CoreWeave plans to enter India with 240 megawatts (MW) of AI data-centre capacity at AdaniConneX’s Taloja campus in Navi Mumbai. The US AI cloud company says the project will occupy three 80 MW buildings, with the first phase targeted for mid-2028. The 240 MW is a planned deployment, not capacity already operating today. CoreWeave’s announcement, dated October 6 in the US, became an India news story on October 7.
The distinction between a power commitment and a working AI cloud is the story’s central test. CoreWeave can sign for a large block of capacity now, but customers will only be able to use it after the buildings, electrical systems, cooling, computing equipment and network are commissioned. That long construction runway makes this an infrastructure and execution story, rather than a simple claim that India suddenly has another 240 MW of live AI compute.
Key takeaways
- CoreWeave says it will be the sole tenant in three planned 80 MW buildings at Taloja, giving it an initial 240 MW allocation.
- An opportunity to expand by a further 240 MW could take the campus allocation to 480 MW. The extra capacity is an option, not a completed deployment.
- The first phase is expected in mid-2028; CoreWeave has not published a full commissioning schedule for all three buildings.
- The company intends to deploy NVIDIA’s Vera Rubin platform and open an India office, but has not disclosed a final investment amount or customer list for the site.
- The practical question is how fast power, cooling, chips and customer contracts can turn the announced capacity into usable services.
What CoreWeave and AdaniConneX actually announced
CoreWeave, a US-based provider of specialised cloud infrastructure for artificial intelligence, said it would collaborate with AdaniConneX, the Indian data-centre joint venture of Adani Enterprises and EdgeConneX. Its first India deployment is planned at the Taloja campus in Navi Mumbai. The three buildings are each described as 80 MW, and CoreWeave would occupy all three as the sole tenant. These are company plans, so the opening date and final configuration may change as construction proceeds.
The company also said it would establish an Indian office and hire locally. That matters commercially because the project is not just an overseas server lease: CoreWeave wants a customer-facing presence near Indian developers and enterprises. Yet neither headcount, committed customer contracts nor an India price list was provided in the announcement. Readers should therefore treat market-size claims as expectations, not booked demand.
Bloomberg’s original report, carried by NDTV Profit, said the investment over the project’s life could run to multiple billions of dollars. CoreWeave did not quantify that figure in its own release. It is therefore a reported estimate rather than an announced capital budget. The Next Web separately covered the project’s location, sole-tenant structure and mid-2028 target, while Crypto Briefing reported the option to scale the allocation to 480 MW. These are three distinct publisher accounts, not three copies of one syndicated wire.
CoreWeave India capacity: commitment, option and delivery
The headline number has three different layers. First, the initial plan is 240 MW across three buildings. Second, CoreWeave has an opportunity to add up to another 240 MW at the campus. Third, the first phase is only expected to become available in mid-2028, with subsequent capacity arriving in phases. Reporting 480 MW as already secured and online would collapse those layers into an inaccurate single figure.
Even the initial 240 MW is best read as a future infrastructure allocation. Data-centre announcements often use megawatts to describe site or facility power rather than the electricity actually consumed by installed IT equipment. The available statement does not provide enough detail to convert the allocation into a count of GPUs or a precise number of model-training jobs. Any such conversion would require assumptions about power usage effectiveness, chip generation, networking and utilisation.
Why the site needs more than a power contract
An AI data centre must bring several systems online together. The building shell and utility connection are only the beginning. High-density GPU rooms need distribution equipment capable of carrying large electrical loads and cooling systems that can handle the heat produced by accelerators. They also need network links between machines and to customers. A delay in any one part can push out the date when capacity can be sold as a cloud service.
CoreWeave says Taloja is being designed for direct liquid cooling in GPU data halls and a non-evaporative chilled-water system. That is a design intention reported by the vendor; independent operating data for the future campus does not yet exist. The system could reduce water consumption compared with evaporative approaches, but the release does not publish audited site-level water or energy efficiency figures. It would be premature to claim a verified environmental benefit from the design description alone.
The company plans to deploy NVIDIA’s Vera Rubin platform. This connects the real-estate commitment to a forthcoming generation of AI hardware, but it also introduces a dependency: the platform, associated networking, rack integration and site commissioning must all line up. A promised chip family should not be mistaken for a published delivery schedule or guaranteed customer performance. For context on how India’s infrastructure suppliers are positioning around the same generation, see our coverage of Yotta’s planned Vera Rubin procurement.
The AdaniConneX role and the India angle
AdaniConneX supplies the physical campus where CoreWeave intends to operate its AI cloud. The arrangement separates property, power and building infrastructure from the cloud platform and customer service layer. That division can help a foreign cloud specialist enter a market without owning every part of a data-centre project, although contract economics, ownership of equipment and financing obligations have not been disclosed publicly.
