Rackspace and Together AI announced an initial multiyear inference infrastructure agreement on November 10, 2026. The agreement is binding, but deployment is planned for 2027 subject to conditions. It does not establish general availability of a new research service.
A definitive agreement with conditional deployment
In its announcement, Rackspace says it expects to operate dedicated NVIDIA GB300 NVL72 systems, including power, liquid cooling, networking and storage. Together AI expects to provide the managed platform. GPU financing had not been finalized; facilities, power and equipment availability are also conditions.
The separate sales LOI is nonbinding
The companies also entered a nonbinding letter of intent to explore joint enterprise offerings. Those need further agreement. Data residency, security and compliance features would depend on the specific offering and customer arrangement.
What universities need to establish in a service arrangement
For research offices assessing a future offering, our reading is that the service arrangement needs to specify data flows and operational responsibilities. Dedicated hardware alone does not answer where prompts, outputs and logs will go or who can use them. That assessment belongs alongside the institution’s research data governance review.
Together AI’s existing privacy documentation illustrates why separate questions matter: prompt storage is on by default, while sharing data for model training is a separate opt-in that is off by default. Its organization settings govern requests made with that organization’s API keys. These documented settings are context for evaluating a service, not confirmation of a future Rackspace configuration.
The zero data retention documentation also distinguishes request content from other records. Enabling the option affects subsequent requests; usage metadata remains, and explicitly uploaded fine-tuning or batch files remain stored until deleted. A university should establish which settings, retained records and deletion arrangements apply to its proposed workload.
NIST’s AI RMF Govern playbook recommends connecting AI risk management to data governance, model testing, monitoring and incident response. Applying that approach here means identifying who approves the workload, who reviews model outputs and who handles operational incidents. The NIST framework for research institutions and an AI governance framework template provide further context for assigning those responsibilities. These are institutional review considerations, not compliance findings about the announced infrastructure.
Sources
- Rackspace: strategic agreement with Together AI, November 10, 2026. Checked 2026-11-15.
- Together AI: Privacy and security. Checked 2026-11-15.
- Together AI: Zero data retention. Checked 2026-11-15.
- NIST AI RMF Playbook: Govern. Checked 2026-11-15.







