L&T's data center unit will host Nvidia hardware on behalf of AI cloud provider Together AI, extending the Indian operator's footprint in dedicated AI workloads. Financial terms, campus locations, accelerator generations, and contract duration remain undisclosed.

What does L&T bring to the deal?

Larsen & Toubro, the Mumbai-based engineering and construction conglomerate, operates one of India's larger domestic data center platforms. The unit serves enterprise, government, and hyperscale customers and has widened its tenant categories as AI demand has reshaped the high-density colocation segment.

L&T enters the AI hosting segment with installed high-density capacity and existing relationships with Nvidia's regional channel partners. The Together AI deployment functions as an anchor tenant at sites built for sustained accelerator operation.

L&T's data center business traces back to the late 2010s. The unit has stood up multi-megawatt ready facilities across multiple Indian metros, targeting both domestic enterprises and global colocation customers entering the South Asian market.

What is Together AI's role?

Together AI runs three product lines:

• Inference APIs for open-source large language models, sold to developer and enterprise users
• Fine-tuning services for adapting open-weight models on customer data
• Dedicated training clusters leased by the month to model developers

The firm has scaled from US-based clusters since early 2023 and now serves customers across North America, Europe, and parts of Asia. The L&T arrangement expands Together AI's footprint into India at a time when Indian enterprises are increasing direct investment in domestic LLM inference capacity.

How has the GPU colocation market shifted?

GPU colocation consolidated into a distinct market segment over 2023 and 2024. The shift stems from AI training and inference workloads, which require rack densities and cooling arrangements that conventional wholesale colocation does not deliver.

Operators with multi-megawatt power envelopes and direct liquid cooling have signed tenants committing to multi-year block allocations. Pricing per kilowatt typically exceeds conventional wholesale rates by a premium reflecting both the build-out cost and the demand imbalance.

Key technical drivers of the shift:

• Sustained rack densities above 40 kW per cabinet
• Direct liquid or immersion cooling
• Power redundancy at the megawatt scale
• Low-latency east-west fabric between accelerator nodes

What does the deal signal for the wider market?

The arrangement confirms two parallel shifts:

• Hyperscalers continue absorbing every megawatt of accelerator capacity they can finance, leaving smaller providers to seek alternative routes
• Operators in geographies outside the US — including India, Singapore, and parts of Europe — are absorbing redirected demand

L&T's commitment to a Western AI cloud signals confidence in the unit's high-density-ready buildout. The deployment's full scope will become visible as the operator files the standard disclosures Indian colocation firms release on power consumption and contract milestones. Industry observers will watch the rollout for indications of further Indian AI infrastructure investment, particularly if combined with renewable-power procurement commitments. A successful first deployment could open the door for additional Western AI cloud operators to follow.