Inception, a burgeoning technology company, has raised $50 million to develop diffusion models aimed at enhancing code and text processing. The round attracted prominent angel investors. Among them were Andrew Ng and Andrej Karpathy, both of whom have since become some of the most recognized names in the tech industry. This investment will help accelerate Inception’s mission to create increasingly efficient artificial intelligence solutions.
Stefano Ermon, another star leader at Inception, made very clear the benefits of their winning competitive strategy in the developing field of artificial intelligence. Since diffusion models are designed to be parallelizable, this offers more flexibility on how to use hardware. This kind of flexibility is proving to be more and more important at a time when demand for AI infrastructure is surpassing supply.
TechCrunch Disrupt in San Francisco is right around the corner —October 13–15, 2026. We can’t wait for you to experience all the incredible innovations in the tech industry – especially Inception’s groundbreaking advancements in diffusion technology! Keeping latency low will be important for their larger models, Ermon noted, and the diffusion approach is going to save on compute costs, too. Latency, or response time of AI applications, is a key performance factor. By reducing latency, we can massively increase the user experience.
“Inception’s models will conserve latency and compute cost using the diffusion approach,” Ermon stated, highlighting how the technology can improve operational efficiency. He further explained that “the qualities of diffusion models become a real advantage when performing operations over large codebases.” These models’ unique ability to compete with large-scale data and surpass existing autoregressive technologies is just a taste of Inception’s capabilities.
Ermon elaborated on the performance metrics of their diffusion-based models, claiming, “We’ve been benchmarked at over 1,000 tokens per second, which is way higher than anything that’s possible using the existing autoregressive technologies.” This critical performance gulf makes Inception a serious contender in the competitive AI workforce.
Ermon asserted that “these diffusion-based LLMs are much faster and much more efficient than what everybody else is building today.” This claim lays bare a change in the AI model development paradigm, where expediency and haste trump all else.