Navi Mumbai is a logical site because it sits within India’s largest established data-centre corridor and close to corporate demand, network infrastructure and the Mumbai financial market. Yet geographic convenience alone does not guarantee low latency for every Indian customer. Service quality will depend on interconnection, routing, availability zones and where users and data actually sit. CoreWeave has not yet published an India-region service catalogue or service-level commitment for this campus.
The partnership is also a meaningful follow-on to AdaniConneX’s broader buildout. Lapaas Voice recently examined Adani’s capital spending on data centres. That earlier story concerns infrastructure investment; this announcement identifies a prospective major AI cloud tenant at one campus. The connection is commercial, but the two stories should not be conflated into proof that all prior capital spending has become productive.
How the India rollout compares with other AI infrastructure
India’s AI infrastructure market contains several different models. Some firms own cloud platforms and lease data-centre buildings. Others operate the real estate and sell power, cooling and connectivity. Some buy GPUs to offer compute as a service. Hyperscalers often operate integrated cloud regions. CoreWeave’s planned Taloja site appears closest to a specialist AI cloud hosted inside a partner’s purpose-built facilities.
That distinction matters to Indian customers evaluating alternatives. A startup training a model may care about accelerator availability and software tooling. An enterprise running inference may care more about predictable latency, data location and long-term service reliability. A government customer may require additional residency and procurement assurances. None of those customer-specific outcomes can be inferred merely from a 240 MW announcement.
India’s existing cloud buildout also provides context. Our report on Microsoft’s Hyderabad AI-ready cloud region concerns facilities already brought into service and described with availability zones. CoreWeave’s Taloja plan is at an earlier stage. The comparison shows why capacity announcements, live regions and customer-ready AI services should be counted separately.
What Bloomberg’s investment estimate does and does not tell us
Bloomberg reported that investment could total multiple billions of dollars over the life of the project. The wording matters: it is a lifetime estimate, not a disclosed up-front payment by CoreWeave. The report does not provide a complete split among data-centre construction, utility works, chips, networking, financing or partner contributions. Investors and customers should resist comparing that estimate directly with one company’s reported quarterly capital expenditure.
The first-party release provides the stronger evidence for the specific site design and timetable. Bloomberg’s report adds a journalist-sourced investment perspective. The Next Web and Crypto Briefing corroborate the core development from their own publications but do not turn an undisclosed contract into a public one. Together, the sources support a story about a large planned India deployment; they do not support claims of a completed campus, signed local customers or a precise return on investment.
The questions that will determine whether the plan succeeds
The first milestone is physical: site works, utility connections and cooling systems must be delivered on time. The second is technical: the planned NVIDIA hardware, networking and CoreWeave software stack must be installed and made reliable at scale. The third is commercial: enough customers must sign contracts at prices that justify the capital and operating cost of a very large AI facility.
Power availability is particularly important. Announced megawatts are valuable only if the relevant grid, substation and backup systems can serve the buildings when they open. The original release describes the project design but does not provide an audited energisation certificate or a construction milestone for each building. Readers should watch for those documents and for an actual service launch before describing any portion of the 240 MW as operational.
Customer demand is another open question. India has active AI developers and enterprises, but their workloads may differ in model size, procurement budgets and data-residency needs. CoreWeave’s local office could help sell to that market, yet the announcement names no anchor Indian customer for Taloja. A sole-tenant building arrangement does not by itself prove final-user utilisation of the compute that may eventually be installed there.
Frequently asked questions
Is CoreWeave already operating a 240 MW AI data centre in India?
No. CoreWeave has announced a planned 240 MW deployment at AdaniConneX’s Taloja campus. The company expects the first phase to become available in mid-2028 and has not said that the full allocation is live today.
Could the CoreWeave India project reach 480 MW?
CoreWeave says it has an opportunity to add up to 240 MW beyond its initial 240 MW plan. That would bring its potential campus allocation to 480 MW, but the additional capacity is an option and has no published commissioning date.
How much will the CoreWeave–AdaniConneX project cost?
CoreWeave did not publish a specific budget in its announcement. Bloomberg reported that investment may run to multiple billions of dollars over the project’s lifetime; that is an attributed media estimate, not a detailed spending commitment from the company.
Source and measurement note
This article uses CoreWeave’s dated company announcement as the primary record and checks the development against original reports by Bloomberg via NDTV Profit, The Next Web and Crypto Briefing. The company release is dated October 6 in the US; Indian publisher coverage appeared October 7. MW figures describe planned data-centre capacity, not measured live IT demand or GPU count. Vendor design and timetable claims remain attributed to CoreWeave.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.



