E2E Networks plans a 5.4 MW Blackwell expansion and has launched its Blaze AI inference service.

India's AI ecosystem is moving from experimental pilots to production deployments. Domestic businesses, AI-native startups, research institutions, and public-sector organisations are increasingly demanding training and inference infrastructure. This shift is driving demand for high-performance graphics processing unit (GPU) infrastructure hosted in India, particularly among organisations that manage sensitive data and large-scale AI workloads.

Indian cloud provider E2E Networks has announced plans to expand its AI infrastructure capacity by approximately 5.4 megawatts (MW). It has also launched Blaze, a pay-per-token AI inference service.

The proposed expansion would increase the company's existing capacity of approximately 9.2 MW to 14.6 MW, growth of nearly 59%.

The expansion, announced on October 26, 2026, will comprise NVIDIA HGX B300 systems and other GPUs based on the NVIDIA Blackwell architecture. E2E Networks launched Blaze on November 2, extending its offering from GPU infrastructure to managed AI inference services.

**How NVIDIA Blackwell will support India's AI infrastructure**

E2E Networks intends to use the additional capacity to support workloads across the AI lifecycle, from foundation model training to enterprise inference. The proposed infrastructure will support three broad areas:

- **Foundation model training:** Compute capacity for training large language models (LLMs), post-training, and developing complex multimodal AI systems.

- **Production inference and agentic AI:** Infrastructure for enterprise AI applications, including autonomous AI agents, enterprise search, and real-time inference workloads.

- **India-hosted AI infrastructure:** GPU infrastructure that enables organisations to run relevant workloads within India, subject to their deployment, security, and compliance requirements.

For enterprise technology leaders and public-sector decision-makers, access to local compute is about more than processing capacity. It also raises questions about data sovereignty, security, and regulatory requirements. Local infrastructure can be particularly relevant for sectors such as banking, healthcare, and government-funded research, where organisations must carefully assess how sensitive data is processed and stored.

According to the company's regulatory filing, the proposed 5.4 MW addition will be financed through a combination of internal accruals and debt. E2E Networks has not disclosed a firm commissioning timeline, the investment required, or the number of GPUs planned for the additional capacity.

**What the expansion means for enterprise AI**

As organisations move beyond AI pilots, their infrastructure requirements are becoming more complex. Training, fine-tuning, and inference workloads all require GPU capacity, but enterprises must also manage compute costs, system performance, and the location of data processing.

Tarun Dua, Founder and Managing Director of E2E Networks, said in the company's announcement that Indian AI developers are moving beyond pilot-stage deployments and consuming AI tokens at increasing scale in production environments. He also pointed to demand for infrastructure that allows organisations to keep their computing workloads and data within their operational boundaries.

For regulated enterprises, infrastructure hosted in India can help address data-residency requirements. However, the location of compute alone does not establish regulatory compliance. Organisations must still evaluate data-handling practices, access controls, security arrangements, and the regulations applicable to their workloads.

**Blaze brings pay-per-token inference to India-hosted infrastructure**

Alongside its infrastructure expansion, E2E Networks launched Blaze on November 2, 2026, as a pay-per-token inference service for open AI models. Blaze gives developers and enterprises access to models through a single OpenAI-compatible application programming interface (API).

Rather than provisioning and managing dedicated GPU infrastructure, customers can use the service for inference workloads, with E2E Networks managing model hosting and scaling. Customers pay according to their token consumption.

The company has identified NVIDIA Nemotron, Sarvam, DeepSeek, Qwen, GLM, and Gemma among the model families supported by Blaze. The service runs on NVIDIA GPU infrastructure hosted in India and integrates open-source components from NVIDIA DSX OS.

According to E2E Networks, the platform uses NVIDIA Dynamo for distributed model serving and other NVIDIA technologies to manage inference workloads and provide hardware-level isolation between customers.

Blaze is designed for inference-intensive applications, including coding agents, voice applications, document AI, optical character recognition (OCR), image and video generation, and language-based services. The company is positioning the service for developers and enterprises seeking to manage inference costs and latency as their AI applications scale.

**How Blaze fits into E2E Networks' AI platform**

Blaze complements E2E Networks' existing TIR AI/ML platform. While Blaze provides API-based inference for open models, TIR supports customers that want greater control over their AI infrastructure, including model pre-training, post-training, and self-hosted inference on NVIDIA GPU clusters. Together, the two offerings are intended to support different stages of AI development, from training and fine-tuning models to running them in production.

E2E Networks says its existing AI infrastructure includes a 1,024-GPU NVIDIA Blackwell cluster that has been operational since May 2026. However, the company has not disclosed detailed per-token pricing, model versions, independently verified performance benchmarks, or Blaze-specific service-level commitments in its launch announcement.

The proposed Blackwell expansion and the launch of Blaze reflect E2E Networks' efforts to address two requirements in India's AI ecosystem: access to high-performance computing infrastructure and consumption-based inference services. For enterprises, the practical value will depend on how quickly the additional capacity becomes available, how Blaze performs under production workloads, and whether its pricing offers a viable alternative to managing dedicated GPU infrastructure.