Rackspace and Together AI announced on Tuesday (10) a multiyear strategic AI infrastructure agreement under which Rackspace will provide and operate dedicated NVIDIA GB300 NVL72 infrastructure to support Together AI's inference workloads. Deployment should happen in 2027, if the conditions set out in the agreement are met.
What the agreement covers
The system will use NVIDIA Spectrum-X Ethernet networking. Rackspace will operate it in a purpose-built, liquid-cooled facility, including power, cooling, networking and storage. According to the announcement, Together AI will provide the managed cloud platform running on that infrastructure, with access to open-source models, inference optimizations and developer tools.
The GB300 NVL72 is NVIDIA's rack-scale system based on the Blackwell Ultra architecture: it combines 72 Blackwell Ultra GPUs and 36 Grace CPUs in a single NVLink domain, designed for large-scale inference. It is the same hardware AI cloud providers have been adopting through 2026.
Letter of intent and financing
Beyond the infrastructure agreement, the companies signed a non-binding letter of intent to explore a joint go-to-market approach for enterprises, combining Together AI's inference, post-training and reinforcement learning with Rackspace's infrastructure and engineering services. Any joint offerings will depend on further agreements and on each customer's requirements.
The announcement also involves financing: on November 6, Rackspace entered into a share purchase agreement with an investor group led by Voss Capital for a $66 million private placement. The company says it will seek separate financing for the purchase of the GB300 NVL72 systems and that there is no definitive financing commitment for the GPUs yet — in other words, deployment depends, among other factors, on that funding.
Context
“Enterprise AI requires more than access to compute. Success depends on bringing compute into production, operating it reliably and delivering sustained value,” said Gajen Kandiah, CEO of Rackspace. Vipul Ved Prakash, founder and CEO of Together AI, said scaling in the enterprise requires infrastructure that can grow with the company's ambitions while delivering performance and reliability. NVIDIA, through Raj Mirpuri, Vice President of Global AI Clouds and Infrastructure Ecosystem, highlighted that the Blackwell Ultra architecture can run inference workloads efficiently.
Founded in 2022, Together AI says it serves more than two million developers with production-scale inference, post-training and reinforcement learning. Geekiki previously covered the company on November 4, when it launched Together Compiler to optimize inference for open models.
Image credit: NVIDIA — diagram of the NVIDIA GB300 NVL72 rack, from the official NVL72 AI Factory reference architecture documentation.
Sources: official announcement (GlobeNewswire, via Morningstar); NVIDIA GB300 NVL72 documentation.
Transparency: this article was produced with the assistance of artificial intelligence, cross-checked against public posts and internet sources, and reviewed against the original sources cited. Please consult the original sources to confirm the information.
By Muse
