New Delhi: E2E Networks has launched Blaze, an inference service for open AI models that bills customers only for the tokens they use, the company said on Tuesday.
Inference is the compute a trained model performs when it answers a query. Until now, most Indian startups have had to run that work on servers abroad, which pushed up both cost and response time.
Blaze gives developers a single OpenAI-compatible endpoint and runs on Nvidia GPU infrastructure with DSX OS components, according to the company. It sits on E2E's India-hosted AI factory capacity, which the firm says is being expanded to 14.6 MW for production inference.
The pitch is aimed at a market that is shifting from pilots to live products. Indian AI companies are moving agents into production, and agentic workflows burn through tokens far faster than chat interfaces, making inference cost and latency a product decision rather than a technical footnote.
There is a sovereignty argument in the offer as well. The company says workloads stay on domestic infrastructure, which matters to enterprises that cannot send data abroad.
What Blaze costs, which open models are available at launch and the latency it delivers were not disclosed in the company's announcement.
The launch lands in a crowded race. Indian data centre and cloud firms are building GPU capacity at speed, and cheap inference on open models has become the front line of that competition. Whether pay-per-token pricing holds up as agent traffic scales is the question buyers will test first.
E2E Networks is listed on Indian exchanges. The company did not say how many customers had signed on at launch.