fal's valuation doubled to $16B in seven months

Four of today's items point at the same layer: the machines that run other people's models. The largest of them is a valuation that has moved four times in sixteen months on one business, and the smallest is the first cheque written to ask whether any of the spending worked.

The ladder under one inference host

fal raised $1.3 billion led by existing investor Bessemer Venture Partners, with Nvidia, Sound Ventures and T. Rowe Price taking part, doubling its valuation to $16 billion seven months after it was worth $8 billion, TechCrunch reported. The rest of that climb is already on the record. fal.ai was valued at $1.5 billion in July 2025 on a $125 million Series C led by Meritech (Business Wire), then at $4.5 billion in December 2025 on a $140 million Series D led by Sequoia, the round Nvidia's NVentures joined for the first time (TechCrunch), and was reported in March 2026 to be in talks at around $8 billion (The Information).

That is $1.5 billion to $16 billion in sixteen months, on the same job: hosting inference. At its $49 million Series B the company disclosed more than a million developers, 50-plus enterprise customers and over 100 million inference requests a day (fal). Nothing in today's record updates those operating numbers, which is worth saying plainly: the valuation is the figure that moved.

The same buyer, twice

Applied Compute raised $350 million at a $3.25 billion valuation in a round led by the angel investor Elad Gil, with Nvidia, Shopify Ventures and HubSpot Ventures taking part (SiliconANGLE). Nvidia is therefore on both of today's inference cap tables, and on fal's Series D before that.

The cost side of the same layer moved too. Together AI launched Together Compiler, which uses teams of AI agents to rewrite the GPU code behind the models it serves (Business Wire). A host whose margin is the gap between what the GPUs cost and what the tokens sell for has an obvious reason to spend engineering on that gap.

The first bill audit

Larridin raised a $30 million Series A led by Pelion Ventures to measure what companies are getting back from AI spending that has climbed into their top ten line items (Business Insider). It is the smallest round in today's investments digest and the one that describes the others: the spending is large enough to show up in the accounts, and the measurement is late enough to be a startup.

Bigger funds, smaller firms

Sequoia Capital said it has raised about $10 billion in new funds while closing its London office and shutting both Scouts and Arc, the batch accelerator that gave admitted founders $500,000 to $1 million up front (TechCrunch). Sequoia was reported in August 2026 to be raising about that amount, which would be the largest commitment in its 54-year history, months after closing a $7 billion expansion fund (TechFundingNews, and TechCrunch on the expansion fund). The raise has landed; the firm paying for it is smaller than it was.

Creandum closed a $750 million eighth fund, its largest, to keep backing early-stage companies as AI seed rounds get bigger (Bloomberg). Creandum closed its seventh at €500 million in 2024, oversubscribed, with more than half the capital from US investors (Tech Funding News).

Put the four together and the shape is a single layer absorbing capital faster than anyone has measured what it returns: $1.3 billion and $350 million into hosting inference on one day, agents rewriting GPU code to make that hosting cheaper, and $30 million for the company that proposes to count it.