E2E Networks Blaze: pay-per-token AI inference service launched in India

E2E Networks, the listed Indian cloud computing company, launched a new service called Blaze on Monday, 2 November 2026. Blaze lets developers and businesses run open-weight artificial intelligence models on graphics processors hosted in India and pay only for the number of tokens they use, instead of renting whole machines. Customers reach the models through a single application programming interface (API) that follows the format popularised by OpenAI, so software already written for that format can be pointed at Blaze with few changes. The company says the service runs on Nvidia GPU infrastructure, with E2E handling the hosting and scaling of the models.

Running an AI model to answer a question or produce text is called inference. It is different from training, which is the costly one-time process of building a model. For most companies the ongoing bill is inference, because every customer query costs computing power. Until now, a business in India that wanted to run an open model such as those released by research labs and tech firms usually had two choices: rent GPU servers and manage the software itself, or send its data to a foreign provider's API.

Blaze tries to offer a third route. A token is a small chunk of text, roughly a part of a word, and AI services commonly charge by the number of tokens they read and write. Paying per token means a company's bill rises and falls with actual use. A start-up testing a voice bot, for example, does not need to keep an expensive GPU switched on all night waiting for callers.

The India-hosted part matters for regulated sectors. E2E markets the service under its "Sovereign Compute" label, aimed at industries such as banking and healthcare, where rules and customer expectations often favour keeping data inside the country. Founder and managing director Tarun Dua described Blaze as the place where customers can run their production inference, and pointed to demand for frontier-class open-weight models served by an Nvidia cloud partner.

The OpenAI-compatible format is a practical choice. Many developers first build their apps against OpenAI's API. Matching that format lowers the cost of switching, which is often the biggest barrier for a smaller provider trying to win customers from global cloud companies.

Peepals timeline: E2E Networks has been expanding its GPU business through 2026, reporting strong second-quarter results driven by AI demand. It already offered TIR, a platform where customers manage their own models and training jobs on Nvidia Blackwell clusters. On 2 November it added Blaze, a managed layer on top, and alongside it set out a plan to grow its AI factory capacity by 5.4 MW to about 14.6 MW.

Peepals calculation: If 5.4 MW of new capacity brings the total to 14.6 MW, E2E's current capacity is about 9.2 MW (14.6 − 5.4), and the expansion adds about 59 per cent (5.4 ÷ 9.2 ≈ 0.59), which matches the company's own description of a nearly 59 per cent rise. Our calculation from the reported figures.

Short-term: Indian developers and start-ups get another way to try open AI models without buying or renting dedicated GPUs, and without sending data abroad. E2E did not publish per-token prices in the launch material we saw, so the real test will be how its rates compare with global providers.

Long-term: If enough businesses move their inference to India-hosted services, domestic providers such as E2E can justify building more data centre capacity, which supports the government's push for computing power located in the country.

Who is affected: Software start-ups, banks, hospitals and other firms building AI features; E2E Networks' shareholders; and global cloud providers competing for Indian AI workloads.

Questions and Answers