Thomas Fuller | SOPA Images | Light Rocket | Getty Images
Video generation startup Luma AI announced it has raised $900 million in a new funding round led by Humain, an artificial intelligence company owned by Saudi Arabia’s Public Investment Fund.
To raise funds, Advanced Micro Devices, Inc. The venture arm and existing investors Andreessen Horowitz, Amplify Partners and Matrix Partners were announced at the US-Saudi Investment Forum on Wednesday.
CNBC confirmed that the company’s valuation is now over $4 billion.
Luma develops a multimodal “world model” that can learn from text as well as video, audio, and images to simulate reality. CEO Amit Jain told CNBC in an interview that these models will scale beyond large language models trained solely on text, making them even more effective and “useful in the real, physical world.”
“With this funding, we plan to expand and accelerate our efforts to train and deploy these global models today,” Jain said.
In September, Luma released Ray3, the first inferential video model that can interpret prompts and create videos, images, and audio. Jain said Ray3’s benchmarks are currently higher than OpenAI’s Sora 2, and about the same level. Google Veo 3.
Launched in May, Humain aims to provide full-stack AI capabilities and strengthen Saudi Arabia’s position as a global AI hub. The company is led by industry veteran Tarek Amin, who previously ran Aramco Digital and before that was CEO of Rakuten Mobile.
The company also announced a number of partnerships at the forum, including the construction of another data center. AMD and Cisco, Global AI (using Nvidia infrastructure) and Elon Musk’s xAI.
Luma and Humain will also partner to build a 2-gigawatt AI supercluster called Project Halo in Saudi Arabia. The expansion will be one of the world’s largest graphics processing unit (GPU) deployments, Jain said.
Big tech companies have invested in supercomputers around the world to train large-scale AI models. In July, Meta announced plans to build a 1-gigawatt supercluster called Prometheus, and Microsoft Nvidia GB300 NVL72 platform in October.
“Our investment in Luma AI, combined with HUMAIN’s 2GW supercluster, will enable the training, deployment, and scaling of multimodal intelligence at the frontier level,” Amin said in the release. “This partnership sets a new benchmark for how capital, computing and power are integrated.”
The collaboration also includes Humain Create, an effort to create sovereign AI models trained on Arabic and regional data. Jain said Luma’s model and features will be rolled out to businesses in the Middle East, as well as focusing on building the world’s first Arabic video model.
He added that because most models are trained by scraping data from the internet, AI-generated content often doesn’t translate well in countries outside of the U.S. and Asia.
“It’s really important to incorporate these cultures, their identities, visual, behavioral and all other expressions into our models,” Jain said.
AI-generated content tools have received significant backlash from entertainment studios over the past year, citing copyright concerns. Luma’s flagship text-to-video platform, Dream Machine, drew some criticism for copying its IP earlier this year, but the company has installed safeguards to prevent unwanted use.
“Even if we really try to deceive, we always improve it,” he said. “We’ve built a very robust system that actually uses the models we trained to detect them.”
WATCH: Humain CEO talks about building ChatGPT’s Arabic rival
