E2E Networks Launches Blaze To Offer Pay-Per-Token AI Inference Services

E2E said its AI factory capacity is expanding to approximately 14.6 MW. This includes a 1,024-GPU NVIDIA Blackwell cluster operational since May, alongside an additional 5.4 MW of Blackwell capacity, including HGX B300 systems.E2E Networks has launched Blaze, a pay-per-token inference service that allows developers and enterprises to access multiple open AI models through a single API. The platform, powered by NVIDIA GPU infrastructure hosted in India, is designed to support production AI workloads.

The launch comes as Indian AI companies move from pilot projects to production deployments, increasing demand for cost-efficient inference and lower latency. E2E said rising token consumption from AI agents, data sovereignty requirements and growing AI-assisted development costs are putting pressure on enterprises to improve token efficiency.

Blaze provides an OpenAI-compatible API to access open-weight models, including NVIDIA Nemotron, Sarvam, DeepSeek, Qwen, GLM and Gemma. Customers pay only for the tokens they use, while E2E manages model hosting and scaling.

“There is customer demand for frontier-class open-weight models running on an NVIDIA Cloud Partner with NVIDIA reference design clusters as Sovereign Compute. Blaze provides this on E2E Cloud in India. Blaze provides easy integration with all E2E Cloud platform services,” said Tarun Dua, Founder and Managing Director, E2E Networks.

The platform is designed for inference-heavy applications such as voice, coding agents, document AI, optical character recognition, image and video generation, and language applications. For latency-sensitive voice and chat workloads, Blaze offers faster time to the first sentence. It also uses key-value (KV) cache reuse to reduce inference costs.

Blaze integrates open-source components from NVIDIA DSX OS, the software layer within the NVIDIA DSX AI Factory Platform. NVIDIA Dynamo provides distributed model serving, including KV cache-aware routing and separate prefill and decode processing.

NVIDIA Grove and KAI Scheduler coordinate inference workloads across E2E’s Kubernetes infrastructure. Meanwhile, NVIDIA BlueField DPUs, DOCA and Infra Controller provide hardware-level isolation between customers, targeting sectors such as banking, healthcare and the public sector.

E2E is also positioning Blaze alongside its existing AI platform, TIR, which supports self-hosted inference, model pre-training and post-training on NVIDIA Blackwell GPU clusters through managed Kubernetes.

Together, the platforms offer customers a route from model training and fine-tuning to production inference. Customers can also serve their own or fine-tuned models through TIR's managed infrastructure.