Google's 890 MW Nuclear Deal Builds No New Reactor

Google says its new deal with Constellation Energy adds the equivalent of a new nuclear reactor to the grid that serves the Mid Atlantic and parts of the Midwest. No reactor will be built. Instead, 11 existing reactor units in Illinois, Pennsylvania and New Jersey will get new turbines, steam generators and digital controls, and together they will squeeze out 890 megawatts more than they produce today. It is one of the least glamorous ways to make clean electricity, and right now one of the fastest.

What was announced

The agreement, announced October 6, 2026, has two parts that are easy to blur together. The first is a 20 year power purchase agreement that funds upgrades at the 11 Constellation units, adding 890 MW of capacity to the PJM Interconnection grid. Constellation puts the investment at more than $4.3 billion and expects the first uprate to be delivered by 2028.

The second is a 15 year energy supply agreement for 2,700 MW from Constellation's existing fleet in PJM. That power already exists. The companies describe it as revenue certainty that keeps operating plants economically viable, which is valuable, but it does not add a single new megawatt to the grid.

So the headline total of roughly 3.6 gigawatts includes about 2.7 gigawatts that were already running. The new supply is the 890 MW. That is still a lot: Google compares it to a large conventional reactor or several small modular reactors.

What an uprate actually is

The Nuclear Regulatory Commission licenses every commercial reactor to run at a maximum power level, and that number feeds the safety analyses for the plant. Raising it is called a power uprate, and it requires NRC approval of a license change. Physically, it can mean replacing or modernizing turbines, steam generators, instruments and control systems so the plant converts more of its heat into electricity, or safely runs hotter.

The appeal is speed. Google's announcement argues that uprates avoid the long timelines and interconnection bottlenecks of building on an empty site. The plant, the transmission connection and the trained workforce already exist. The upgrade happens inside a fence that has been licensed for decades.

The cheapest new reactor is sometimes a few hundred extra megawatts hiding inside the old one.

How big is 890 MW, really

The NRC offers the best yardstick. Over the past 20 years, the agency says, approved uprates have added about six gigawatts to the national grid, the equivalent of six large reactors. It also says the industry is interested in requesting more uprates that could add another two gigawatts in the coming years.

Set against that, one corporate deal covering 890 MW is a large slice of what regulators expect the entire fleet to pursue. That comparison is mine, built from the two published numbers, and the categories are not perfectly matched, since the NRC figure counts approved license changes and Google's counts planned capacity. Still, it shows the scale of what a single buyer can now pay for.

The grid behind it is big too. PJM, the regional operator, serves 67 million people, according to Google. The deal is framed as a response to PJM's "Bring Your Own Power" proposal, the idea that large new consumers, such as data centers, should bring new supply with them instead of drawing down capacity everyone else relies on.

What to watch

Two things will decide whether the 890 MW shows up on time. The first is the NRC. Every uprate needs its own license amendment, and the agency says it is reviewing past cases to process new requests as efficiently as possible while maintaining safety. The second is execution: the companies give a date only for the first uprate, by 2028, and no public schedule for the other units. Until those dates exist, the 890 MW is a commitment, not a delivery.

The deal also includes a demand flexibility clause, under which Google's consumption can be shaped and non critical load curtailed during stressed grid events. That detail gets less attention than the megawatts, and it may matter as much. A data center that can step back on the hottest afternoon of the year is worth more to the grid than one that cannot, whatever it buys.

The AI power story is usually told as a race to build: new reactors, new gas plants, new transmission. This deal is a reminder that some of the fastest capacity was already standing, in plants that were simply never pushed to their limit. The question for the next few years is how many of those hidden megawatts remain, and who gets to pay for them first.