While OpenAI retreated from video generation, LTX just built a business model that solves the actual problem: nobody can afford to iterate.
The Summary
- LTX, a Lightricks spinoff, launched open-source AI video software that runs on a single laptop, free until developers hit $10M in revenue
- The pricing model attacks the real cost barrier: generating one minute of video costs $3-10+ at xAI, Google, and Runway, and users need dozens of attempts to get usable output
- This is revenue-share capitalism meeting the agent economy: make money only when your customers make money
The Signal
OpenAI shutting down Sora wasn't a vote of no confidence in video generation. It was a referendum on the economics. When you charge per generation and the models aren't precise enough to nail it on the first try, you've built a slot machine, not a tool. CEO Zeev Farbman gets it: the core problem isn't capability, it's iteration cost.
The math is brutal. If you're building an agent that needs to generate product videos, and each attempt costs $5-10, and you need 20 iterations to get something usable, you just burned $100-200 per final output. Scale that to hundreds of products or personalized variants, and suddenly your AI agent's operating costs look like a Series A burn rate.
LTX's move to open-source with deferred revenue-sharing flips this. Developers can iterate for free during the expensive R&D phase. When they cross $10M in revenue, meaning they've built something that actually works at scale, LTX takes a cut. This is the pricing model that makes autonomous agents economically viable. An agent that generates marketing videos doesn't need to justify its costs in month one. It needs to prove unit economics over time.
The deeper play: running on a single laptop means no cloud vendor lock-in, no API rate limits, no surprise bills. For companies building agents that need to generate hundreds or thousands of videos, that's the difference between a feature and a business model.
The Implication
Watch for more AI infrastructure companies to adopt revenue-sharing models. The pattern is clear: if you want developers to build agents that actually ship, you can't charge them per API call during the messy middle. LTX is betting that helping 100 companies get to $10M is worth more than nickel-and-diming 10,000 developers who never make it past the prototype. If they're right, this becomes the standard for how AI infrastructure gets priced in the agent economy.
Source: The Information