AI Needs Energy. From Uranium to Cooling: The Power Stack Behind AI 👇

Goldman Sachs now sees data-center power demand rising 220% by 2030 vs. 2023, adding 905 TWh. Morgan Stanley sees U.S. data-center demand reaching 74 GW by 2028, with a 49 GW power-access gap.

$CEG

Constellation $CEG is the nuclear-heavy power seller for hyperscalers wanting clean baseload, not intermittent supply. Its edge is scale: the largest U.S. nuclear fleet plus the $26.6B Calpine acquisition, adding 27 GW of gas assets. Nuclear gives 24/7 carbon-free power. Gas adds dispatchable flexibility for data-center timelines.

$TLN Talen

$TLN is turning from a merchant generator into a long-term AI power landlord. Susquehanna gives it a 2.2 GW nuclear anchor, while Amazon’s expanded agreement can pull up to 1,920 MW through 2042. Colocated data centers get firm power. Talen gets contracted cash flows instead of pure merchant volatility.

$VST Vistra

$VST is a large competitive power producer sitting in the sweet spot: nuclear, gas, and power markets with rising scarcity value. Meta’s 20-year nuclear agreement supports a 2.6 GW AI data-center load in Ohio. Cogentrix adds 5.5 GW of gas plants, giving Vistra more dispatchable capacity for hyperscale demand.

$NEE NextEra Energy

$NEE is the clean-power scale player for hyperscalers chasing renewable PPAs. Solar plus batteries sit at the core: 2.2 GW of solar and 1.3 GW of storage were added to new project contracts in Q1. Management also flagged ~12 GW of potential data-center demand, positioning NextEra as the utility-scale clean energy supplier.

$ETR Entergy

$ETR is the regulated utility turning Gulf South industrial load into an AI infrastructure strategy. Meta’s Northeast Louisiana data-center project pushed Entergy toward a $57B four-year capex plan, with $27B aimed at new generation. Its 7–12 GW data-center pipeline makes territory selection a major part of the thesis.

$BE Bloom Energy

$BE sells on-site fuel cells for data centers stuck behind grid queues. Oracle’s Project Jupiter is the flagship: up to 2.45 GW of solid oxide fuel-cell capacity for a private microgrid in New Mexico. Instead of waiting years for interconnection, hyperscalers can bring power onto campus and scale faster.

$CAT Caterpillar

$CAT is the old-school industrial winner hiding inside the AI power stack. Data centers need prime power, backup power, and cooling-linked generation. Its Joule Capital deal targets 4 GW of G3520K generator sets for a high-performance compute campus in Utah. Indiana capacity expansion supports multi-year distributed power demand.

$GEV GE Vernova

$GEV is the heavy electrical architecture supplier for the AI grid buildout. Data centers need turbines, transformers, grid gear, and electrification systems before racks go live. Q1 included $2.4B of data-center equipment orders, reportedly more than its full prior-year total. AI load is turning grid hardware into strategic infrastructure.

$PWR Quanta Services

$PWR is the contractor wiring the physical AI economy. Hyperscalers need high-voltage lines, substations, utility upgrades, and skilled crews. Quanta is tied to multi-year grid programs with utilities such as AEP and NiSource, while management frames the opportunity as a $2.4T infrastructure market through 2030. Picks and shovels, literally.

$POWL Powell Industries

$POWL makes custom switchgear and control rooms for power-dense data centers. AI campuses need safe distribution of extreme electrical loads, not commodity boxes. Powell landed its first individual data-center megaproject above $75M and passed $100M in data-center orders for the quarter. Backlog near $1.6B shows demand moving from talk to purchase orders.

$MTZ MasTec

$MTZ is a dual-track contractor for AI infrastructure: power delivery plus fiber connectivity. Data centers need electrons and network routes. Its work with Lumen supports medium-mile fiber builds linking rural compute clusters back to major hubs. MasTec benefits when hyperscalers need transmission pathways, renewables interconnects, and wireline infrastructure together.

$AMSC American Superconductor

$AMSC is a grid-stability specialist for constrained power regions. AI campuses create dense, sensitive loads, so voltage control and power quality matter. Its grid segment now drives most company sales, while management points to power electronics, materials, and semiconductor-related demand tied to data-center resiliency. Less flashy. Very bottleneck-adjacent.

