Together AI Cloud Regions, September 2026: IBM, Equinix, and Sovereign Buildouts in India, Saudi Arabia, and the UAE
Together AI is simultaneously deploying capacity across IBM Cloud, Equinix's global colocation fabric, and sovereign-backed facilities in Chennai, Saudi Arabia, and the UAE — a buildout financed by the $800 million Series C that valued the company at $8.3 billion.
In September 2026, Together AI's regional buildout reached a visible inflection point when Equinix and NVIDIA joined the company to launch the Inference Exchange — a service that routes enterprise customers to Together AI's catalog of more than 200 open-source models through Equinix's interconnected colocation network, offering secure, low-latency NVIDIA compute. The launch is the most recent addition to an infrastructure footprint expanding rapidly since July 2026, when Together AI closed an $800 million Series C at an $8.3 billion valuation, led by Aramco Ventures and co-invested by NVIDIA and Vista Equity Partners, with annual order commitments reportedly exceeding $1.15 billion at close.
That financing has since underpinned a series of concrete deployments. In August 2026, Together AI and IBM formalized a $240 million multi-year agreement to deploy a large-scale NVIDIA HGX B300 inference cluster on IBM Cloud, reportedly targeting completion in early 2027, with Together AI reportedly expecting the capacity to sell out before it goes live. At roughly the same time, the company was reported to be actively looking abroad for additional data-crunching capacity — a signal that domestic resources were running short of demand. Aramco Ventures' lead position in the Series C connects Together AI directly to Saudi Arabia's sovereign capital base and foreshadowed the Gulf-region partnerships that followed.
The sovereign and emerging-market dimension of that expansion is substantial. In Chennai, Larsen & Toubro secured what it described as a "mega" order to build India's largest NVIDIA B300 AI factory on behalf of Together AI — a unified cluster of more than 10,000 GPUs — with Indian media reporting the contract value at approximately Rs 10,000 crore, or roughly $1.2 billion. In Saudi Arabia, Together AI and HUMAIN announced a strategic partnership to accelerate AI infrastructure and cloud services in the Kingdom; reports place AMD accelerators, Cisco networking equipment, and Together AI's software stack at the core of HUMAIN's national AI backbone. In the UAE, Abu Dhabi's Mubadala announced a tie-up with Together AI to pursue AI infrastructure investment opportunities in the country. Taken together, the three relationships in the Gulf and South Asia represent a significant geographic diversification for a US-founded neocloud.
The capacity push is paired with platform moves designed to deepen customer retention. In September 2026, Together AI expanded its fine-tuning platform and cut training prices, putting it in more direct competition with proprietary model vendors. The company also introduced discounted preemptible compute instances for price-sensitive, interruptible workloads, and continued optimizing its ThunderKittens HPC tensor library for NVIDIA's next-generation Vera Rubin GPU architecture — positioning the stack for the next accelerator cycle even as current capacity sells ahead of deployment.
The picture that emerges is of a company transforming rapidly from a US-centric neocloud into a multi-continent GPU infrastructure provider, but doing so against a backdrop of material execution risk. Running major construction projects in India, Saudi Arabia, and the UAE simultaneously — while managing the IBM Cloud deployment and the Equinix partnership — places significant demands on organizational bandwidth and exposes Together AI to political, regulatory, and construction timelines it does not fully control. Sovereign investment brings strategic reach but also dependency: Aramco Ventures' lead position means Together AI's Gulf market access runs partly through a state-linked channel, introducing complexity if geopolitical conditions shift. Three signals will clarify how reliably this expansion converts ambition into operational reality: whether the IBM B300 cluster delivers on its reportedly early-2027 timeline; whether the Chennai factory brings its full 10,000-GPU capacity online on schedule; and how quickly the Equinix Inference Exchange draws enterprise adoption outside the United States, which will test whether distributed colocation translates geographic presence into revenue.