The announcement came without fanfare—perhaps deliberately so—on a late May morning in San Francisco. Reactor, a startup most developers had never heard of, declared it had raised $59 million across seed and Series A rounds and was now open for business. The pitch: making real-time generative video actually work.
It's an ambitious claim in a market that's suddenly crowded with ambitious claims. But Reactor's founders, Alberto Taiuti and Bryce Schmidtchen, aren't exactly newcomers to the infrastructure grind. Both spent years on Apple's early Vision Pro team, wrestling with the kind of low-latency challenges that can make or break immersive experiences. Taiuti also co-founded Luma AI, where he served as CTO building the scaffolding beneath that company's 3D and video generation platform. Schmidtchen brought kernel optimization chops from the trenches of AR and VR development—the unglamorous work of making things run fast when milliseconds matter.
What they're building now is less flashy than the AI models dominating headlines, but potentially more essential: a unified platform that handles GPU provisioning, session orchestration, and streaming so developers don't have to. Think of it as AWS for the emerging world of real-time generative video—infrastructure that sits between cutting-edge AI labs and the applications trying to use their models in production.
The target? Sub-50 millisecond frame latency. In practice, that's the difference between a video model that feels responsive and one that feels like molasses.
Following the Money
Lightspeed Venture Partners led both Reactor's seed round and its Series A, a vote of confidence that suggests the firm sees something more than vaporware. Bucky Moore, a partner at Lightspeed, framed the opportunity in characteristically bullish terms: "Real-time systems will define the next era of AI, but the infrastructure gap makes these technologies inaccessible."
Whether that gap is as wide as Lightspeed suggests is up for debate. The firm pointed to recent research advances—DeepMind's Genie 3, StreamDiT—as evidence that world model capabilities are accelerating. Fair enough. But research breakthroughs and production readiness are different animals entirely, something every startup pitching infrastructure eventually learns.
The investor roster reads like a who's who of venture firms betting on AI's next wave: WndrCo, Amplify Partners, Sky9 Capital, FPV Ventures, Abstract Ventures. Jeffrey Katzenberg, the DreamWorks co-founder who now runs WndrCo, is joining as a board observer—a signal that Hollywood money is watching this space closely.
Amazon Web Services also showed up, though not as an investor. Instead, AWS signed on as a strategic partner, providing compute and distribution for real-time generative video workloads and giving the startup access to Amazon's inference economics and custom chips designed for low-latency video workloads. Jason Bennett, AWS's VP and Global Head of Startups and VC, highlighted those purpose-built chips in the announcement, though he stopped short of making any exclusivity commitments.
A Suddenly Busy Category
Reactor's emergence comes amid what can only be described as a feeding frenzy in world model infrastructure. Decart pulled in $300 million earlier in May, while Yann LeCun's Advanced Machine Intelligence—backed by the credibility of a Turing Award winner—secured a staggering $1.03 billion seed round in March. Those numbers suggest investors believe the market for this technology will be enormous. They also suggest competition will be fierce.
Reactor went live with the funding announcement, offering usage-based pricing billed by model type—a fairly standard approach, though the devil will be in the details for developers watching their cloud bills. The company already has at least one notable early partner: Overworld, which is building interactive world models on Reactor's infrastructure. Whether others follow depends on how well the platform delivers on those sub-50ms promises.
The team, at least on paper, has the credentials. Beyond the founders, Reactor has pulled engineers from Apple, Netflix, Meta, Google, Adobe, Replicate, and Microsoft—a roster that suggests the company can write competitive compensation packages and sell a compelling vision. LinkedIn profiles indicate the headcount sits somewhere between 11 and 50 employees, which is either lean and focused or understaffed, depending on how quickly the platform needs to scale.
Real-Time or Really Hard?
For all the venture capital enthusiasm, real-time generative video remains more promise than proven business model. The technical challenges are formidable: maintaining consistent quality while hitting aggressive latency targets, managing costs that can spiral with GPU-intensive workloads, and convincing developers to build on infrastructure that's still proving itself.
Reactor is betting that abstracting away that complexity will unlock a wave of applications that are currently impractical or impossible. Interactive gaming environments that generate on the fly. Video calls with AI-enhanced backgrounds that don't lag. Training simulations that respond in real-time. The use cases sound compelling in a pitch deck. Making them economically viable is another question altogether.
Still, the $59 million buys Reactor time to find out—and in a market moving this quickly, perhaps that's what matters most. Developers curious enough to kick the tires can access the platform at reactor.inc, where the real test begins: does it actually work the way the press release says it does?
The infrastructure is live. Now comes the hard part.