$FLNC Fluence Energy

$FLNC builds utility-scale batteries for the 24/7 AI power problem. Solar and wind are useful, but data centers need smoothing, shifting, and reliability. Fluence delivered 2.3 GWh of storage in fiscal Q1, holds a $5.5B backlog, and is pursuing a ~36 GWh data-center pipeline with its high-density Smartstack platform.

$EOSE Eos Energy

$EOSE is a long-duration battery play using zinc-based chemistry for safer, multi-hour storage. AI campuses moving off-grid or into constrained regions need more than short backup. Its TURBINE-X partnership targets up to 2 GWh across hyperscale projects, pairing batteries with gas turbines. MN8 adds a 750 MWh solar-plus-10-hour storage angle.

$GNRC Generac

$GNRC is evolving from home backup into behind-the-meter AI microgrids. The EPC Power partnership links SBE Block batteries, ARC controllers, and industrial inverters. Goal: handle fast load swings from AI training, smooth disturbances, and switch between grid power and on-site generation. Uptime becomes a power-electronics problem.

$FSLR First Solar

$FSLR is the domestic solar module supplier for hyperscale clean-energy procurement. Its thin-film panels feed utility-scale projects tied to long-term corporate PPAs. Sun Streams 2 in Arizona, a 150 MW project linked to Microsoft data-center power, shows the model: lock clean generation before AI campuses consume more grid capacity.

$VRT Vertiv

$VRT sells the life-support layer inside AI data centers: power distribution, thermal management, and liquid cooling. Blackwell-class GPU racks can push 100 kW+ density, making air cooling less viable. MegaMod HDX targets prefabricated liquid-cooled deployment, while Strategic Thermal Labs adds dry coolers and heat-exchange depth.

$ETN Eaton

$ETN is moving from electrical components into full grid-to-chip power architecture. The NVIDIA collaboration around Beam Rubin DSX connects Eaton directly with AI Factory reference designs. Medium-voltage switchgear expansion adds manufacturing capacity, while the Boyd Thermal acquisition gives Eaton a bigger cooling footprint inside hyperscale facilities.

$NVTS Navitas Semiconductor

$NVTS targets the efficiency bottleneck inside AI server power supplies. GaN and SiC help shrink components, reduce losses, and raise power density. The company has shown 3.2 kW, 4.5 kW, 8.5 kW, and 12 kW PSU designs aimed at high-load AI racks. Less wasted power means less heat and better rack economics.

$CCJ Cameco

$CCJ is uranium supply exposure for the nuclear AI buildout. Hyperscalers want 24/7 clean baseload, and reactors need fuel before they produce a single megawatt. Cameco’s Westinghouse stake connects it to reactor expansion, including the U.S. government and Brookfield partnership targeting at least $80B of new nuclear projects.

$LEU Centrus Energy

$LEU sits at the enrichment bottleneck. Advanced reactors and SMRs need secure domestic fuel supply, especially HALEU. Centrus selected Geiger Brothers for a major Piketon, Ohio expansion built around AC100M centrifuges. The commercial LEU backlog near $2.3B gives the company a direct line into nuclear fuel scarcity.

$UEC Uranium Energy

$UEC offers unhedged U.S. uranium exposure for a nuclear market suddenly tied to AI power demand. ISR projects in Wyoming and South Texas give it domestic supply optionality. No heavy long-term hedge book means more upside if uranium pricing stays tight as utilities, hyperscalers, and reactor developers compete for fuel security.

$BWXT BWX Technologies

$BWXT is the nuclear hardware supplier in the stack. It manufactures pressure vessels, precision components, and specialized nuclear systems for commercial and advanced reactors. Utilities extending plant lives and uprating output need suppliers like BWXT. SMR developers also need qualified nuclear manufacturing, a much smaller club than software investors realize.

$OKLO Oklo

$OKLO is trying to become a data-center utility through fast-fission microreactors. Aurora, its 15 MW project at Idaho National Laboratory, is the anchor concept. The model is simple but hard: deploy clean power near compute, bypass grid congestion, and sell electricity through long-term PPAs to operators such as Switch.

$SMR NuScale Power

$SMR is the modular nuclear platform bet. Its TVA and ENTRA1 Energy partnership targets up to 6 GW of SMR capacity across the TVA region. A 12-module plant design can scale in blocks, matching industrial and AI load growth over time instead of forcing one massive gigawatt-scale nuclear decision upfront.