AI's next bottleneck is electricity and real estate

What did Meta and Reliance announce?

Meta agreed to lease a 168 megawatt AI data centre that Reliance Industries will build in Jamnagar, Gujarat, with the announcement landing on 9 June in the United States and 10 June Australian time. Jamnagar becomes Meta's first AI-enabled facility in India, leased with an option to scale, and delivery is expected within two years. Reliance acts as the single provider across design, construction, utilities, renewable power supply and network connectivity, while Meta covers the full cost of the power and water.

"This world-class facility in Jamnagar will help us scale our AI infrastructure globally while deepening our long-term investment in India's economy," Mark Zuckerberg said in the announcement. Meta frames the site as part of its push toward what it calls personal superintelligence.

Why has power become the constraint on AI?

Electricity explains the shape of the deal. Meta separately contracted nearly 1 gigawatt of clean energy across India through CleanMax, at 837 megawatts across Rajasthan and Karnataka, and Fourth Partner Energy, at 88 megawatts spread over Tamil Nadu, Karnataka, Maharashtra and Uttar Pradesh. The Jamnagar site itself runs entirely on renewable power and cools with desalinated seawater, which folds water supply into the equation alongside land and generation capacity.

Compute now goes where the electrons and the coastline are, and the algorithms follow the infrastructure rather than the other way around. Leasing built-to-suit capacity from an energy conglomerate also moves faster than building alone, since Reliance brings the land, the grid connections and the approvals that a foreign hyperscaler would otherwise spend years assembling.

What does Jamnagar signal about where capacity goes?

Jamnagar closes a six-year arc. Meta put 5.7 billion US dollars into Jio Platforms in 2020, formed a joint venture with Reliance in 2025 to sell enterprise AI built on Llama models, and has now added the compute layer that both earlier deals run on. India supplies one of Meta's largest and fastest-growing user bases, and local infrastructure simplifies compliance as Indian regulators sharpen their scrutiny of how foreign platforms handle data.

Reports trace early talks to 2024, when Meta explored leasing capacity in Chennai before the scope outgrew the plan, and the 2025 joint venture, roughly 70 per cent owned by Reliance, sells enterprise AI built on Meta's Llama models to Indian businesses. Meta's Project Waterworth, the 50,000 kilometre subsea cable announced in February 2025, ties the same map together by linking the United States, India, Brazil and South Africa.

What does the capacity race mean for AI search?

A hyperscaler does not sign a built-to-suit lease on another continent, with dedicated power contracts and seawater cooling, for a product line it expects to fade. The capacity race is the clearest available evidence that AI assistants are infrastructure now, funded and housed like power stations rather than features. TechCrunch reported the companies have not disclosed the value of the agreement or the workloads the site will run, and the silence on price says plenty about how routine deals of this size have become.

For AI search, capacity is destiny. Every megawatt of new inference capacity makes assistants faster and cheaper to run inside more products, which expands the surface area where brands are recommended, cited or skipped. ChatGPT already counts 900 million weekly users, the Gemini app reached 900 million monthly, and AI Overviews touch 2.5 billion people a month, and the data centres being signed in 2026 are what carries those numbers higher. Google's I/O announcements push from the demand side at the same time, since information agents that monitor the web around the clock for a billion AI Mode users consume standing compute rather than per-query bursts. Our June briefing on the fragmenting answer market covers the platform numbers in detail.

What should Australian brands take from it?

Australian brands should read the infrastructure news as a permanence signal. Citation positions inside AI answers are still cheap to win in most Australian categories, and the buildout guarantees the audience on the other side of those answers keeps growing. The practical sequence sits in our guide to which AI platforms to optimise for first, and the paid side is arriving in parallel through ChatGPT's advertising expansion into Australia. Optimising early against a growing surface beats bidding late against a crowded one, and the companies pouring concrete in Gujarat have already made their version of that call.