L&T NVIDIA B300 AI Factory: A 10,000-GPU Mega Deal

The L&T NVIDIA B300 AI Factory became official on August 13, 2026. For India’s AI ambitions, it is a landmark moment. Larsen & Toubro, the country’s engineering giant, secured a mega order worth Rs 10,000 to 15,000 crore (up to $1.57 billion). In return, it will build India’s largest single-cluster AI infrastructure. The customer is Together AI, a US cloud company. The site is Chennai.

On August 13, Larsen & Toubro announced a win for its AI and data centre business, Vyoma.AI. Through its subsidiary LTN Compute, the group had landed India’s largest single-cluster AI infrastructure order. According to the company’s announcement, the plan is an integrated NVIDIA B300 AI Factory at Vyoma’s Chennai data centre campus.

The word integrated is there for a reason. Instead of a plain server hall, the facility bundles hyperscale data centre infrastructure, accelerated computing, high-performance networking, ultra-low-latency interconnects and parallel storage into one stack. As a result, the deal also marks L&T’s first real step into the AI factory business.

The headline figures explain why Indian business media spent the whole day on this story. Here is the deal in one view.

For context, this is one of the largest AI computing deployments announced in India so far. Moreover, L&T classifies a mega order as anything above Rs 10,000 crore, a label it rarely gets to use for a digital business.

NVIDIA uses the term for a data centre purpose-built to produce one thing: AI. Rather than hosting websites or databases, an AI factory runs training, fine-tuning and inference around the clock. Instead, it counts its output in tokens served rather than pages loaded. So NVIDIA pitches it as infrastructure that lets any enterprise operate like a hyperscaler.

The concept is also becoming a financial story. In fact, NVIDIA says it has lined up partners such as Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilise over $500 billion for AI factory financing. In other words, investors now treat compute as an asset class, and the Chennai project plugs India straight into that trend.

The B300 sits in NVIDIA’s Blackwell Ultra generation, the current top of its data centre line. In the DGX B300 system, eight of these GPUs share 2.1 TB of memory and deliver 144 petaflops of FP4 compute. Compared with the previous B200, NVIDIA claims 1.5x the FP4 performance. It also promises up to 50x more inference throughput per megawatt than older Hopper systems.

Inference is where most AI money now flows, so that throughput claim is the one Together AI’s customers will feel first. Meanwhile, the same Blackwell family reaches all the way down to the desk. The NVIDIA DGX Spark is the tiny sibling of the machines Chennai will run at rack scale.

Together AI is a San Francisco based company that runs what it calls an AI Acceleration Cloud. It rents out GPU compute and serves hundreds of open-source models for training, fine-tuning and inference. As a result, it is a popular alternative to building your own cluster. In short, the Chennai AI Factory will power exactly that platform.

Why India? The global GPU crunch has cloud providers hunting for sites with cheap power, fast construction and room to grow. In addition, India offers all three, plus a huge developer market of its own. Therefore the partnership gives Together AI capacity, and it gives L&T a marquee anchor tenant.

India already hosts GPU clusters, and the government’s IndiaAI Mission subsidises access to tens of thousands of accelerators. That same national push set the agenda at the India AI Summit. Policy, infrastructure and talent were the three pillars on show. However, nothing announced so far matches this project as a single cluster. That is why L&T, Reuters and the trade press all describe it as India’s largest.

L&T conceived the Chennai campus as a gigawatt-scale AI infrastructure site. Phase 1 alone targets 250 MW of capacity, with 150 MVA of power readiness. Therefore the 10,000-GPU deployment is the opening move, not the end state.

For L&T, the order opens a new business line at exactly the right time, since AI infrastructure spending is one of the few budgets still growing everywhere. For India, the L&T NVIDIA B300 AI Factory proves the country can build and run world-class AI compute at home. Until now, that meant renting clusters abroad.

The contrast with the United States is sharp. While projects there stall under local pushback, India is courting these builds with land, power and policy. If Chennai delivers, expect the next mega order to arrive much faster than this one did.

If the hardware side has you curious, start with our picks for the best GPU for AI at home. And for the other side of the data centre boom, read our report on AI data center public opposition in the US.

*Sources: Business Standard, Yahoo Finance, Electronics For You, NVIDIA, best-ai.news.*