Artificial intelligence agent startup NinjaTech AI Inc. today launched Ninja Enterprise to let large companies run what the startup calls AI employees inside their own cloud environments for a fixed yearly fee.

The graphics processing units those agents need are part of the deal. Ninja Enterprise is aimed at the unpredictable spending that NinjaTech says tends to stall corporate AI projects after the pilot stage. Token-based pricing is hard to forecast. Long-running agents make that worse, because a per-token bill keeps climbing for every hour one of them spends on a task.

“Most vendors keep AI on a usage meter,” Chief Executive Babak Pahlavan said. With a fixed bill, he contends, AI can spread through an organization once the trial phase is over.

For customers running Ninja Enterprise in their own cloud, the fixed price also covers the computing power behind the software, which NinjaTech said it has never supplied before. Buyers don’t have to line up a separate vendor and contract for it, as most enterprise AI projects require.

NinjaTech sources the GPU and inference capacity itself, through Microsoft Corp. working with Fireworks AI Inc. or through Amazon Web Services Inc. Capacity is reserved for each customer on a single-tenant basis, so there are no servers to buy.

The company claims open-weight models running on that capacity cost about a 10th as much, end to end, as comparable deployments on frontier models from the big AI labs. Customers can still move any workflow onto a model from Anthropic PBC or OpenAI Group PBC when that suits the job better.

The cloud setup doesn’t suit businesses whose most sensitive work can’t touch an external network at all. For those customers, NinjaTech licenses a second version that runs inside their own air-gapped environment on hardware they supply. In either setup the company says customer data is never pooled or used to train an outside model.

The AI employees themselves are built to take a goal and work through it unattended at any hour. Staff talk to the agents in Slack and Microsoft Teams. Packages are sized for 100, 500 or 1,000 AI employees, and customers can add more on demand.

Implementation comes under the same contract as the platform, delivered for healthcare customers by integration specialist Optimum Healthcare IT LLC. Everyone else works with Infosys Ltd. The Indian information technology services company invested in NinjaTech through its Infosys Innovation Fund in August, when it also agreed to act as the startup’s exclusive professional services partner for at least a year.

NinjaTech is a venture capital-backed startup that Pahlavan co-founded in 2022 after 11 years at Google LLC. SRI Ventures and DCVC were early backers.

Ninja Enterprise is available now. Buyers can start with a fixed-scope pilot before signing an annual capacity contract.

Image: NinjaTech AI