L&T Wins Order for 10,000-GPU Together AI Factory
Larsen & Toubro announced on August 13 that it secured a Rs 10,000-15,000 crore mega order to build a 10,000-GPU NVIDIA B300 AI Factory for Together AI at its Chennai campus. The Ken reports that the project does not involve E2E Networks, despite L&T's 2024 acquisition of a 21% stake in the Indian GPU cloud provider. The facility is intended to support large-scale inference, fine-tuning, and training workloads.
Larsen & Toubro (L&T) has secured a mega order, classified by the company as worth Rs 10,000-15,000 crore, to build a 10,000-GPU NVIDIA B300 AI Factory for US-based Together AI at L&T's Chennai data centre campus. L&T announced the order on August 13 through its LTN Compute unit and Vyoma.AI data centre business.
NDTV and the Free Press Journal report that L&T did not disclose the contract's precise value, but its "mega order" classification covers the Rs 10,000-15,000 crore range. The facility is described as India's largest single-cluster AI infrastructure project and is intended to power Together AI's cloud platform for inference, fine-tuning, and model-training workloads.
The installation is to be hosted at Vyoma.AI's Chennai campus. According to NDTV, the site's first phase is designed for 250 MW of capacity, with power infrastructure readiness of 150 MVA. The reported design combines accelerated compute, high-performance networking, low-latency interconnects, parallel storage, and AI infrastructure operations.
E2E is absent from the new project
The Ken reports that L&T acquired a 21% stake in Indian GPU cloud provider E2E Networks for Rs 1,079 crore in 2024, partly to gain access to E2E's cloud and AI platform. A senior executive in L&T's data-centre division told The Ken that E2E supplied the cloud and AI layer while L&T already had physical-infrastructure capabilities.
However, The Ken reports that E2E is not involved in the Together AI cluster. The publication quoted the same L&T executive as saying, "The direction has changed." It also reported that E2E's Blackwell cluster is housed in an L&T data centre as a tenant, rather than serving as the basis for L&T's current AI infrastructure project.
The Ken further reports that E2E's GPU fleet had operated at 35-40% utilization, versus an 80% internal target, while the company was purchasing additional GPUs and using bank loans to cover a cash shortfall. Those figures and the account of E2E's finances are reported by The Ken; L&T's announcement of the Together AI order does not address E2E's operations or explain its absence from the project.
GPU depreciation remains a cloud economics issue
The same report describes Nvidia as having arranged financing capacity of up to $500 billion through six financing companies to support selected cloud providers buying GPUs, including protection against some depreciation losses. The Ken reports that E2E is not among those providers.
For GPU-cloud operators generally, the project illustrates why scale alone is not the full infrastructure proposition. GPU utilization, capacity scheduling, networking, storage throughput, reliability guarantees, and the pace of hardware refresh all affect whether an accelerator fleet can generate returns before newer hardware changes the market. The Chennai project is notable because it combines a large B300 cluster with power and data-centre capacity sized for expansion, while Together AI provides the cloud-platform workload destination cited in L&T's announcement.
Key Points
- 1L&T's Rs 10,000-15,000 crore Together AI order centers on 10,000 NVIDIA B300 GPUs, creating a major Indian AI infrastructure deployment.
- 2The Ken reports E2E Networks is absent despite L&T's 2024 stake, highlighting the commercial importance of cloud-platform differentiation beyond GPU supply.
- 3Across GPU clouds, utilization and depreciation determine fleet economics as strongly as accelerator count, networking, storage, and data-centre power capacity.
Scoring Rationale
A reported 10,000-GPU B300 cluster and up to Rs 15,000 crore order make this a substantial AI infrastructure buildout in India. It is particularly relevant to practitioners and operators tracking regional GPU availability, cloud capacity, and the economics of operating accelerated-compute fleets.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
