Orange has expanded its collaboration with OpenAI to deploy the latest open-weight reasoning models within its own infrastructure and Africa stands to gain
Orange has expanded its partnership with OpenAI, adding new technical capabilities and deployment options focused on data control, regulatory compliance, and language inclusion across Europe, the Middle East, and Africa. As a result of the wider partnership Orange becomes one of the first global partners to deploy OpenAI’s new open-weight reasoning models, gpt-oss-120b and gpt-oss-20b, across its trusted infrastructure.
These models will be hosted and operated within Orange’s own data centres, meaning it can offer sovereign AI operations tailored to local regulatory landscapes. This latest phase builds on an agreement first announced in November 2024, when Orange revealed it would work with both OpenAI and Meta to fine-tune open-source models to understand regional African languages, beginning with Wolof and Pulaar.
At the heart of the latest deal is Orange’s ability to flexibly deploy OpenAI’s open-weight models across its 26-country footprint, from cloud data centres in France to edge servers and on-premises infrastructure. This naturally allows Orange to customise models for specific use cases, including network management, enterprise solutions, and customer service, while keeping sensitive data within its own infrastructure.
“This collaboration with OpenAI is foundational to our strategy of using state-of-the-art AI models that are both trusted and responsible,” said Orange chief AI officer Steve Jarrett (above). “This strategy drives new use cases to address sensitive enterprise needs, help manage our networks, enable innovating customer care solutions including African regional languages, and much more.”
“These oss-gpt models with mixture-of-experts, reasoning, and agent capabilities allow us to deploy state-of-the-art use cases safely and securely on our own infrastructure, not the public cloud,” he said. “These secure use cases can include our own network data as well as many applications for our B2B customers who want powerful sovereign AI solutions like Orange Business Live Intelligence.”
Open AI chief operating officer Brad Lightcap added: “Our ongoing work with Orange shows how forward-thinking businesses can use our open models to solve real-world problems – from boosting network efficiency to improving African language support, they are delivering meaningful benefits directly to their customers.”
Bet-hedging
Orange is not a one-trick pony when it comes to AI partners. In February 2025, the operator announced a strategic partnership with Paris-based Mistral AI to co-develop AI infrastructure optimised for telecom networks across Europe. The two are assessing how large-scale AI affects network performance and to plan for future infrastructure needs. The collaboration includes integrating Mistral’s models into Orange’s enterprise tools and Mobile Pro services, with a focus on maintaining data sovereignty.
Meanwhile, Google’s Gemini tools also appear within Orange’s internal AI toolkit. Orange has adopted a GenAI toolbox for more than 50,000 employees—branded “Dinootoo”—that includes models from OpenAI (GPT‑x, DALL‑E), Google (Gemini), Anthropic (Claude), Mistral, and Meta (LLaMA). This internal innovation stack enables secure document analysis, chat, code generation, meeting summarisation, and image creation.
African link
Africa has featured prominently in both phases of the Orange–OpenAI partnership. In November, Orange announced it would fine-tune OpenAI’s Whisper and Meta’s Llama 3.1 models to recognise African regional languages currently unsupported by any generative AI systems. The project will begin with Wolof and Pulaar, spoken by more than 20 million people across West Africa, with the longer-term aim of enabling AI to understand and respond in all African languages across Orange’s 18-country footprint in the region.
Crucially, the fine-tuned models will be made available as open-source under a non-commercial licence, allowing governments, NGOs, and public institutions to deploy them in sectors like health, education, and local administration. Orange also intends to partner with local startups and developers to expand and refine these models.
The latest announcement reinforces this direction, confirming that Orange will integrate the latest OpenAI models into its own infrastructure. This deployment will support natural language interaction in local African languages for customer support, marketing, and operational use – building on the linguistic foundation established in the earlier release.
“These open models will also help drive digital inclusion in our African countries where Orange has data centers but where public clouds won’t exist for many years,” said Jarrett.
Responsible deployment
Orange’s AI strategy hinges on what it calls “Responsible AI” – a pragmatic, even frugal approach that deploys the right tool for the right problem. Instead of relying on one-size-fits-all solutions, Orange customises AI models to meet specific operational needs, a principle that extends from customer service to network automation and enterprise AI offerings.
By deploying OpenAI’s open models locally, Orange emphasises that it retains control over data processing and energy consumption, aligning with European data sovereignty principles.
