Together AI Heads to Abu Dhabi: Renting a Model or Running Your Own
Zeta42 · 3 October 2026
Mubadala and Together AI will explore AI infrastructure in the UAE, with Together AI setting up in Abu Dhabi. What open models close to home mean for organisations.
What Mubadala and Together AI announced
Mubadala Investment Company, an Abu Dhabi sovereign investor, and Together AI have announced a strategic partnership to explore AI infrastructure and ecosystem opportunities in the UAE. As part of it, Together AI will establish a local presence in Abu Dhabi, bringing its inference and research capabilities closer to regional demand and to developers building AI applications across the UAE and the wider region, the companies said.
Together AI was founded in 2022 and is headquartered in San Francisco. It describes itself as an AI Native Cloud for building and running generative AI with open-source models, offers a platform for inference, model shaping and accelerated compute, and says it powers more than a million developers. The partnership follows Mubadala's $100 million participation in Together AI's Series C financing round, announced on 1 July 2026. The two will work together to identify opportunities for developing and deploying AI in the UAE, in support of the emirate's ambition to build sovereign capabilities across the AI value chain.
What inference is, and why its location matters
Training is how a model is built. Inference is the work of running it: each time someone asks a question or hands over a document, the model runs inference to produce the answer. For an organisation that uses AI rather than builds it, inference is the part it pays for every day, and the part where its data moves.
That makes location a practical question. Every request carries a prompt, and often a file, to wherever the model runs. The closer that machine is, the shorter the trip, and whoever operates it is handling your data for the length of the request. When the announcement talks about bringing inference "closer to regional demand", this is the practical side of it.
Open models and closed models
A closed model is rented. You send requests to its maker's service and get answers back, and the model itself stays under its maker's control. An open model can be downloaded and run by whoever holds it, on hardware of their choosing. Vipul Ved Prakash, Together AI's founder and chief executive, made the case in the announcement:
"Developers want open models because they can own, shape and deploy them however they want, but that only works if the infrastructure underneath is fast, efficient, and actually there when you need it."
He also said: "The demand for open-source AI is outpacing the compute available to serve it." His point is that an open model is only as useful as the hardware it runs on, and that the hardware has not kept up with demand.
Three ways to run a model
For an organisation in the UAE, the options look roughly like this:
- Rent a closed model. No hardware and the quickest start. Every prompt and file leaves your network, and you work within the provider's terms, pricing and model changes.
- Run an open model on a hosted platform. This is Together AI's business. You choose the model and can shape it, while someone else runs the compute. The Abu Dhabi presence is meant to bring that inference closer to the region.
- Run an open model on your own network. The model sits on hardware you control, and prompts and files stay inside the building. You take on the hardware and its upkeep.
In practice, the data usually decides. A team drafting public marketing copy can rent a model without much worry. An HR team working through salary files, or a legal team reviewing contracts, has to ask who else can see the request.
What close to home can mean on your own network
Open models are what make the third route possible, since a closed model generally stays with its maker. This is where Zeta42 works. Its AI infrastructure service runs AI models on-premise for data residency, using desk-side AI computers that hold and run real models on the client's own network, so nothing leaves the building.
That does not make the Mubadala and Together AI news less relevant to organisations that keep their models in-house. More open-model capacity in the region, and the deeper technical expertise the partnership aims to build in the country, help whichever route a business takes. We expect many to mix them: rented models for public work, open models run locally for the documents that should never travel. Close to home can mean Abu Dhabi. For the most sensitive data, it can also mean the next room.
Source: zawya.com