Anthropic has reportedly agreed to purchase approximately $10 billion worth of compute capacity over six years from Volta, a data center developer, according to reporting from TechCrunch on August 4 that was also covered by Bloomberg. The deal centers on a data center in Norway, reported to be linked to Bitdeer, with a capacity of roughly 133 megawatts, and is expected to be built around Nvidia's upcoming Vera Rubin platform. Neither Anthropic nor Volta has confirmed the exact terms on the record.
What the reporting says, and what it doesn't
TechCrunch's sourcing puts the commitment at roughly $10 billion over about six years, with Anthropic buying dedicated capacity out of a single Norwegian facility rather than owning or building the data center itself. Bloomberg's coverage lines up with the same broad strokes. Until one of the parties speaks, those numbers stay in the reported column, large, directionally consistent across two outlets, and still soft enough that they could shift before construction wraps.
The site itself is reported at around 133 megawatts, which is a serious amount of power for one facility, big enough to sit in the category of purpose-built AI infrastructure rather than a converted general-purpose data hall. The facility is reportedly built around Vera Rubin, Nvidia's next architecture after Blackwell. Locking in capacity around a platform that isn't broadly available yet has become a familiar move among frontier labs: get in line early for scarce next-generation silicon rather than wait for it to become available to everyone at once.
The reporting describes the facility as linked to Bitdeer, a name that might ring a bell from crypto mining rather than AI. Bitdeer has been shifting parts of its infrastructure business toward AI compute hosting as crypto-mining economics have gotten less favorable in some of its markets. Exactly how Volta and Bitdeer relate on this specific site isn't spelled out in what's been reported so far, so I won't guess at the ownership structure. "Linked to" is doing a lot of work in that sentence, and I've seen coverage flatten it into something firmer than the sourcing supports.
If you're keeping a spreadsheet of Anthropic's compute commitments, pencil this one in with a question mark on every column: dollars, years, megawatts, chip generation, and counterparty. Two reputable outlets saying roughly the same thing is more than rumor-mill noise. It still isn't a signed term sheet on Anthropic's letterhead.
Volta is young, and Norway keeps showing up
Volta was only founded earlier this year. It's listed as an Nvidia Cloud Partner, the designation Nvidia hands out to infrastructure providers it works closely with on deploying its newest hardware, a signal of access, not necessarily of scale. If the reported terms hold up, this is a remarkably large early contract for a company barely a year old, and it says something about how much capital is chasing AI data center capacity right now that a startup this new can land a commitment this size.
Nordic sites have become a recurring destination for AI compute buildouts over the past couple of years, mostly for reasons that have nothing to do with AI itself: cold climates cut cooling costs, and the region has relatively abundant renewable power for its population size. Neither TechCrunch nor Bloomberg spells out why this particular site landed in Norway, but it's the same logic that's driven a string of similar announcements elsewhere in Scandinavia.
Nothing in the reporting suggests Anthropic is putting all its eggs in one Norwegian basket. The deal is framed as part of a push to line up compute from multiple providers, which tracks with how every other frontier lab has approached the same problem, spread the dependency across enough vendors that no single data center operator or cloud provider becomes a chokepoint.
I've stopped being surprised when a company you've never heard of shows up attached to a ten-figure AI infrastructure number. The capital is chasing floor space and power as hard as it's chasing model talent. Volta's age is still striking: landing something on this scale before most startups have finished furnishing an office is either a testament to how hungry Anthropic is for dedicated capacity, or a reminder that "reportedly" covers a lot of ground between a signed commitment and a conversation that got ahead of itself.
What nobody has answered yet
How Anthropic intends to use 133 megawatts of Vera Rubin capacity once it's live, training runs on future Claude models, inference capacity for existing demand, or some mix of both, isn't in the coverage. That answer would tell you a lot about where Anthropic thinks its bottleneck is right now. Until Anthropic says more, it's an open question.
Compute deals at this scale have a way of getting adjusted before construction even wraps, so I'd treat the six-year framing as a starting point rather than a locked-in contract. The $10 billion figure and the term should both be read as reported numbers, not confirmed ones, until one of the parties says otherwise.
Anthropic isn't alone in chasing dedicated capacity outside the traditional big three cloud providers. As more of these bespoke, single-tenant deals get struck with newer developers like Volta, the traditional hyperscalers may respond by sweetening their own long-term agreements, or simply cede this particular segment to developers willing to build to a single enormous customer's specifications. Either way, this deal is one more data point in a reshuffling of who actually owns and operates the physical infrastructure behind frontier AI, even if the specific numbers shift once official confirmation arrives, if it ever does on the record.
For readers trying to track Anthropic's compute story specifically, file this next to whatever else the company has lined up with AWS, Google, and anyone else in the queue, not as a replacement for those relationships, and not as proof that Norway is suddenly Anthropic's primary home for training. It's one reported facility, one reported dollar figure, one reported chip bet. Large enough to matter. Incomplete enough that the next disclosure could rewrite half the details.
How I'd file this until someone confirms it
My working label for this story is "credible, incomplete." TechCrunch and Bloomberg both describing a roughly $10 billion, six-year Anthropic commitment to Volta's Norway site is enough to take seriously. It's not enough to treat the megawatt figure, the Bitdeer link, or the Vera Rubin bet as locked.
I've covered enough infrastructure announcements that never quite matched the eventual contract to keep a soft pencil on deals like this. Construction timelines slip. Chip roadmaps slip. Sometimes the "deal" was a memorandum that never converted. Sometimes it converts and then quietly gets resized.
What I do take as directionally solid is Anthropic hunting dedicated capacity outside a pure hyperscaler relationship, and Norway remaining a magnet for AI power densification. If Volta's facility shows up in a later Anthropic filing or an on-the-record confirmation, we can harden the numbers. Until then, this stays in the reported column where it belongs.
- Edge Computing
- Funding
