CoreWeave's 240 MW India Plan Is a Target, Not a Running Site
CoreWeave plans to build a 240 megawatt (MW) AI computing site in Navi Mumbai with AdaniConneX, its entry into India. The announcement sits in the company's newsroom alongside a set of other capacity and customer items. Some of those describe finished steps and others describe intent. For anyone tracking AI infrastructure, that difference matters more than the headline figure.
A megawatt is a unit of electrical power. It measures how much electricity a facility can draw, which makes it a practical yardstick for the size of a data center, the buildings full of servers that run AI.
On capacity, CoreWeave says it has brought additional AI computing capacity online in Europe, extending its high-performance global network. It also says it has extended its global buildout into Southeast Asia with new facilities. Both describe work that has already happened. The India deployment is different. It is a plan, and a megawatt figure is the easiest milestone to check later. Those checkpoints are construction, energization (the point where a facility is switched on and drawing power) and customer load. For a 240 MW site, the operative questions are power procurement, grid connection timing and how quickly the capacity is contracted once it exists.
The customer announcements are more varied. Wayve will train its end-to-end foundation models (large AI models that learn a whole task, such as driving, in one system) on CoreWeave Cloud to accelerate commercial deployment for autonomous vehicles. That is a stated plan, not a running workload. IMC will deploy CoreWeave's AI cloud platform to power its next generation of research infrastructure, also a forward-looking statement. Rescale, by contrast, has expanded to CoreWeave Cloud to provide infrastructure for AI, HPC (high-performance computing, the large-scale calculations used in science and engineering) and simulation workloads. It is the one customer item phrased as a completed step.
The Leidos collaboration carries a different structure. CoreWeave supplies its AI-native platform, while Leidos leads mission integration to deliver AI capacity for national security. The split gives compute delivery to one party and the mission layer to another. For an investor or counterparty, revenue, integration risk and contract terms sit with different organizations. A reader should not assume the cloud provider controls the full chain from hardware to deployed application.
The broader context here is the spread of end markets. Autonomous driving model training, research computing, HPC and simulation, and national security work have different usage patterns, procurement cycles and tolerance for downtime. In principle, a provider serving all of them depends less on any single buyer category. That only holds if the planned deployments and collaborations turn into contracted, running capacity.
In my view, the most useful way to read a batch like this is as a schedule of tests, not a set of results. The India plan will be judged on whether the 240 MW arrives on time and at what cost in power. The European and Southeast Asian additions will be judged on how fully they are used. The customer programs will be judged on whether training runs, research clusters and simulation pipelines move from announcement to sustained load. General experience with large capacity projects suggests that power supply and construction timelines are the usual sources of slippage, and that customer ramps are often slower than the first press release implies. None of that is specific to CoreWeave.
Taken together, the items describe a company extending its physical footprint into new markets while signing workloads across several sectors. What they do not yet show is how much of that capacity is energized, how it is priced, or when it will carry revenue. Those numbers will decide whether the breadth reads as strength or as a long list of intentions.
