Google and Amazon Fund 1.1 GW of Nuclear Power Uprates in a Week

Two 20-year nuclear contracts landed in the space of a week. Amazon signed on 30 September 2026 and Google followed today, and together they fund roughly 1.1 GW of added output through nuclear power uprates at reactors that are already running.

The detail that matters most sits in the financing. Tech balance sheets are covering the cost of this new firm capacity, not utilities or ratepayers.

The timing has a cause. PJM Interconnection, the grid operator covering much of the US mid-Atlantic and Midwest, ran a capacity auction in July 2026 that cleared about 6.8 GW short of its reliability requirement. It was the first time the entire regional transmission organisation (RTO) fell short, and that gap is why bilateral deals like these now carry so much weight.

Here is what you need to know: how uprates work, what these two contracts add in real megawatts, why PJM’s shortage made them possible, and what they do and do not mean for your nuclear and uranium exposure.

Two deals, 1.1 GW: what Google and Amazon are actually buying

The deals compared

Set the two agreements side by side and the pattern becomes clear quickly.

Dividing capital by uprated megawatts (derived arithmetic, not a reported figure) gives about $4.8 million per MW for Google and about $15.8 million per MW for Amazon. That gap is not a pricing error. Amazon’s money covers upgrades across the whole 1,790 MW Calvert Cliffs plant, and its 20-year commitment gives Constellation the revenue certainty to extend the plant’s life by two more decades.

Much of Amazon’s money buys continued existing output rather than new megawatts. Read “capital committed” and “capacity added” as separate numbers.

The technology and workforce layer

The Google deal reaches well beyond the uprates. A separate 15-year supply agreement covers a further 2,700 MW from Constellation’s broader PJM fleet without being tied to specific plants, and it includes load-shaping and demand-response features that reduce consumption when the grid is under strain.

A five-year partnership will also put Google Cloud’s Gemini Enterprise to work across Constellation’s nuclear fleet. Planned uses include site selection, power flow modelling, permitting and interconnection planning, generation optimisation and infrastructure security.

The template argument Constellation CEO Joe Dominguez said the agreement could serve as a template for privately funded collaboration between tech and energy companies, with benefits that extend across the grid.

Constellation expects the deal to sustain about 4,400 existing jobs and create around 7,200 temporary construction roles. Investors responded quickly: CEG rose nearly 14% in morning trading today, after gaining about 2.7% overnight following the Amazon news.

How nuclear power uprates work, and why they have limits

An uprate raises the electricity an operating reactor produces while staying inside its licensed limits. No new reactor is built. Engineers replace or upgrade equipment so more of the reactor’s heat becomes power on the grid.

Constellation’s upgrades target three areas:

• Turbines, to convert more steam energy into electricity
• Steam generators, to improve heat transfer from the reactor
• Digital control systems, to run the plant more precisely and efficiently

The Nuclear Regulatory Commission (NRC), the US federal nuclear safety regulator, sorts uprates into three categories.

Historically, US uprates have come as small gains at many reactors over several decades, often alongside licence extensions and major component replacement. They have proved technically feasible, but they need close NRC oversight and heavy capital spending. Calvert Cliffs follows this pattern, with Constellation describing upgrades across the entire 1,790 MW plant.

The research behind this analysis did not include aggregate US uprate capacity or a typical cost per kilowatt, so no fleet-wide benchmark is offered here. The combined 1,080 MW is still small set against the overall US nuclear fleet.

The uprate category tells you how much of a project’s schedule and cost depends on NRC review and heavy equipment. That is where any delays would appear first.

Cost and schedule outcomes depend heavily on sequencing, and Energy Northwest cut its Columbia uprate estimate from $1.2 billion to $700 million by aligning capital modifications with refuelling outages already on the calendar.

Why PJM’s shortfall turned tech companies into nuclear financiers

Uprates are not new. Hyperscalers paying billions for them is new, and that shift traces back to PJM’s capacity market. The capacity market is the auction in which PJM pays generators to guarantee power availability years in advance.

