July 2026 Dan Ives Power 30 Research Report
The Bottleneck Shifts from Demand to Delivery
Dan Ives and the Wedbush research team have updated the Dan Ives Power 30 Research Report (the “AI Power Report”), adding Ormat Technologies (ORA) and Celestica (CLS) while removing Dominion Energy (D) and American Tower (AMT). The July 1, 2026 update reflects Wedbush’s view that power delivery rather than customer demand has become the constraint that decides what gets built.
The AI Power Report launched in March 2026 on the conviction that the AI buildout is a physical infrastructure story and that the physical bottleneck is power. Three months on, Wedbush sees that thesis accelerating. Hyperscaler capital commitments have grown, interconnection constraints have tightened, and the companies that own the generation, grid, and hardware assets the AI stack runs on have become harder to replicate.
AI Power Report Updates
Additions to the AI Power Report
- Ormat Technologies (ORA): Added to Power Generation & Fuel Supply as the only vertically integrated geothermal operator in the world, with roughly 65% global share in binary cycle technology and 95%-plus plant uptime. ORA has signed its first direct data center PPA (power purchase agreement), a 20-year deal with Switch, plus an agreement with NV Energy to supply Google’s data centers with up to 150 MW, and first quarter adjusted net income rose 93.5% year over year. Geothermal runs at a roughly 92% capacity factor, on par with nuclear, and fills a baseload gap no other member addresses.
- Celestica (CLS): Added to Equipment & Power Management as a hardware co-design partner to hyperscalers, building AI servers, 800G and 1.6T networking switches, and custom rack systems at scale. First quarter revenue rose 53% to $4.05 billion; the hyperscaler-facing segment, now 80% of revenue, grew 76%, and full-year 2026 guidance was raised by $2 billion to $19 billion. CLS covers the hardware manufacturing layer no other member occupies.
Removals from the AI Power Report
- Dominion Energy (D): Removed following NextEra Energy’s announced all-stock acquisition, a $66.8 billion transaction Wedbush describes as the largest power-sector deal on record, with a 12 to 18 month path to close. D shares now trade on the deal outcome rather than a standalone infrastructure thesis, and NEE remains in the list.
- American Tower (AMT): Removed as CoreSite, the data center business that qualified AMT for inclusion, represents roughly 9% of revenue, with the remainder in wireless towers that carry no direct AI power thesis. Data center REIT exposure is consolidated in Equinix (EQIX) and Digital Realty (DLR).
The Power Wall Is the Story
The bottleneck has shifted from demand to delivery. The five largest hyperscalers are committing a combined $690 billion in 2026 capital expenditures, nearly all of it driven by AI, yet the defining question is no longer whether customers want capacity. It is whether the grid can deliver enough electricity to run it. Microsoft has described an $80 billion Azure backlog driven by power availability rather than weak demand. U.S. data center power demand is on track to reach 76 GW (gigawatts) by year-end 2026, up from roughly 50 GW in 2024, and Wedbush’s bottom-up model still projects consumption of 470 TWh (terawatt-hours) by 2030, roughly 10% above the International Energy Agency base case.
The Grid Cannot Absorb It Quickly
More than 2,600 GW of generation and storage capacity sits in U.S. interconnection queues, nearly double the country’s entire installed capacity, and entering the PJM queue today means no interconnection agreement before 2028 at the earliest. Utilities are committing capital to match. S&P Global forecasts aggregate U.S. energy utility capital spending of $1.3 trillion for 2026 through 2030, a 29% increase over prior estimates, with Duke Energy planning $102 billion through 2030 and Southern Company $81 billion through 2029.
Policy Is Moving Toward Baseload
The NRC (Nuclear Regulatory Commission) Part 53 rule, effective April 29, 2026, is the first new reactor licensing framework since 1989. In June 2026 the DOE (Department of Energy) issued a $17.5 billion conditional loan commitment to finance long-lead components for ten large-scale AP1000 reactors, the largest direct federal financing of the nuclear supply chain since the original U.S. fleet was built. Geothermal and nuclear are winning hyperscaler PPA attention because both run at roughly 92% capacity factors against about 23% for solar and 35% for wind, and Google has now signed agreements in both.
What Wedbush Is Watching in Q2 Earnings
Wedbush framed the second quarter earnings cycle as the first real stress test of the thesis. The five signposts are whether Microsoft, Alphabet, Meta, and Amazon hold or raise capex guidance, whether Vertiv’s EMEA bookings recover, whether GE Vernova’s margins confirm new slot reservations pricing above backlog, whether Quanta’s book-to-bill holds above 1.5x, and whether utility earnings from Southern Company, NextEra, Constellation, and Vistra include new large-load PPA announcements.
Conclusion
The AI story is now a story about electrons, steel, and turbines, and the July update reflects where that constraint is binding hardest. The roster changes add 24/7 clean baseload and the hardware manufacturing layer while removing a merger situation and a name with limited direct AI power exposure. The thesis the AI Power Report was built on is now playing out in public.
Important Information
This content draws from the July 2026 AI Power Report by Dan Ives and the Wedbush Research team. Excerpts are used under fair use for educational and informational purposes. All rights reserved to Wedbush Securities Inc.
This material is for informational purposes only and does not constitute investment advice or an offer to buy or sell any security. The views expressed are those of the authors and are subject to change. Investing involves risk, including loss of principal. Past performance is not indicative of future results.
References to specific securities are for illustrative purposes and do not constitute a recommendation.
