Ramp, Linear and Hebbia, robot training at Physical Intelligence and 1X, video generation at Runway, drug discovery at Chai Discovery.
Our read: private infrastructure companies rarely publish revenue, and publishing a run rate alongside a raise is a deliberate signal in a month when the market started questioning whether anyone is actually paying for AI. Note which product carries the growth. Sandboxes exist because models write code now, so the demand that tripled here is agents executing work, not models being trained.
Dell fell 11% Wednesday, dragging HPE, Oracle and Super Micro with it, after Barclays downgraded Dell, HPE and NetApp to the equivalent of hold from buy "on enterprise AI timing." Oracle dropped 9% and is now down 33% for the year; CoreWeave fell 15%, wiping out its gains for 2026. The analysts wrote that they have not seen a "meaningful AI inflection" in enterprise spending, and attributed the strong enterprise revenue the market has been rewarding to the global memory shortage instead — infrastructure vendors have largely passed higher memory prices on to customers. The supporting numbers are blunt: businesses are allocating about 5% of their IT budgets to AI, and only about a quarter of enterprise AI spend goes to hardware.
Our read: this is the counterweight to the story above, and both can be true. Dell is up more than 300% this year and HPE about 160%, so a downgrade on timing is a repricing of expectations rather than a report of collapse. But if a meaningful slice of enterprise server growth is memory inflation, then normalizing memory prices cuts both ways on the way down — the same asymmetry we flagged when Broadcom's CEO said the slowdown call changed nothing about AI demand.
Anduril is letting employees sell up to $1.3 billion of stock at $94.80 a share, in a tender offer that values the defense technology company at roughly $90 billion. The offer opened October 13 and closes November 9, and for the first time it reaches current and former employees across 16 countries — earlier tenders were limited to the United States, Britain and Australia. Bloomberg reported the terms first and Anduril confirmed the valuation to CNBC. It is a steep step up from May, when a $5 billion round led by Thrive Capital and Andreessen Horowitz doubled the company's valuation to $61 billion; Anduril reported $2.2 billion in revenue last year, and signed a Navy shipyard contract worth up to $2.9 billion this month.
Our read: tender offers have quietly replaced the IPO for late-stage private companies, and the price is the signal — $90 billion puts a defense contractor in the same conversation as the frontier labs, which is a claim about where autonomy software is valued, not just hardware.
What to watch: whether the memory-cost explanation for server revenue survives the next two quarters of earnings, and whether Modal's next disclosure separates inference revenue from sandbox revenue.
If the enterprise AI refresh turns out to be a memory-price story, does that change what you're buying this quarter? Tell us in the comments.
