JAKARTA, 11 AUG 2026 — CoreWeave is coming to Indonesia with three data centres and 360 megawatts of contracted power. The company's first move into Asia-Pacific hinges on a single sentence about ownership.
"CoreWeave will own and operate the compute environment across all three sites," the company said on 4 August. The buildings will sit in Indonesia. The machines inside them will not be Indonesia's.
What was announced
Three facilities, 360MW of contracted IT power, expected online in 2028. CoreWeave says it will build and train a local team to run the sites and to support technical talent development in the country. Sachin Jain, the chief operating officer, framed the choice around the market rather than the incentives: "Indonesia is investing heavily in AI education and digital capacity, and it's paying off."
The omissions are as telling as the details. CoreWeave has not disclosed where the three sites will be, how capacity splits between them, or what it is spending. Reuters reported the company expects to spend billions of dollars without being able to give an estimate.
So the verifiable content is a number, a year and an ownership structure. Everything else is a plan.
How big 360MW is for CoreWeave
It helps to put the figure against the company's own book. As of 31 March 2026 CoreWeave operated 49 data centres with a gigawatt of active power and 3.5GW under contract, and it has said it expects to pass 1.7GW active by the end of this year.
Roughly a tenth of contracted capacity, arriving two years out. For an entire region's first allocation from a company of this size, that is a position rather than a commitment — enough to be present, not enough to be dependent.
"Contracted" is doing a lot of work in that sentence
The 360MW is contracted power, not delivered power, and CoreWeave's own book shows how far apart those can be. On 31 March the company held 3.5GW under contract and had one gigawatt actually running. Contracted capacity was three and a half times what was energised.
That gap is normal in the industry and it is not evidence of anything going wrong. Power is contracted years ahead because it has to be. But it does mean a contracted megawatt is a claim on the future rather than a machine that exists, and a 2028 date sits at the far end of a capital cycle nobody can currently forecast.
Two things could change the number before then. Demand for AI compute could soften, in which case contracted capacity is the first thing renegotiated. Or it could hold, in which case 360MW may look modest and get revised upwards. Either way, the figure announced in 2026 is not the figure that will be switched on.
The ownership question is the regional one
Southeast Asian governments have spent three years competing to host AI infrastructure, and the competition is usually framed as attracting investment. Land, power, tax treatment, and a queue of hyperscalers.
Hosting and owning are different things, and the gap between them is where the economics live. A country that hosts 360MW gets construction, a power offtake agreement, property tax, a few hundred operating jobs and the training CoreWeave has promised. A country whose firms own compute gets to decide what runs on it, at what price, and for whom.
Indonesia is getting the first. That is not a criticism of the deal — it is what nearly every data centre announcement in the region has meant, including the ones in Singapore and Malaysia — but the language of "digital sovereignty" that surrounds these projects tends to obscure it.
The one component that does transfer is operational skill. CoreWeave says it will build and train the team that runs the sites; running GPU fleets at this scale is a capability the region lacks. Whether that produces engineers who could later build something of their own, or a staffing arrangement that ends when the contract does, is not decided by anything in the announcement.
Power is the constraint that decides this
360MW is a substantial block of generation in any market. It is the part of the plan least under CoreWeave's control and most under Indonesia's, because the contracted power has to come from somewhere and be delivered reliably to a load that does not tolerate interruption.
This is where regional data centre projects usually slip. A site announced with a 2028 date is depending on grid connection timelines, and across Southeast Asia those have proved a harder constraint than either capital or demand.
Nothing in the announcement addresses where the power comes from or how it will be generated, and until that is public the 2028 date is an aspiration with a number attached.
What to watch
The site locations, first. Indonesia's data centre capacity has concentrated around Jakarta and Batam, and Batam's appeal has always been proximity to Singapore's demand rather than Indonesia's own. Three sites near Batam would describe a different project from three sites spread across Java.
Then the power arrangements, because they are the schedule. And then whether any Indonesian institution — a state enterprise, a bank, a university consortium — ends up contracting capacity in volume, which is the only mechanism by which hosting turns into something more durable.
The announcement was written for CoreWeave's investors: a new region, a large number, and a date far enough out to be safe. Read from Jakarta it is a different document, and most of what Indonesia would want to know is not in it.
