Businesses are drowning in video they never watch. Store cameras, broadcast archives, production footage - petabytes of it sits unused on servers, becoming what the industry calls dark data. Now two former Google Japan executives think they've cracked the code on making it useful. InfiniMind, their Tokyo-based startup, just raised $5.8 million to build AI infrastructure that converts massive video libraries into searchable, queryable business intelligence - and they're already processing 200-hour video archives for enterprise customers.
The numbers tell the story: by 2025, video already makes up over 80% of all consumer internet traffic, according to industry research. But for enterprises, most of that footage becomes what technologists call dark data - information collected automatically but never analyzed or used. InfiniMind co-founder and CEO Aza Kai saw this problem firsthand during nearly a decade at Google Japan, where he worked across cloud, machine learning, ad systems, and video recommendation models.
"My co-founder, who spent a decade leading brand and data solutions at Google Japan, and I saw this inflection point coming while we were still at Google," Kai told TechCrunch. By 2024, the technology had matured enough that Kai and COO Hiraku Yanagita left to build the solution themselves.
The Tokyo-based startup just closed a $5.8 million seed round led by UTEC, with participation from CX2, Headline Asia, Chiba Dojo, and an AI researcher at a16z Scout. The funding validates InfiniMind's bet that enterprises are ready to unlock decades of archived video - if the technology can actually handle the scale and complexity.
What changed wasn't just falling GPU costs, though annual performance gains of 15-20% over the past decade helped. The real breakthrough came from vision-language models between 2021 and 2023. Earlier video AI could tag objects in individual frames, but it couldn't track narratives, understand causality, or answer complex questions about content. For clients with petabytes of broadcast archives, even basic questions went unanswered.
"Earlier approaches could label objects in individual frames, but they couldn't track narratives, understand causality, or answer complex questions about video content," Kai explained. Until recently, models just couldn't do the job at enterprise scale.
InfiniMind's first commercial test came in Japan with TV Pulse, which launched in April 2025. The AI-powered platform analyzes television content in real time, helping media and retail companies track product exposure, brand presence, customer sentiment, and PR impact. After pilot programs with major broadcasters and agencies, it's already generating revenue from wholesalers and media companies.
But the real product is DeepFrame, InfiniMind's long-form video intelligence platform. It can process 200 hours of footage to pinpoint specific scenes, speakers, or events - the kind of query that would take human analysts weeks to answer manually. Beta launches in March 2026, with full commercial release set for April.
The competitive landscape is fragmented. Companies like TwelveLabs offer general-purpose video understanding APIs for consumers, prosumers, and enterprises. InfiniMind is betting on a different approach: enterprise-specific use cases including monitoring, safety, security, and content analysis, delivered as a no-code solution.
"Our solution requires no code; clients bring their data, and our system processes it, providing actionable insights," Kai said. "We also integrate audio, sound, and speech understanding, not just visuals. Our system can handle unlimited video length, and cost efficiency is a major differentiator."
That cost efficiency matters. Most existing solutions prioritize accuracy or specific use cases but don't solve the economic challenge of processing petabytes of video. InfiniMind's infrastructure is built to scale without breaking enterprise budgets - a pitch that's resonating with early customers sitting on massive archives they can't afford to analyze with current tools.
The company is now relocating its headquarters to the U.S. while maintaining operations in Japan. The Japanese market provided what Kai calls "the perfect testbed" - strong hardware infrastructure, talented engineers, and a supportive startup ecosystem that let the team fine-tune the technology with demanding customers before going global.
The seed funding will accelerate DeepFrame development, expand engineering infrastructure, and fuel hiring as InfiniMind chases customers across Japan and the U.S. Target sectors include media companies with broadcast archives, retail chains with thousands of store cameras, and security operations drowning in surveillance footage.
"This is an exciting space, one of the paths toward AGI," Kai said. "Understanding general video intelligence is about understanding reality. Industrial applications are important, but our ultimate goal is to push the boundaries of technology to better understand reality and help humans make better decisions."
InfiniMind's timing looks sharp. Vision-language models finally work at enterprise scale, GPU economics make petabyte-scale processing viable, and companies are sitting on video archives they can't ignore anymore. The ex-Google pedigree doesn't hurt, but what matters is whether DeepFrame can deliver on the promise of turning dark data into actual business intelligence. With paying customers already using TV Pulse in Japan and a U.S. expansion funded, we'll know soon enough if enterprises are ready to make their video archives searchable. The March beta will be the real test of whether the technology can handle the complexity and scale that's kept this problem unsolved for so long.
