Meta's Reliance deal puts 168 megawatts of Indian compute behind the company's global AI push

On the morning of 10 June 2026, Meta confirmed its first AI data centre partnership in India, signing a deal with Mukesh Ambani's Reliance to develop a 168-megawatt facility that the company says will serve its global AI computing needs and can be expanded over time. The announcement, reported by TechCrunch and confirmed by Reuters, marks the social-media company's first significant compute footprint in the world's most populous country and pulls India's largest conglomerate deeper into the global AI infrastructure race.

The deal lands at an awkward moment for Meta. While the company's open-weight Llama models have kept it in the conversation, prediction markets are giving the firm only a 15% chance of producing a top-tier frontier AI model by the end of the year, according to a Polymarket contract tracked on 9 June 2026. The Reliance partnership is, in effect, a wager that compute, not model architecture, will be the binding constraint on Meta's standing in the next phase of the AI build-out.

A 168-megawatt first step

TechCrunch, citing people familiar with the arrangement, described the facility as a 168-megawatt build with provisions to scale. That is a meaningful but not enormous footprint: it is roughly the capacity of a single mid-sized hyperscale campus in the United States, and a small fraction of the multi-gigawatt clusters now under construction by Microsoft, Google and Amazon in Virginia, Texas and the Gulf. What makes the project interesting is less the headline number than the location and the partner.

Reliance brings three things that are hard to replicate for a foreign entrant. It controls land, power contracts and fibre in several Indian states, including Gujarat and Maharashtra, where hyperscale demand is outrunning supply. It has a long-standing relationship with Indian state governments that are eager to host AI capacity. And, through its Jio Platforms unit, it can offer Meta preferential access to the country's mobile data market — by some measures the largest in the world by volume — at a moment when Meta's WhatsApp and Instagram businesses are the company's fastest-growing revenue lines. The Indian Express also reported the partnership on 10 June, framing it as an AI-powered data centre deal with national reach.

The structure of the agreement has not been disclosed in detail. It is not clear whether Reliance is acting as a pure real-estate and power partner, a co-investor, or a longer-term capacity off-taker that will resell compute to Indian enterprise customers. Meta's statement, as carried by Reuters, emphasised that the facility will support the company's global AI computing needs, language that suggests Meta itself is the primary tenant rather than Reliance.

Why India, and why now

The conventional read is that Meta is following the playbook of its US peers, who have spent the past two years anchoring large compute clusters in jurisdictions with cheap power, capable construction labour and friendly regulators. India offers all three. It also offers something the Gulf and the US Midwest do not: a domestic market of more than 800 million internet users, most of them accessing Meta's services on low-cost Android handsets served by Jio's network.

That last point matters for the political economy of the deal. Indian policymakers have spent the last two years pushing back against the dominance of foreign cloud and AI providers on Indian soil, both through the production-linked incentive scheme for semiconductors and through stricter data-localisation rules. A deal that anchors a hyperscale facility in India, under Indian corporate control, and that promises to make Indian users' AI workloads run on Indian soil, fits neatly inside the government's industrial-policy frame.

The Reliance partnership also gives Meta a degree of insulation from the geopolitical crosswinds now blowing through the AI supply chain. US export controls on advanced accelerators, tightened in 2024 and 2025, have not been applied to India in the same way they have been applied to China, but the policy mood in Washington has soured on the spread of frontier-grade compute to any jurisdiction outside a trusted circle. Building through a domestic champion, rather than a US- or Singapore-based joint venture, is a quieter way to keep the lights on.

The counter-narrative: 15% odds and a crowded market

The most obvious counterpoint is the one Polymarket is already pricing. Meta is currently treated as an outside bet to ship a frontier model by year-end, trailing OpenAI, Anthropic, Google DeepMind and, by some accounts, the Chinese labs. A 168-megawatt facility in Jamnagar or Nagpur will not change that ranking on its own. Frontier-lab status is still set by model benchmarks and by the talent density around a small number of research campuses in California and, increasingly, Hangzhou and Shenzhen. Compute is necessary but not sufficient.

There is also a less flattering read. Meta has spent the better part of three years trying to compress its AI capex into a single year-end print, the way a retailer compresses inventory into a holiday quarter. The Reliance deal extends that spending over a multi-year horizon and shifts some of it offshore, which helps the optics of 2026 and 2027. Whether it helps the actual product roadmap is a separate question, and one the markets will answer in the next two earnings cycles.

The Reliance side of the partnership carries its own risks. The company is investing heavily across renewables, telecom and retail, and a hyperscale build-out is a different kind of bet: lower margins, longer payback, and exposure to a single tenant's decisions. If Meta decides the model race is lost and reallocates capacity to its existing US footprint, Reliance is left with a 168-megawatt shell and a Power Purchase Agreement to honour.

What the deal actually signals

Read closely, the Reliance agreement is less a marquee AI infrastructure story than a piece of a broader platform-anchoring strategy. Meta needs Indian compute not only to train its next generation of models, but to serve inference for WhatsApp, Instagram and Facebook in a market where latency, data-residency rules and the cost of round-tripping traffic across the Indian Ocean all matter. A domestic facility is, in that sense, an extension of the company's existing edge-compute footprint rather than a moonshot.

The global picture this sits inside is one of compute going where the users are, rather than the other way around. Microsoft has been building in India through its partnership with Adani; Google has expanded its Mumbai and Delhi regions; Amazon has been present in the country for nearly a decade. Meta is late to that particular table. The Reliance deal is the price of admission.

What remains genuinely uncertain is the technical specification. The Indian press coverage has not said what accelerator silicon the facility will run, and Meta has not disclosed it. That detail, more than the 168-megawatt headline, will determine whether the campus is competitive with Meta's existing Oregon and Texas builds or whether it sits a generation behind. Indian regulators have so far been permissive on accelerator imports, but the US export-control regime is the binding constraint, and Washington has been twitchy. The deal will be judged in 2027 not by what was announced in June 2026, but by what actually gets switched on.

Desk note: the wire read on this story has framed the Reliance partnership as a pure AI infrastructure announcement. The more honest read is that it is a hybrid: part compute, part Indian market positioning, part cover for a 2026–2027 capex cycle. Monexus is treating it as the third.