CoreWeave leases 240MW in Navi Mumbai to put Vera Rubin GPUs in India from mid-2028

Picture an AI team in India renting CoreWeave GPUs this quarter. Those GPUs sit in another country, because CoreWeave has never had a data centre in India ^[1]. Chief executive Mike Intrator said last month that the company is "largely sold out" of capacity, Bloomberg reported ^[16]. The Navi Mumbai lease gives that team a local option, starting roughly 20 months after the October 6 announcement ^[2][3][21].

The pitch came from Sachin Jain, CoreWeave's chief operating officer. "We are investing in the infrastructure, technology, and local presence needed to support that growth, bringing 240MW of capacity, Nvidia Vera Rubin, and CoreWeave's expertise to customers building and scaling AI in India," Jain said ^[9]. What is actually under way, per the company, is a lease plus an Indian office and local hiring ^[11]. CoreWeave linked the investment to the government's IndiaAI Mission ^[11].

Measured against India, the lease is large. Its 240MW is about 11% of the 2.2GW of operational capacity the country built over three decades ^[13][20]. Wood Mackenzie forecasts 12GW by 2030 ^[13]. It also projects AI-specific capacity to grow almost 24-fold to 6.5GW ^[14]. Working back from that multiple puts today's AI-specific base near 270MW, close to the size of this single lease ^[19].

The reports give different figures for how far the campus can grow. TNW reported an option to add up to another 240MW ^[22]. DatacenterDynamics described the option as doubling capacity, for a potential total of 580MW ^[23]. Doubling 240MW gives 480MW, leaving 100MW between the two accounts ^[24].

AdaniConneX, a 50:50 joint venture of Adani Enterprises and EdgeConneX ^[4], is also one of the companies working on Google's India AI hub. The Guardian reported that hub has been cleared for 2.51GW ^[15]. Adani said early this year that it was targeting 5GW of data centre capacity in India ^[6].

In my view Navi Mumbai belongs in a capacity plan as a 2028 line item. The trade-off is planning around a site with one dated phase and an expansion figure the two reports disagree on ^[3][24]. Two questions sort the rest. First, whether the workload has to run inside India under rules or contracts the team already lives with. Second, whether it can wait for mid-2028 and a phased ramp.

A workload that must stay in India and can wait goes on the Navi Mumbai shortlist. Phase dates and the option size are the first things to get from CoreWeave in writing. One that must stay and cannot wait gets nothing from this lease, and its choices are limited to the 2.2GW already operating ^[13]. Work that can run anywhere and can wait gains one more region to compare on price. For work that can run anywhere and cannot wait, the plan stays as it was.