Image credit: Exceptional Technology — AI-generated illustrative scene for E2E Networks’ Blaze, a cloud service with no official product photograph available.
For most Indian teams building with artificial intelligence, the hardest bill to predict is not the model — it is the machine underneath it. Rent a cluster of graphics processors and you pay whether your users show up or not; run your own serving stack and you spend your engineering weeks keeping it alive. On 2 November 2026, Delhi-based E2E Networks launched Blaze, a service built to change that arithmetic: pay only for the tokens you use, and let somebody else run the factory.
One endpoint, many models
Blaze is a pay-per-token inference service for open models. Instead of renting GPUs and running their own serving stack, the company says, developers and enterprises send requests to a single OpenAI-compatible endpoint and choose from a catalogue of open-weight models — including NVIDIA Nemotron, Sarvam, DeepSeek, Qwen, GLM and Gemma — covering text, voice, optical character recognition, multimodal work and image generation. E2E manages the hosting and the scaling; the customer is billed, in rupees, only for the tokens consumed.
That single-endpoint idea is the heart of the product. A team building a voice assistant, a coding agent, a document-reading tool or an image generator no longer has to stitch together a different serving setup for each model. One API, one bill, and the freedom to switch models as better open weights appear. “Blaze is where you can run your production inference,” said Tarun Dua, founder and managing director of E2E Networks.
Built for inference-heavy work
Blaze is designed for the products where inference cost and speed decide whether the business works at all: voice and chat, coding agents, document AI and OCR, and image and video generation. For latency-sensitive voice and chat applications, the company says, Blaze offers a fast time to the first sentence, while open-weight models benefit from high KV-cache reuse that brings inference cost down. As Indian AI companies move from pilots to production, the company says, token consumption is climbing fast — and so is the pressure on chief information officers to make every token count.
Keeping the compute in India is part of the pitch too. Blaze runs on NVIDIA GPU infrastructure hosted in India, and the company says the service is built for customers with strict data rules — banks, healthcare providers and public-sector bodies among them — who need to know where their data is processed.
More Read
The engine underneath: NVIDIA DSX OS
Under the hood, Blaze integrates open-source components from NVIDIA DSX OS, the open, modular software layer within the NVIDIA DSX AI Factory Platform. NVIDIA Dynamo provides distributed model serving, with KV-cache-aware routing, disaggregated prefill and decode, and fast data transfer between GPUs through NVIDIA NIXL. NVIDIA Grove and the KAI Scheduler coordinate inference workloads across E2E’s Kubernetes fleet, while NVIDIA BlueField data processing units, NVIDIA DOCA and the NVIDIA Infra Controller enforce hardware-level isolation between customers.
“Access to AI computing and inference services is essential to advancing India’s AI ecosystem,” said Vishal Dhupar, managing director for South Asia at NVIDIA, welcoming a service that makes accelerated computing more accessible to India’s developers and enterprises.
From training to tokens on one platform
Blaze does not arrive alone. It sits alongside TIR, E2E’s AI platform for self-hosted inference and model pre-training and post-training on high-performance NVIDIA Blackwell GPU clusters through managed Kubernetes. The two together, the company says, give customers a flexible path: train and fine-tune on TIR with greater control over infrastructure and customisation, then serve production traffic through Blaze’s token-efficient API — or serve their own or fine-tuned models on TIR’s managed Kubernetes with integrated NVIDIA Dynamo.
The capacity behind both is growing. E2E says its AI factory capacity is expanding to about 14.6 MW, including a 1,024-GPU NVIDIA Blackwell cluster live since May 2026 and an additional 5.4 MW of Blackwell capacity including HGX B300 systems and other Blackwell GPUs.
From a 2009 cloud bet to an AI factory
E2E Networks began in 2009 in New Delhi as a contractless cloud computing player aimed at value-conscious startups, and the company says its cloud powered the early scaling of many of today’s well-known Indian startups. It listed on NSE Emerge in 2018 in an issue oversubscribed 70 times, moved to the NSE main board in 2022, and launched its TIR AI platform in 2023. Blaze is the next step in that journey: the same India-first cloud instinct, now aimed squarely at the production inference bill.
For India’s AI builders, the promise is simple and worth celebrating. The models are increasingly open, the talent is home-grown, and now the factory floor that serves those models at scale — billed by the token, hosted in India — is open for business too.
Keep reading: Maths, Code and 22 Languages: Inside Soket AI’s EKA-1, India’s New Homegrown Foundation Model · Built From Scratch, Running Offline: Inside ZenteiQ’s Dhīr Models for Indian Engineering · More Industrial Automation & Technology stories
Like stories like this? Add Exceptional Technology as a preferred source on Google to see more of our reporting in your search results and Discover feed.
