Google reportedly nears a $1 billion Constellation nuclear deal days after Amazon's 690 MW contract

Bloomberg's figure is "$1 billion or more", sourced to people familiar with the matter ^[1], and on its own it fits deals of very different sizes. Take Amazon's terms as the template: 690 MW for 20 years ^[10], with the reactor running every hour of every year. The plant would deliver about 120.9 million megawatt-hours over that term, and $1 billion spread across them comes to about $8.27 per megawatt-hour ^[17]. Put the same $1 billion on a 96 MW block, the size of Google's Georgia agreement ^[3], over the same 20 years and the figure becomes about $59.46 ^[18]. The gap between the two is the ratio of the volumes, roughly seven to one ^[16]. Duration, megawatts and source plants are not public ^[2]. Constellation declined to comment, and Google could not be reached ^[8].

Constellation's own history gives a second comparison. In January 2025 it announced a $1 billion deal to supply federal agencies through the General Services Administration over ten years ^[12], about $100 million a year ^[20]. Crypto Briefing describes that deal as a record ^[12]. Google's reported floor matches it ^[1].

Suppose Google's power comes from an operating plant named in the contract, as Calvert Cliffs is in Amazon's ^[10]. Then Google is buying from the same pool of existing reactor output as its rivals. Amazon, Google, Meta and Microsoft already hold roughly 8.5 GW of it ^[14], and Amazon's one contract is about 8% of that ^[21]. Suppose instead the money pays for added megawatts. Then Google is doing what it has done so far: the Georgia agreement adds capacity at Vogtle and Hatch ^[3], and the NextEra deal restarts a plant in Iowa ^[13]. A third outcome is that the deal changes or slips, since the account rests on unnamed people ^[1].

I think the first outcome is the likelier one, because Google's add-capacity route has produced small increments that are still pending. Its Georgia block is about a seventh of what Amazon took from one plant run by the largest US nuclear operator ^{[16] [9]}, and it still needs regulatory approval ^[3]. Crypto Briefing argues that Google's interest here is scale and reliability, since its current nuclear partners sell smaller increments ^[6]. It also notes that Google has mostly bought nuclear power elsewhere ^[5]. The counter-case is in the price. Over 20 years, $1 billion buys a 96 MW block at about $59 a megawatt-hour ^[18]. If so, Google is still buying in pieces near 100 MW, just from a bigger seller.

AI data centers need power around the clock ^[7]. A power purchase agreement locks a buyer into electricity from a specific source on agreed terms for years ^[15], and Amazon's runs for 20 ^[10]. For the buyer, that is a cost fixed for two decades. For the seller, it is the basis for spending: Constellation expects the Amazon contract to unlock more than $3 billion of infrastructure investment in Maryland ^[11]. A Google contract that names an operating plant and a volume near Amazon's would support the view that hyperscalers now compete for a limited stock of existing nuclear output. A volume near 96 MW, or money aimed at uprates, would count against it ^[3].