Nscale and Anthropic discuss a proposed $45 billion compute deal

Bloomberg reported that Nscale and Anthropic are negotiating a deal under which Nscale would sell computing power to Anthropic. People familiar with the negotiations put the proposed deal’s potential value at about $45 billion.[1] The agreement is not final: TechCrunch reported that it had not been finalized and that details could change.[2] For teams assessing the announcement, that distinction matters. The reported value describes a possible agreement, not a completed contract or confirmed spending commitment.[1][2]

The new talks come after Nscale and Anthropic announced another $45 billion compute deal a little over a month before the report.[2] TechCrunch said that, if the proposed deal goes through, it would be the latest in a long string of giant compute deals Anthropic struck over the past year.[2] The evidence does not give the proposed deal’s specific capacity, delivery schedule, or duration, so the headline value alone does not show how much computing power would be delivered or when.[1][2]

The talks are unfolding amid investor concerns that companies are spending too much money for too little profit, TechCrunch reported.[2] At the time of its report, both Nscale and Anthropic were pursuing IPOs.[2] TechCrunch also reported investor pushback over a proposed $30 billion valuation for Nscale.[2] Nscale said it had not determined the number of shares or price range for its proposed IPO.[3] These details provide context for evaluating the proposed deal, but do not establish how investors or IPO plans will respond to it.

Nscale announced a $3.36 billion pre-IPO convertible financing on September 25, 2026, and said the round was led by Third Point, with NVIDIA among the new and existing investors supporting it.[4] Nscale said its platform had more than $103 billion in total contracted value and described its offering as combining software, computing capacity, and power.[4] Separately, the company signed a letter of intent with Microsoft for 1.35 gigawatts of AI compute capacity at its Monarch campus in West Virginia, with deliveries planned in tranches beginning in late 2027.[5]