For a company most of the world has never heard of, AM Intelligence is spending like a buyer that expects to matter. On October 5, the Indian artificial-intelligence infrastructure firm said it had placed two binding orders for roughly 20,000 of Nvidia’s next-generation Vera Rubin graphics processing units, a purchase worth about $4 billion that will feed data centers in India and Malaysia.
The orders come barely two months after the company committed to 9,000 Rubin GPUs for a facility in Hyderabad. Taken together, the two deals lift AM Intelligence’s committed Vera Rubin capacity to about 29,000 chips, according to people familiar with the matter, and position the firm to run what it describes as the largest operational compute capacity in India by the second quarter of 2027.
AM Intelligence is an unusual entrant in the global race for AI hardware. It was founded by the team behind Greenko, the Indian renewable-energy developer that assembled one of the country’s largest clean-power portfolios. That pedigree matters now. AI data centers are constrained less by the cost of the chips themselves than by the cost and availability of the electricity required to run them, and a founder group that knows how to finance and build power projects has a running start.
The new orders represent about 70 megawatts of planned capacity, part of a broader expansion of roughly 100 megawatts across India and Malaysia, people familiar with the deal said. The Vera Rubin line, Nvidia’s successor to its Blackwell chips, is the foundation of that build-out, offering more compute per watt at a moment when power is the binding constraint on AI expansion.
For Nvidia, the order is another data point in a widening pattern. The chipmaker’s most advanced systems are increasingly sold, in ever larger blocks, to a new class of customer: not the established cloud providers that dominated the past decade, but well-capitalized infrastructure firms and state-backed players building compute from scratch. AM Intelligence is one of several such buyers emerging from India, where Yotta, a domestic data-center operator, has outlined GPU purchase plans worth tens of billions of dollars.
The timing reflects a market in which supply remains tight. Global chip shortages have kept lead times long, and Nvidia’s allocation decisions now shape which regions and which customers get to scale first. By locking in orders before the Vera Rubin line ships at volume, AM Intelligence is effectively buying a place in a queue that will determine who gets to train the next generation of models.
Analysts said the economics are still being tested. A $4 billion commitment to chips is a large wager for a company whose cloud-computing operation is young, and the payoff depends on winning enough AI workloads to keep the new capacity busy. Data-center operators across Asia have announced ambitious build-outs in recent months; the harder question is whether demand will arrive as quickly as the hardware does.
The company is betting that India’s push into AI, backed by government incentives and a deep pool of engineering talent, will generate that demand. Malaysia offers a second lane, marketed as a lower-cost hub for the same work. Between the two markets, AM Intelligence is trying to build a regional compute franchise rather than a single national one, the people said.
The Vera Rubin architecture is part of the appeal. The chips are designed for the largest training runs, and early customers have described the step up from Blackwell as one that matters most to the frontier labs. A firm that can offer that class of hardware inside India, without the latency and cost of renting it from a Western cloud provider, is wagering that there is a local market willing to pay for it.
There is a debt-and-power story underneath the chip story. Financing 29,000 high-end GPUs means financing the megawatts to run them, and the Greenko lineage is meant to signal that AM Intelligence can line up both. Whether the company can convert that advantage into steady cloud revenue is the open question, and it will not be answered until the first Vera Rubin systems come online in Hyderabad.
The first systems are not due until early next year. The Hyderabad facility is scheduled to take delivery of its Vera Rubin systems in the first quarter of 2027, with the India and Malaysia expansion to follow. That gives AM Intelligence a runway measured in quarters rather than years to line up customers for capacity that will not earn a rupee the moment the chips arrive. In a market where rivals from Reliance to Yotta are racing toward the same finish, the edge goes to whoever can switch on first.
None of this is certain. Orders can be shifted, build-outs can slip, and the AI market has already shown how quickly one year’s shortage becomes the next year’s surplus. But the direction is clear. Capital is flowing into compute outside the United States, and a new set of companies is forming to collect it. AM Intelligence has now put $4 billion on the table, the kind of commitment that signals it intends to be more than a footnote in the next chapter of the chip business.
