The Pulse
E2E Networks Launches Blaze, an India-Hosted AI Inference API
E2E Networks launched Blaze, a pay-per-token service that routes requests to open AI models through one OpenAI-compatible API on GPU infrastructure hosted in India. The company named six model families at launch and said the service is avai
AI.info Team · Written with AI · fact-checked ·
Six model families behind one API
E2E Networks has launched Blaze, a pay-per-token inference service that connects developers and companies to six open-model families through one OpenAI-compatible API. The company named NVIDIA Nemotron, Sarvam, DeepSeek, Qwen, GLM and Gemma in its November 2 filing to India’s stock exchanges. It says the service uses NVIDIA GPU infrastructure hosted in India and handles model hosting and scaling for customers.
Blaze is aimed at customers who want to send inference requests without first arranging GPU capacity and serving infrastructure themselves. E2E says users pay only for the tokens they consume. The filing lists both domestic and international markets and says the product is launched globally, but does not give token prices or identify specific model versions.
Tarun Dua, founder and managing director of E2E Networks, described the service as a production option for open-weight models: “Blaze is where you can run your production inference.” He said customers are asking for those models on NVIDIA reference-design clusters operating as sovereign compute, and that Blaze runs on E2E Cloud in India.
NVIDIA software handles serving and isolation
The service combines NVIDIA GPU infrastructure with open-source components from NVIDIA DSX OS, the software layer within NVIDIA’s DSX AI Factory Platform. E2E says NVIDIA Dynamo handles distributed model serving, including routing that accounts for KV-cache reuse and separate prefill and decode stages. NVIDIA NIXL moves data between GPUs, according to the company.
NVIDIA Grove and KAI Scheduler coordinate inference workloads across E2E’s Kubernetes fleet. For customer separation, the release points to NVIDIA BlueField data-processing units, DOCA software and NVIDIA Infra Controller, which it says enforce hardware-level isolation. E2E identifies banks, healthcare providers and public-sector organisations with strict data rules as potential beneficiaries, but the filing does not provide independent performance or security test results.
Vishal Dhupar, NVIDIA’s managing director for South Asia, said: “Access to AI computing and inference services is essential to advancing India’s AI ecosystem.” He added that NVIDIA’s accelerated-computing platform gives cloud providers such as E2E a foundation for making AI more accessible to Indian developers and businesses.
Blaze sits alongside E2E’s TIR platform
E2E positions Blaze alongside TIR, its AI platform for training and model deployment. Blaze offers API-based inference for open-weight models; TIR gives customers more direct control through managed Kubernetes, self-hosted inference, and pre-training and post-training on NVIDIA Blackwell GPU clusters. The filing says customers can also run their own or fine-tuned models on TIR with integrated NVIDIA Dynamo.
The distinction is between calling a managed model endpoint and managing more of the model and infrastructure stack. That gives E2E a way to serve customers who want a ready API as well as teams that need to customize or host models themselves. The filing does not specify which Blaze models are live at launch or describe how customers select among them.
A 14.6 MW AI factory plan
Blaze turns part of E2E’s GPU capacity into a metered inference service. The company says its AI-factory capacity is expanding to about 14.6 megawatts, including a 1,024-GPU NVIDIA Blackwell cluster that has been live since May 2026 and an additional 5.4 MW of Blackwell capacity, including HGX B300 systems and other Blackwell GPUs.
The filing does not say how much of that capacity Blaze uses, how many customers have signed up, or what response times the service delivers. Its stated offer is narrower and concrete: one compatible API for named open-model families, hosted on NVIDIA GPU infrastructure in India, with charges based on tokens used.