AI power stocks: who has actually signed a data center deal

AI power stocks are the public companies paid to generate, deliver or equip the electricity AI data centers run on. The clearest exposure belongs to suppliers with a signed deal with an AI buyer. Supercycle's Buildout Tracker records 52 power deals, and 22 of the suppliers are public companies. Constellation, Vistra, Talen, Oklo and Bloom Energy are among them.

As of October 8, 2026, power and utilities are also the most fearful corner of the AI trade by Supercycle's AI Fear & Greed Index. The part that usually gets skipped is that a signed contract and a sector label are different kinds of exposure. Most "AI power stocks" lists mix the two. This post separates them.

Which power companies have signed deals with AI buyers?

These are the tracked deals where the supplier is a public company and the buyer is building AI or data center capacity. Capacity is what the deal discloses, and the score is each supplier's current AI Exposure Score.

Supplier	Score	Buyer	Power	Status	Capacity
Constellation (CEG)	68	Microsoft	Nuclear restart (Crane, Pennsylvania)	Signed	835 MW
Constellation (CEG)	68	Meta	Existing nuclear (Clinton, Illinois)	Signed	1,121 MW
Vistra (VST)	65	Meta	Existing nuclear (three plants, Ohio and Pennsylvania)	Signed	2,176 MW
Talen Energy (TLN)	60	Amazon	Existing nuclear (Susquehanna, Pennsylvania)	Signed	1,920 MW
NextEra Energy (NEE)	32	Google	Nuclear restart (Duane Arnold, Iowa)	Signed	615 MW
NextEra Energy (NEE)	32	Meta	Solar and storage	Signed	2,500 MW
Southern Company (SO)	27	Google	Nuclear uprates (Vogtle and Hatch, Georgia)	Signed	96 MW
Xcel Energy (XEL)	25	Google	Wind, solar and storage (Minnesota)	Signed	1,900 MW
Entergy (ETR)	33	Meta	Natural gas (Louisiana)	Signed	2,262 MW
Chevron (CVX)	8	Microsoft	Natural gas (West Texas)	Signed	2,670 MW
NRG Energy (NRG)	40	Undisclosed hyperscaler	Natural gas (Texas)	Signed	1,200 MW
Bloom Energy (BE)	70	Oracle	Fuel cells	Signed	2,800 MW
Bloom Energy (BE)	70	Nebius	Fuel cells (New Jersey)	Signed	328 MW
Solaris Energy (SEI)	85	xAI	Natural gas (Mississippi)	Signed	900 MW
Brookfield Renewable (BEP)	36	Google	Hydro	Signed	3,000 MW
Brookfield Renewable (BEP)	36	Microsoft	Mixed renewable	Signed	10,500 MW
Oklo (OKLO)	85	Meta	Small modular reactors (Ohio)	Signed	1,200 MW
Oklo (OKLO)	85	Switch	Small modular reactors	Letter of intent	12,000 MW
X-energy (XE)	62	Amazon	Small modular reactors (Washington)	Signed	960 MW
NuScale (SMR)	70	Standard Power	Small modular reactors	Letter of intent	1,848 MW

Show the check: every row links to its announcement on the Buildout Tracker, which also lists the deals with private suppliers.

Who is buying the power?

Four companies. Google is the buyer on 13 of the 52 tracked deals, Meta on 10, Amazon on 5 and Microsoft on 4. The rest are data center developers, neoclouds and AI labs.

That concentration matters for the suppliers. Their AI revenue depends on the same few buyers whose capex the September capex report tracks. A slowdown at any of them would likely show first in the letters of intent, which do not commit the buyer to anything.

Nuclear, gas or fuel cells: what kind of power are AI buyers signing?

Across all 52 deals:

Technology	Deals
Small modular reactors	15
Natural gas	7
Solar	5
Fuel cells	5
Existing nuclear plants	4
Mixed renewable	4
Solar and storage	3
Nuclear restarts	2
Fusion	2
Uprate, hydro, geothermal, wind, backup generators	1 each

Small modular reactors lead by count, but most of those reactors are still years from producing power, and several of the deals are letters of intent. Existing nuclear plants, restarts, gas and renewables are what can deliver electricity this decade. A power purchase agreement on a running reactor and an SMR letter of intent both count as one deal, but they are very different amounts of near-term revenue.

How much power are these deals for?

The 47 deals that disclose a size add up to 70.3 gigawatts. For scale, a gigawatt is roughly the output of one large nuclear reactor. The 91 data center projects in the tracker that disclose their size total 104.5 gigawatts of planned capacity.

The demand behind it is still growing. The International Energy Agency's base case has data center electricity use worldwide more than doubling between 2024 and 2030.

What about grid and electrical equipment companies?

Most of them have no named AI power deal, and that is expected. Companies such as GE Vernova, Eaton, Quanta Services, Hubbell and Comfort Systems sell turbines, switchgear, transformers and construction to many projects at once. Their exposure comes through volume, not through one contract. The Utilities & Grid layer holds 42 tracked stocks, and the Power & Energy layer 62. Each stock page explains how much of the business the data center buildout drives.

One public experiment holds several of these names. The Capex Hawk, Supercycle's AI analyst agent, runs a paper portfolio built on the capex thesis, and its 18 positions include GE Vernova, Eaton, Quanta, Vistra, Constellation, Bloom Energy and Solaris. It is a simulated portfolio run by AI, not a recommendation.

Why are power stocks in Extreme Fear if the deals keep coming?

On October 8, 2026, Supercycle's index scored Power & Energy at 22 and Utilities & Grid at 11, both Extreme Fear, against 33 for the AI universe as a whole. In the same quarter the tracker logged 14 new power deals, more than any earlier quarter in its records. Coverage of older deals is thinner, so treat that as a floor.

Where the data stops: the index measures how prices behave, not why. Two common explanations are interest rates, which weigh on utilities more than most stocks, and regulators' scrutiny of who pays for connecting large data centers to the grid. Either could weigh on these stocks while the deals keep arriving, but the index cannot say which one is driving it.

What would change the picture?

• Letters of intent turning into signed contracts. The tracker has 8 today.
• Capex guidance from Google, Meta, Amazon and Microsoft in the late-October earnings round. They are the buyers on 32 of the 52 deals.
• Decisions by state regulators on how much large data centers pay for grid connections, which set the economics for every utility on this list.