Cassava Technologies has announced that it will offer Nvidia’s Nemotron open models as a managed service on its AI Factory, the company’s Nvidia-powered computing platform.
The models will be available through the Cassava AI Multi-Model Exchange (CAIMEx), where businesses and service providers can access them as APIs rather than setting up and running their own model infrastructure.
Cassava says Nemotron is built for reasoning and for using software tools, which are the abilities AI agents need to work through multi-step tasks and interact with company systems. Nvidia has also released the model’s weights, training data, and methods, so developers can adapt it for specific uses.
Two other additions were announced alongside the models. The first is the CAIMEx Token Optimiser, which Cassava says identifies content a model has already processed and reuses it where appropriate. It applies to both Cassava-hosted and third-party models, though Cassava has not published figures on the savings.
The other is Nvidia OpenShell, which Cassava is using to combine computing, controlled agent activity, and independent security enforcement, a response to agents gaining access to business systems.
“Expanding our AI Factory across our markets is about creating the capacity and economics needed for businesses to build and deploy AI solutions at scale.” Group Chief Technology & AI Officer Ahmed El Beheiry, Cassava Technologies.
What It Means for Kenya
Cassava first announced its AI Factory in March 2025, with data centers in South Africa, Egypt, Kenya, Morocco, and Nigeria in line for upgrades with Nvidia hardware. The stated aim was to keep data within Africa’s borders.
The company later launched GPU rental services and, in November 2025, agreed to provide computing capacity to Rockefeller Foundation grantees across eight countries, including Kenya.
The practical question for Kenyan organizations is where AI work gets done. Many firms using AI tools today rely on foreign cloud providers, and sending customer data abroad can complicate compliance with the Data Protection Act, 2019.
Locally hosted services are one way to address that, along with cost and speed.
