Luma AI will become a key customer of HUMAIN’s Project Halo, a 2-gigawatt AI supercluster in Saudi Arabia designed to train foundational models at scales far beyond today’s large language models. Existing backers including Andreessen Horowitz, Amplify Partners and Matrix Partners, along with AMD Ventures, participated in the round.
The deal positions Luma for rapid expansion of its flagship Ray3 video-model technology, already deployed in creative studios and Adobe products, into new sectors such as robotics, simulation and global real-time inference. CEO Amit Jain described the partnership as a critical enabler of Luma’s goal to “learn from a quadrillion tokens of information — roughly the collective digital memory of humanity.”
The choice of Saudi Arabia for the compute build is tightly aligned with the kingdom’s AI-industrial strategy. HUMAIN CEO Tareq Amin stated the collaboration “combines capital, compute and capability at the frontier”. The PIF-backed entity seeks to establish a value chain across large-scale data centres, high-performance infrastructure, advanced models and real-world solutions for enterprise and government clients.
Market watchers note that Luma’s valuation now exceeds US$4 billion. The funding round places the company in direct competition with other frontier AI labs working on multimodal general intelligence beyond text. Analysts say the race is intensifying among firms seeking to go from large language models to systems that understand video, audio, imagery and robotics.
Luma developed its Ray3 model and “Dream Machine” platform to generate professional-grade video and imagery. With the new funding and access to Project Halo’s anticipated infrastructure, Luma plans to ramp up training on peta-scale datasets that include video and robot data — rather than only textual tokens. The supercluster is estimated to offer capacity hundreds to thousands of times larger than existing compute dedicated to LLMs.
The deployment of HUMAIN’s infrastructure is also part of broader U. S.–Saudi technology diplomacy. The announcement took place during the U. S.–Saudi Investment Forum in Washington, D. C., coinciding with the visit of Crown Prince Mohammad bin Salman and underscoring Riyadh’s intent to become a global AI power. HUMAIN already has announced partnerships with AMD and Cisco to build out data-centre capacity serving billions of users across the Middle East, Africa and Asia.
However, risks remain. Building “world models” that learn from multimodal data and apply to real-world robotics and simulation is a far more complex challenge than scaling text-based LLMs. Experts caution that technical bottlenecks around data diversity, compute efficiency, control and safety remain significant. Moreover, reliance on large-scale infrastructure raises questions about centralisation, energy consumption and geopolitical dependencies.
Topics
Spotlight