Luma raises $900M, led by Humain
Luma AI has raised a $900 million Series C led by Saudi-backed Humain, which is also partnering with Luma on a 2GW compute supercluster to train and run its multimodal models.
Luma AI announced a $900 million Series C round led by Humain, with significant participation from AMD. Existing investors Andreessen Horowitz, Omniva, Amplify Partners and Matrix Partners also took part. The company did not disclose a valuation.
Alongside the round, Luma and Humain are partnering on Project Halo, a 2GW compute supercluster. Luma says deployment begins in Q1 2026 and should finish by 2028–29. The cluster will handle both training and inference.
The company's case is that general intelligence needs to be multimodal. In its words, "AGI is multimodal and reality is the dataset of AGI." Luma argues that text-only LLMs can't simulate the physical world, and that models should be trained jointly on text, video, audio and images. As evidence, it points to Ray3, which it calls the world's first reasoning video model.
Luma says the next generation of models is meant to support engineering simulation, film production, interactive tutoring and the "foundations of general purpose robot brains." One example it gives is designing systems like rocket engines "in a year instead of a decade." The announcement offers no timelines or technical details for those goals.
Luma now has more than 130 employees. It says adoption is accelerating among Fortune 500 companies in media, entertainment and advertising, through its models and its Dream Machine platform. It plans to spend the money on expanding research and engineering and on growing its teams to deploy its models more widely.
The announcement doesn't give the cost of Project Halo, how much of its capacity Luma will get, or whether AMD's investment is tied to hardware supply.