Indian cloud infrastructure company E2E Networks has launched Blaze, a pay-per-token AI inference platform that allows developers and enterprises to access multiple open-weight AI models through a single API. Launched on November 2, 2026, the service runs on NVIDIA GPU infrastructure hosted in India and is designed to simplify the deployment of production AI applications.
Blaze provides access to models including NVIDIA Nemotron, Sarvam, DeepSeek, Qwen, GLM and Gemma through an OpenAI-compatible API. Instead of managing GPU infrastructure and model hosting independently, customers can use E2E’s managed platform and pay according to the tokens they consume.
What Is E2E Networks’ Blaze AI Platform?
Blaze is an AI inference service that allows businesses to run open-weight AI models without independently managing the underlying GPU infrastructure.
AI inference refers to the process of using a trained AI model to generate responses, analyse information, write code or complete other tasks. As businesses increasingly integrate AI into their products, inference has become an important part of operating AI applications.
Traditionally, organisations have needed to arrange GPU capacity, deploy models, manage inference servers and scale infrastructure according to demand. These requirements can increase operational complexity and costs, particularly for startups and businesses handling unpredictable AI workloads.
Blaze aims to simplify this process through a managed inference platform.
Its key features include:
- A single OpenAI-compatible API for accessing multiple open-weight models.
- Pay-per-token billing based on consumption.
- NVIDIA GPU infrastructure hosted in India.
- Managed model hosting and automatic scaling.
- Optimised inference for coding agents, voice applications and document processing.
- Support for enterprise AI workloads requiring local infrastructure.
The platform is designed for both Indian and international customers, according to E2E Networks’ regulatory filing.
Which AI Models Are Available on Blaze?
One of Blaze’s main features is its ability to provide access to multiple AI models through a common API.
Instead of integrating different providers individually, developers can use a unified interface to work with supported models.
The company has identified the following model families:
| AI Model | Developer | Primary Focus |
|---|---|---|
| NVIDIA Nemotron | NVIDIA | Reasoning and enterprise AI |
| Sarvam | Sarvam AI | Indian language AI |
| DeepSeek | DeepSeek | Reasoning and coding |
| Qwen | Alibaba | General-purpose AI and coding |
| GLM | Zhipu AI | Language and reasoning |
| Gemma | Open-weight AI applications |
The availability of multiple models gives developers flexibility to select systems according to their application requirements.
For example, a coding application may use a model suited to programming tasks, while a multilingual customer-support application may benefit from models with stronger language capabilities.
E2E manages the hosting and scaling of these models, allowing customers to focus on building applications rather than maintaining inference infrastructure.
How Blaze Uses NVIDIA GPU Infrastructure
Blaze is built using NVIDIA GPU infrastructure and integrates open-source components from NVIDIA DSX OS, the software layer within NVIDIA’s DSX AI Factory Platform.
The platform uses several NVIDIA technologies to improve model serving, manage workloads and optimise GPU utilisation.
NVIDIA Dynamo for Distributed Model Serving
Blaze uses NVIDIA Dynamo to manage distributed AI inference workloads.
The technology supports KV-cache-aware routing, which helps reuse previously computed information during repeated interactions. This is particularly useful for AI agents and applications that process long conversations or documents.
It also separates two major stages of inference:
- Prefill: Processes the input prompt and prepares the information required by the model.
- Decode: Generates the output tokens one by one.
Separating these processes can help improve resource allocation and inference efficiency.
NVIDIA Grove and KAI Scheduler
E2E also uses NVIDIA Grove and KAI Scheduler to coordinate inference workloads across its Kubernetes infrastructure.
These technologies help manage GPU resources and distribute workloads across available computing capacity.
Hardware-Level Customer Isolation
For enterprise customers, Blaze integrates NVIDIA BlueField Data Processing Units (DPUs), NVIDIA DOCA and NVIDIA Infra Controller.
These technologies are used to provide hardware-level isolation between customers.
This is particularly relevant to organisations in banking, healthcare and the public sector, where data security and infrastructure separation are important requirements.
Blaze Targets AI Agents, Coding and Document Processing
E2E Networks is positioning Blaze for applications that require frequent AI inference and efficient token consumption.
