India’s AI industry is getting a major infrastructure boost. Cloud provider Vultr has announced a 20 MW AI supercluster in Chennai, while Indian cloud company E2E Networks plans to add 5.4 MW of NVIDIA Blackwell-based capacity and has launched its Blaze AI inference service.

These are not consumer products like a new smartphone or app. They are large computing systems that can help startups, enterprises, researchers and government bodies build and run AI tools inside India.

The announcements are important because advanced AI needs huge amounts of computing power. Local access to that power can reduce delay, improve data control and make it easier for Indian companies to move AI projects from testing to real-world use.

Vultr’s 20 MW AI supercluster in Chennai

Vultr has announced a new 20 MW AI supercluster in Chennai, powered by AMD Instinct MI355X GPUs. The company says the initial setup will include around 10,000 GPUs and will be part of its new Chennai cloud region.

A megawatt, or MW, is a measure of electrical power. So, 20 MW indicates the scale of electricity needed to operate the servers, GPUs, cooling equipment and supporting infrastructure. It does not mean the system delivers 20 MW of AI performance.

A supercluster is a large group of connected high-performance computers. In this case, the GPUs can work together on demanding AI jobs such as training large language models, fine-tuning a company’s AI model or processing thousands of AI requests at the same time.

Vultr has said the Chennai capacity will start at 20 MW, with further expansion planned from 2027. The company is also targeting a much larger AI infrastructure presence in India over the next few years. However, these future targets should be viewed as plans, not completed capacity.

What businesses can do with the Chennai facility

A startup building an Indian-language AI assistant may need to train and test its model on powerful GPUs. A bank may need to run fraud-detection models while keeping sensitive data close to home. A university may need computing resources for scientific research.

Instead of buying and maintaining a large number of expensive GPUs, these organizations can rent cloud computing power. This makes advanced AI infrastructure more accessible, though the final cost will depend on the workload, usage time and service terms.

Vultr’s use of AMD GPUs also adds another option to a market often associated with NVIDIA hardware. More infrastructure choices may give businesses greater flexibility when selecting AI platforms.

E2E Networks plans Blackwell expansion

E2E Networks has proposed adding approximately 5.4 MW of AI computing capacity using NVIDIA HGX B300 systems and other GPUs based on NVIDIA’s Blackwell architecture. The proposed addition would take its AI factory capacity from about 9.2 MW to around 14.6 MW.

Blackwell is NVIDIA’s newer GPU architecture designed for AI workloads. It is intended to support large model training, fine-tuning and inference. Inference is the stage where a trained model answers a user’s question, generates text, analyses a document or creates an image.

E2E has said it plans to fund the expansion through internal funds and debt. The company has not disclosed a final investment amount, GPU count or commissioning date, so the expansion should be treated as a plan rather than an already completed deployment.

E2E Networks was founded by Tarun Dua, who is also its founder and managing director. The company is positioning itself as an India-hosted cloud provider for AI developers and businesses that want their workloads closer to local users.

E2E Networks plans 5.4 MW NVIDIA Blackwell expansion of India’s AI Factory https://t.co/OW224nNgDi

— AIM Investments (@AimInvestments) October 26, 2026

Blaze AI service – how it works

E2E Networks launched Blaze on November 2 as a pay-per-token AI inference service for open models. A token is a small unit of text processed by an AI model. Depending on the model, a word may be one token or several tokens.

With Blaze, developers can use one OpenAI-compatible API to access selected open models, including families such as Sarvam, DeepSeek, Qwen, Gemma and NVIDIA Nemotron. Instead of hosting and scaling the models themselves, customers send a request through the API and pay according to the tokens used.

For example, a customer-support tool can send a user’s question to an AI model through Blaze and receive a drafted response. E2E manages the underlying GPU infrastructure and model serving.

The service is aimed at use cases such as voice tools, coding assistants, document analysis, optical character recognition and image or video generation. Its India-hosted infrastructure may be especially useful for organizations with strict data-handling requirements.

E2E Networks has launched Blaze, a pay-per-token artificial intelligence inference service for open AI models. The product is aimed at developers and businesses that want to use powerful AI models through an application programming interface, or API, without renting and managing GPUs themselves.

The launch is important because many Indian companies are moving from AI experiments to real products. A chatbot, document-reading tool or coding assistant may work well in a pilot, but running it for thousands of users requires reliable infrastructure, speed and predictable billing. Blaze is designed to address that production stage.

What is Blaze AI inference service

Blaze gives users access to supported open AI models through a single API that follows the OpenAI-compatible format. This can make adoption easier for developers who already use software tools built around that format.

E2E Networks says Blaze runs on NVIDIA GPU infrastructure hosted in India and uses NVIDIA DSX OS, a software layer for serving AI models at scale. The service is designed for tasks such as chat applications, document processing, voice tools, coding agents, image generation and video-related workloads.

The company has listed open-model families including NVIDIA Nemotron, Sarvam, DeepSeek, Qwen, GLM and Gemma. The exact models available, speeds and pricing may change as the platform develops.

How pay-per-token billing works

A token is a small unit of text that an AI model reads or produces. It can be a word, part of a word, punctuation mark or a short piece of code.

For example, when a user asks an AI assistant to summarize a 20-page report, the model first processes the text in the report. These are input tokens. It then creates the summary, which becomes output tokens.

Under a pay-per-token model, the customer pays according to the amount of text or other model usage processed. This is different from renting a GPU server by the hour. With a GPU rental, a company may pay even during quiet periods. With token-based billing, costs rise mainly when the AI service is actively used.

This can be useful for a small business with irregular demand. A legal-tech startup, for instance, may use Blaze to summarize contracts only when customers upload documents. It does not need to keep a dedicated GPU running all day.

Why it matters for Indian AI companies

For developers, the attraction is convenience. They can select a model, generate an API key and add AI features without handling server deployment, scaling or model updates themselves.

For larger organizations, data location may also matter. Companies working with sensitive customer data often consider where their AI workloads are processed. E2E is positioning Blaze as an India-hosted option, although businesses should independently review data handling, retention policies, security controls and legal compliance before using any AI service.

Blaze also gives developers a choice beyond closed-model platforms. Open models can be attractive when companies want more flexibility in model selection, fine-tuning or cost planning.

Similar AI inference technologies

Blaze enters a crowded market. OpenAI, Anthropic and Google offer API-based access to their own models, generally billed by usage. These services are popular for teams that want strong proprietary models and managed platforms.

For open-model inference, companies such as Together AI, GroqCloud, Hugging Face and OpenRouter offer different combinations of model access, fast inference, routing and developer tools. Cloud providers including Amazon Web Services, Google Cloud and Microsoft Azure also offer managed AI model services.

The key difference is not just price. Developers should compare model quality, response speed, context limits, uptime, data controls, regional hosting, API compatibility and support. A lower token price is useful only if the model performs well for the actual task.

What users should watch next

Blaze’s success will depend on clear pricing, reliable performance and a useful catalogue of models. Developers should test the same prompt across two or three services before making a long-term choice.

AI inference is becoming a basic business utility, much like cloud storage or payment processing. If Blaze can offer dependable India-based access to capable open models, it may become a practical option for startups and enterprises building AI products at scale.

Facts Input- E2E Announcement, E2E PDF, Stockssena, E2E Pricing