It was only a matter of time…
An AP story broke on Friday, October 24, 2025 on the behind-the-scenes negotiations between Santee Cooper and Brookfield Asset Management to resume construction of the husks of the two failed AP1000 units resting outside Columbia, South Carolina:
“Santee Cooper’s board agreed Friday to start six weeks of negotiations with Brookfield Asset Management that they hope will lead to a deal that lets the private company build the nuclear plants at the V.C. Summer site near Jenkinsville at their own risk to generate power that they could mostly sell to whom they want, such as energy-gobbling data centers.”
This is a smart move compared to funding data center construction directly. Even if the data center boom were to bust and turn to gravel in the mouth tomorrow, there would be a captive audience willing to buy 2,000 MW of nuclear-generated electricity for at least the next 80 years.
“Santee Cooper said Brookfield preliminarily agreed to provide the utility with some of the power generated. But that and probably thousands of other details will have to be negotiated. In a twist, Brookfield took over the assets of Westinghouse Electric Co., which had to declare bankruptcy because of difficulties building new nuclear reactors.”
This tells me Brookfield is keeping their options open. They’re looking for the highest bidder. Right now, they probably think that means hyperscale data centers. But it might not look that way in 5 years halfway through construction.
“After eight years in the elements, all the equipment and the structure of the plant, which was less than halfway finished, will need to be carefully inspected before it can be used. The permits to build and the licenses to operate the nuclear plants will need to be renewed, likely starting from scratch, Clements said.”
This signals one thing: millions in fees are about to be spent on specialty engineering design, and then billions more on construction. I’m sure Brookfield or those in its orbit will be combing the grid for nuclear electrical engineers who can pick up the details fast and hit the ground running.
“‘I still believe that the cost, technical and regulatory hurdles are too big to lead to completion of the project,’ Clements said, adding the agreement appears to let Brookfield walk away if it decide [sic] it’s not feasible.
Santee Cooper heard from 70 bidders and received 15 formal proposals to restart construction of the reactors. Interest in the project has grown as power demand in the U.S. surges with the increase in data centers as artificial intelligence technology develops.”
The regulatory hurdles of resuming construction can be overcome. Holtec has proven that with Palisades. The initial licensing and operating agreements of an AP1000 can be secured; Vogtle 3 and 4 proved that. Feasibility has been proven, and the merit of acquiring two 80-year, revenue-generation engines in the most political-friendly environment that commercial nuclear power has seen in 50 years was enough to bring 70 bidders to the table.
The article also points out that $9 billion was lost on the two failed VC Summer units, ending by pointing out the $17 billion cost overrun required to complete Vogtle Units 3 and 4.
Comparing Asset Class Economics
Let’s put these costs on the table and compare them to data center economics.
Vogtle 3 and 4 are estimated to have cost around $37 billion. That capital cost will be depreciated for between 60 and 80 years and generate over 2,000 MW annually all that time.
Contrast that with the recent project that Meta announced with Blue Owl Capital to build the “Hyperion” data center. That’s a $27 billion deal, and it’s projected to be complete by 2030.
Meanwhile, Brookfield released a document in August of this year explaining their investment philosophy in “building the backbone of AI.” In that document is a critical statement:
“Today, compute infrastructure, including GPUs, interconnects and supporting hardware, accounts for up to 50% of total capital expenditure in next-generation data center deployments.”
So, in Meta’s $27 billion deal, about $13 billion could go to chips (Nvidia) and networking gear. And what’s the depreciation period on that?
In January of 2025, Meta lengthened its hardware depreciation period to 5.5 years. Amazon had extended the depreciation period (which is a proxy for useful life) of its hardware to 6 years before recently reversing and reducing it to 5 years. Any computer gamer who has always been on the bleeding edge of graphics hardware might reduce the useful lifespan to 3 years, or possibly even less.
So, after roughly just 5 years of operation, Meta will be faced with the question of what to do about the $13 billion in aging, obsolete Nvidia chips running its AI servers. This is the same decision faced by every hyperscale data center built or currently under construction.
Jensen Huang’s eyes are seeing dollar signs, for sure. At least for as long as the music keeps playing.
But even if data center customers dry up, there will be a continual and growing demand for electricity in the United States. Having a 60-year, power-generating asset means you don’t have to be limited to selling to just one type of customer.
The economics of revenue and profit generation in a regulated utility market are well understood. The economics of generating profits from investing hundreds of billions of dollars in AI infrastructure have not yet materialized.
This is why Brookfield is interested in funding the completion of two partially constructed nuclear power plants.
And, should the AI bubble burst while the two AP1000s are under construction, then companies like Eaton who are selling their gear to data center customers faster than they can manufacture it will undoubtedly find themselves with an unexpected surplus and large, accumulating inventory.
Switchgear and electrical distribution equipment will go on sale big time. That might actually reduce Brookfield’s construction budget and sweeten the long-term ROI.