The company has identified several major use cases.
| Application | How Blaze Can Be Used |
|---|---|
| Coding agents | Code generation, debugging and automated programming tasks |
| Voice AI | Real-time conversational applications |
| Document AI | Extracting and processing information from documents |
| OCR | Converting images and scanned documents into machine-readable text |
| Image generation | Generating images through AI models |
| Video generation | Supporting AI-powered video creation workflows |
| Language applications | Chatbots, assistants and multilingual applications |
For voice and chat applications, the platform is designed to deliver faster time to the first sentence.
Blaze also uses key-value (KV) cache reuse to reduce redundant computation. When an application repeatedly sends similar context, previously calculated information can be reused rather than processed entirely from scratch.
This can improve efficiency in applications involving long conversations and repeated agent interactions.
However, E2E has not published independent comparative benchmarks establishing specific latency or cost improvements across these workloads.
How Pay-Per-Token Pricing Works
Blaze uses a consumption-based pricing model in which customers pay for the tokens they consume.
Tokens are the basic units of text processed by AI models. Both incoming prompts and generated responses contribute to token consumption.
For example, an application that processes thousands of customer questions will consume tokens based on the amount of input information and the responses generated by the AI model.
Under Blaze’s model, customers pay according to their actual usage rather than independently provisioning dedicated GPU infrastructure for every workload.
This approach could benefit startups and enterprises that experience fluctuating AI demand.
However, E2E has not disclosed detailed per-token prices, model-specific rates or comparative cost figures in its launch announcement. Therefore, the platform’s actual cost advantage will depend on its pricing structure and the models selected by customers.
E2E Networks Expands Its NVIDIA Blackwell AI Infrastructure
The Blaze launch is part of E2E Networks’ broader investment in AI computing infrastructure.
The company said its AI factory capacity is expanding to approximately 14.6 megawatts (MW).
This includes a 1,024-GPU NVIDIA Blackwell cluster that became operational in May 2026, alongside an additional 5.4 MW of Blackwell capacity, including NVIDIA HGX B300 systems and other Blackwell GPUs.
The expanded infrastructure is intended to support growing demand for AI model training, fine-tuning and production inference.
E2E Networks is also positioning Blaze alongside its existing TIR AI platform.
While Blaze provides API-based inference for open-weight models, TIR allows customers to run self-hosted inference, model pre-training and post-training through managed Kubernetes infrastructure.
Together, the platforms provide customers with infrastructure covering different stages of AI development, from model training to production deployment.
Why India’s AI Industry Needs More Inference Infrastructure
India’s AI ecosystem is gradually moving beyond experimentation towards commercial deployment.
Startups are developing AI-powered applications, enterprises are integrating AI into business operations, and developers are increasingly building autonomous agents capable of completing multiple tasks.
These developments are increasing demand for GPU computing and inference services.
However, operating AI applications at scale involves several challenges, including infrastructure costs, response latency, GPU availability and data residency requirements.
India-hosted inference infrastructure could help organisations that want their computing workloads processed within the country.
For startups, managed inference platforms can reduce the need to purchase and maintain expensive GPU infrastructure. Enterprises can also benefit from consumption-based access when demand fluctuates.
Nevertheless, local hosting alone does not guarantee regulatory compliance. Businesses must still evaluate security arrangements, data handling policies and applicable regulatory requirements.
The Bigger Picture
E2E Networks’ Blaze launch highlights the growing competition in the AI infrastructure market. As AI models become more accessible, the ability to deploy them efficiently and affordably is becoming just as important as developing increasingly powerful models.
The shift towards AI agents is also changing infrastructure requirements. Unlike traditional chatbots, autonomous agents may make repeated model calls, interact with external tools and process large amounts of information. This can significantly increase token consumption and inference costs.
By combining multiple open-weight models, NVIDIA GPU infrastructure and consumption-based billing, E2E is positioning itself as an Indian provider of managed AI inference services. Its success will depend on pricing competitiveness, reliability, model availability and performance under real-world workloads.
Looking Ahead
E2E Networks will look to attract AI startups, developers and enterprises seeking managed inference services without operating their own GPU infrastructure. The company’s expanding NVIDIA Blackwell capacity and integration with NVIDIA’s software ecosystem could strengthen its position in India’s growing AI infrastructure market.
For customers, the important factors will be the actual cost per token, response latency, available model versions and service reliability. As more businesses move AI applications into production, platforms such as Blaze could become an important part of the infrastructure supporting India’s next generation of AI products.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
