Google and Constellation’s 3.59 GW Power Deal: Why AI Data Centers Are Repricing Nuclear Capacity

AI infrastructure is becoming a power-market story

The AI investment cycle is increasingly constrained by electricity rather than only by chips. Google and Constellation Energy made that shift unusually visible on October 6 with a long-term power agreement covering 3,590 megawatts across the PJM market, including 890 megawatts of new nuclear capacity created through upgrades to existing reactors.

The market immediately treated the announcement as a major earnings and valuation event for Constellation. CEG shares surged by double digits and became one of the strongest stocks in the S&P 500. Utilities also outperformed as investors extrapolated the deal into a broader thesis: hyperscalers are willing to sign very long contracts and support large capital investments to secure reliable power for data centers.

This matters because the AI trade is expanding beyond semiconductors. Power generators, nuclear operators, grid equipment vendors and transmission developers are becoming direct beneficiaries of the same capital cycle.

What the agreement actually includes

The first component is a 20-year power purchase agreement tied to 890 megawatts of new nuclear capacity. Constellation plans to create that capacity by upgrading 11 nuclear units in Illinois, Pennsylvania and New Jersey. The company said the program represents more than $4.3 billion of investment, with the first uprate expected to be delivered in 2028.

This is not a restart of a closed reactor and it is not a new greenfield nuclear plant. The strategy is to increase thermal and electrical efficiency at operating units. That distinction matters because uprates can add firm capacity faster than building a new plant from scratch, although they still require engineering work, regulatory approvals and capital.

The second component is a 15-year energy supply agreement covering an additional 2,700 megawatts in PJM. That power is not described as new nuclear capacity in the same way as the 890 megawatts, but it gives Constellation long-duration commercial support for existing generation.

Together, the agreements cover 3,590 megawatts. The companies also expanded a technology relationship in which Constellation will use Google Cloud and Gemini Enterprise for areas such as site selection, power-flow modeling, outage planning and operational optimization.

Why this is strategically important for Google

Google’s problem is not simply finding electricity. It is finding large amounts of electricity that are reliable, available on the timeline of data-center construction and compatible with corporate clean-energy goals.

AI training and inference loads are different from many traditional commercial loads. They can be large, concentrated and sensitive to interruptions. That increases the value of firm generation such as nuclear power.

By supporting uprates, Google is effectively helping bring incremental supply to the grid rather than only competing with existing customers for a fixed pool of electricity. That is politically important because rising data-center demand has created concern that residential and industrial customers could be forced to absorb higher grid costs.

The structure also gives Google a hedge against power scarcity. A 20-year agreement may look long compared with normal technology cycles, but data-center campuses and grid assets are long-lived. If AI computing demand remains strong, early access to firm power can become a competitive advantage.

Why this can change the valuation framework for Constellation

For Constellation, the agreement creates both growth and revenue visibility.

The 890-megawatt uprate program gives the company a path to expand output from existing nuclear assets without waiting for entirely new plants. The 2,700-megawatt supply contract supports the economics of the current fleet. Long-duration contracts can reduce exposure to short-term wholesale power-price swings and make future cash flows easier to underwrite.

The broader implication is that nuclear fleets may increasingly be valued not only as power plants but as scarce infrastructure capable of supporting data-center growth. That scarcity can improve contract terms, justify new capital spending and increase the strategic value of existing licenses, sites and grid connections.

The deal also reinforces a pattern already visible across the industry: large technology companies are signing directly with power producers instead of relying only on standard utility supply. This changes the negotiating balance between hyperscalers, generators and grid operators.

Market debate: structural scarcity or an expensive peak?

The bullish case is that U.S. power demand is entering a multi-year growth cycle after decades of relatively flat consumption. Data centers, manufacturing reshoring and electrification are all competing for generation and transmission capacity. Existing nuclear plants are attractive because they are already connected to the grid and can provide round-the-clock output.

The bearish case is that investors may be capitalizing very distant cash flows too aggressively. Nuclear uprates take time, capital costs can rise, and regulatory approvals can slow delivery. If AI efficiency improves faster than expected or data-center buildouts are delayed, some projected power demand could arrive later than current valuations assume.

There is also a customer concentration issue. A power producer that signs more long-term contracts with hyperscalers gains visibility, but it also increases exposure to a small group of giant buyers. Contract design, credit protection and cost-sharing therefore matter.

For Alphabet investors, the debate is different. Long-term power contracts can secure capacity, but they also commit capital and may raise the fixed-cost base of the AI buildout. The return depends on whether the compute funded by that power generates sufficiently high cloud, advertising and AI-product revenue.

Risks investors should track

The first risk is execution. Constellation must deliver reactor upgrades on schedule and budget. The first capacity is expected in 2028, so the market is paying today for benefits that will be realized over years.

The second risk is regulation. Nuclear uprates and grid changes require approvals and coordination. Delays could reduce the economic value of the contract.

The third risk is power-market politics. Data-center demand is becoming a consumer issue in regions where electricity prices are rising. The agreement is designed to show that new demand can bring new supply, but public scrutiny of large-load customers will remain intense.

The fourth risk is valuation. After a double-digit stock move, CEG investors need to separate the strategic quality of the deal from the price already embedded in the shares.

What to watch next

Watch the schedule for the first uprate in 2028, project-level capital spending, and any updates on regulatory milestones. Investors should also monitor whether Google and Constellation expand their strategic framework into additional generation, storage or demand-response projects.

A second signal will be whether other hyperscalers sign similarly structured agreements. If long-term contracts for new firm capacity become standard, the market may continue to revalue nuclear assets and grid-connected generation.

Finally, watch PJM capacity prices and transmission constraints. The tighter the grid becomes, the more valuable incremental firm megawatts are likely to be.

Conclusion

The Google–Constellation agreement is important because it turns AI electricity demand into a concrete, investable infrastructure contract. The 890 megawatts of new nuclear capacity and the additional 2,700-megawatt supply agreement show that hyperscalers are moving from general clean-energy targets toward direct, long-duration procurement of firm power.

For Constellation, that can support both growth and valuation. For Google, it is an attempt to secure a scarce input before data-center demand becomes even more competitive. And for the broader market, the message is clear: the next phase of the AI trade is increasingly about megawatts, grid access and long-lived energy assets, not only GPUs.