E2E Networks has launched Blaze, a pay-per-token inference service for open AI models, offering developers and enterprises a single OpenAI-compatible API to access models including NVIDIA Nemotron, Sarvam, DeepSeek, Qwen, GLM and Gemma.
The service runs on NVIDIA GPU infrastructure in India and uses components from NVIDIA DSX OS. E2E manages model hosting and scaling, with customers charged according to token usage. The company said Blaze is designed for inference-heavy applications such as coding agents, voice, document AI, OCR and image and video generation.
Blaze uses NVIDIA Dynamo for distributed model serving, with features including KV-cache-aware routing and separate prefill and decode processing. NVIDIA Grove and KAI Scheduler coordinate workloads across E2E’s Kubernetes infrastructure, while NVIDIA BlueField DPUs and related software are used to isolate customer workloads.
E2E said Blaze is intended to address demand for lower-latency inference and India-hosted compute as businesses move AI applications from pilots into production. The service is aimed at organisations seeking to keep data and workloads within India.
Blaze will operate alongside E2E’s TIR AI platform, which supports self-hosted inference, model training and fine-tuning on managed Kubernetes infrastructure. E2E said its AI factory capacity is expanding to about 14.6 MW, including a 1,024-GPU NVIDIA Blackwell cluster that has been operational since May 2026 and an additional 5.4 MW of Blackwell capacity.
Tarun Dua, Founder and Managing Director of E2E Networks, said the service responds to customer demand for open-weight models running on NVIDIA Cloud Partner infrastructure in India.
NVIDIA South Asia Managing Director Vishal Dhupar said access to AI computing and inference services is important to the development of India’s AI ecosystem.