In 2006, a PhD student at Carnegie Mellon named Henry Kang built robots that recorded their surroundings, identified the objects in them, and remembered where to look later. Twenty years on, he has raised $15 million to do the same favor for people — and for the AI agents now rummaging around on their behalf.
Clipto, the three-year-old startup Kang founded in 2023, announced on Aug. 31 that it had closed an all-equity round at a $250 million post-money valuation. HSG — the firm formerly known as Sequoia China — led, joined by GL Ventures, EnvisionX Capital, Palm Drive Capital, 522 Ventures, and angel investors Hans Tung and Lu Zhang.
The size is unremarkable by 2026 standards. What is not: Clipto says it crossed $15 million in annual recurring revenue at the start of this year and is profitable on a net-income basis. It raised, in other words, roughly one year of revenue it was not obviously short of — a posture almost nobody else in applied AI can strike.
What Clipto Actually Does
Clipto indexes the media a person already owns — videos, audio, images, meeting recordings and documents sitting on their machine — and builds a semantic index across them, so the whole archive can be queried in plain language: describe the moment, get the timecode. Processing runs locally on the device, Kang says, with no requirement to ship files to a cloud service first.
“The real insight is that in this AI era, we don’t have a content shortage,” Kang told TechCrunch. “The opposite is true. We have too much content. We have too much video footage sitting on our computers that isn’t being used.”
The company built the first version for video creators drowning in footage scattered across drives. That is no longer the bulk of the base. Creators now account for only about one-quarter to one-third of users, Kang said; the rest are lawyers, doctors, researchers, marketers, HR staff, professors and students. Clipto says more than 30 million people have used its products since launch, with hundreds of thousands paying. Kang declined to give a precise subscriber count or average revenue per customer.
The headcount is the number that should stop you: just over 20 employees across the San Francisco Bay Area, Hong Kong and Singapore. Roughly $750,000 of ARR per person.
The Agent Hook
About two weeks before the raise, Clipto shipped support for Model Context Protocol, the standard that lets AI applications reach outside data sources. Alongside the funding it launched native Windows and Android apps, extending a footprint that had been web, Mac and iOS.
MCP is the strategic center of the round. It means a user can point ChatGPT or Claude at their own private archive and have the agent search it — with, Kang says, access gated on an explicit user request and authorization, and retrieval confined to whatever scope the user grants.
“Generative AI has transformed how content is created, but it still struggles to understand the enormous amount of personal media people already have,” Kang said in the funding announcement. “We believe AI needs a memory layer, not just a reasoning layer.”
Why This Matters
Foundation models are extraordinary reasoners with no idea what is on your hard drive. Closing that gap is the connective-tissue problem of this phase of the industry, and there are two plausible architectures for it. One is that platform owners index everything in their own clouds — Apple Photos, Google’s Personal Intelligence in Gemini, Adobe’s media intelligence in Premiere. The other is a neutral layer on the user’s device that rents access to whichever model the user prefers. Clipto is a pure bet on the second.
That bet is what the investors are underwriting. “We believe AI’s next frontier isn’t just generating more content, it’s making sense of the enormous amount of content people already have,” said Holly Zheng, Founding Partner at EnvisionX Capital, in the announcement. “Clipto is tackling one of the hardest problems in AI: turning years of unstructured video into AI-readable memory while keeping data private and on-device.”
The profitability is the other signal, and arguably the louder one. The dominant AI business model of the past three years has been to raise enormous sums against a compute bill and defer unit economics. Clipto inverted it: it charged for a discrete piece of utility software, kept the team small, ran inference on hardware it does not pay for, and reached positive net income before taking growth capital. A $250 million valuation on $15 million of ARR is roughly 17 times revenue — expensive for software, cheap for AI, and priced as though the buyer believes the revenue is durable rather than promotional.
The obvious counterargument is the one TechCrunch put plainly: it is not yet clear whether AI media search supports a stand-alone category or simply becomes a checkbox inside products people already use. Clipto’s answer is breadth plus neutrality — it searches across video, audio, images and documents regardless of which vendor’s service they live in, and hands the result to any agent. Whether that is a moat or a head start is the whole question.
What to Watch
Three things. First, whether OpenAI and Anthropic ship first-party local file connectors of their own; MCP is an open standard, which cuts both ways, and Clipto’s defensibility rests on index quality and on-device model efficiency rather than on the connector. Second, conversion: 30 million users against hundreds of thousands of subscribers is a rate under one percent, and Windows and Android now have to do better than Mac and iOS did. Third, the enterprise turn — the new capital is earmarked for on-device models, compute, more agent integrations and, per FinSMEs, developer API onboarding for media and entertainment platforms. Consumer profitability funding an enterprise motion is a good story. It is also where a lot of good consumer companies have gotten lost.
“The real insight is that in this AI era, we don't have a content shortage. The opposite is true. We have too much content.”— Henry Kang, Founder and CEO, Clipto