The sequence ran as follows:

1. July 2026: The base auction for the 2028/29 delivery year cleared about 6.8 GW below the reliability requirement, at close to its cap of roughly $325/MW-day.
2. Backstop proposal: PJM’s Resource Backstop Program proposed a higher cap of $555/MW-day for a one-time emergency procurement.
3. 29 September 2026: The Federal Energy Regulatory Commission (FERC) approved the procurement only in part.
4. Aftermath: PJM postponed or scrapped the planned backstop auction.

PJM’s independent market monitor was blunt in its assessment:

The emergency backstop auction proposal was described as “deeply flawed” by PJM’s independent market monitor.

Reuters has connected this stress to rising interest in bilateral deals from large customers. Hyperscalers want firm, around-the-clock carbon-free power, and they can apply their credit strength to existing sites instead of greenfield builds. Constellation presented the Google deal as a direct response to PJM’s “Bring Your Own Power” proposal.

The fairness question remains open. Critics argue long-term contracts can privatise uprate benefits while other customers carry wider grid costs. Constellation says the deal is privately funded, benefits the whole grid and aligns with the White House ratepayer protection pledge, and no consumer-advocate responses to these specific deals were found.

If you pay power bills or follow utility costs, the change is this: access to new firm capacity is increasingly decided by contract rather than by auction.

What hyperscaler-funded uprates mean for nuclear and uranium investors

The case for a structural shift

Two readings compete:

• Structural: 20-year contracts, multi-billion-dollar retrofits, delivery timelines stretching to 2032 and a CEO describing the model as replicable all point to lasting change in nuclear economics.
• Cyclical: Commentary links the deals to the AI data-centre boom, and that demand could prove cyclical.

The contract length favours the structural reading. Investors do not commit to two decades of offtake for a temporary load spike, and persistent PJM shortfalls give large buyers a continuing reason to contract directly.

Risks and open questions

The risks are concrete but mostly unquantified:

• NRC review, equipment procurement and installation across multiple units could delay delivery.
• Revenue is concentrated among a small number of tech counterparties, which matters if AI or cloud spending slows.
• Cost-allocation and access disputes could invite regulatory pushback.
• Co-location arrangements still face open questions over interconnection and capacity markets.

Uranium is where the evidence is thinnest. Uprates modestly increase fuel requirements and extend plant lives, which supports long-term demand in principle. Roughly 1.1 GW is small against the global fleet, though, and no named analyst has linked these deals to fuel-cycle equities.

The evidence supports reading these deals as a shift for operators like Constellation more than a near-term catalyst for uranium producers. Weigh your exposure accordingly, and note that earlier hyperscaler nuclear precedents fell outside this research.

Earlier hyperscaler nuclear precedents exist outside the US, notably Google’s agreement funding a life extension at Finland’s Loviisa plant, which first started up in 1977.

Past performance does not guarantee future results. Forward-looking statements are speculative and subject to change based on market developments and company performance.

What these deals settle, and what remains open

The two contracts settle one question. Hyperscalers are willing to fund firm capacity from existing reactors on 20-year terms.

They do not close PJM’s gap. About 1.1 GW arriving between 2028 and 2032 is a small step against a 6.8 GW shortfall.

PJM’s reliability shortfall is driven largely by data centre load slated to connect in 2029 and 2030, which alone could exceed what Pennsylvania consumes today, and that scale explains why 1.1 GW of uprates cannot close the gap.

Four variables are worth watching:

• First uprate power, targeted for 2028
• NRC review progress across the units involved
• How FERC and PJM follow through on market design
• Whether more hyperscaler deals follow the template

For operator investors, the question is whether contracted revenue visibility justifies the execution risk. For ratepayers, it is whether contract-based access becomes the norm.

This article is for informational purposes only and should not be considered financial advice. Investors should conduct their own research and consult with financial professionals before making investment decisions.