Access to large, dedicated AI compute materially changes the cost and locality calculus for frontier-model training, enabling non-North-American R&D hubs to host long-running experiments and scale pretraining. Per reporting by Reuters, Zawya/Refinitiv and Arab News, HUMAIN, a PIF-backed company, and Cohere announced a collaboration in which HUMAIN will designate at least 50 megawatts (MW) of dedicated AI compute capacity to support Cohere's next-generation foundation models. The partnership, announced during a diplomatic visit reported by Reuters and Zawya, is expected to have the deployment live by Q4 2027 and include the ability to scale over the next five years, according to Reuters and Arab News. Reporting by Zawya and Arab News adds that the deal will also cover enterprise AI solutions and sovereign, Arabic-language and domain-adapted foundation models, and Zawya describes the commitment as Cohere's first major international compute deployment outside North America.
Meta releases Muse Spark 1.1, claiming it surpasses OpenAI and Google models