AI Data Centers and Heat Overload US Power Grids: PJM Capacity Costs Jump 60%
Bloomberg describes a "perfect storm" for US power grids: the AI data center boom has collided with record heat in July 2026. In PJM, the largest grid (13 states plus Washington, D.C.), data centers pushed capacity procurement costs up by more than 60%, to a record $16.4 billion. Bloomberg forecasts that data centers will consume a fifth of US electricity by 2035.
AI-processed from Bloomberg Tech; edited by Hamidun News
Bloomberg Tech described a "perfect storm" for the US power grid on July 23, 2026: the simultaneous boom in AI data centers and record summer heat are pushing grids into emergency modes, and PJM Interconnection, the largest operator, has raised its capacity procurement cost by more than 60%.
What is happening to US power grids
US power grids can no longer cope with the double load from AI and the weather. In early July 2026, electricity demand in PJM Interconnection, the country's largest grid, hit a new record: temperatures in the Mid-Atlantic states topped 38 °C (100 °F), and a grid already stretched by the data center boom was pushed to its limit. Back in June 2026, PJM updated its emergency plan because of data center growth, and on July 14, 2026, the grid fell short of its target for capacity supply amid surging AI demand. PJM serves 13 states and Washington, D.C.
Key figures from Bloomberg:
- Capacity procurement costs in PJM rose by more than 60%
- The procurement record stands at $16.4 billion, of which about $6.3 billion is attributable to data centers
- Electricity demand in PJM hit an all-time high in early July 2026
- Temperatures in the Mid-Atlantic region exceeded 38 °C
- Emissions at Amazon and Google (Alphabet) rose in 2025 because of the power demands of AI
Why it is called a "perfect storm"
The "perfect storm" arises because heat hits data centers and the grid at the same time. According to Bloomberg, cooling accounts for about 40% of a data center's energy consumption even at normal temperatures, and that share grows in extreme heat. That is why summer peaks are the most vulnerable moment: extreme heat raises household demand for air conditioning and, at the same time, increases the energy servers need for cooling. The result is a vicious circle: data centers demand peak power precisely when an overloaded grid has the least to give.
"A perfect storm is brewing: more and more grid failures and power outages,"
Bloomberg's Bloomberg Tech newsletter said on July 23, 2026.
How much electricity AI will take
Data centers will consume about 20% of all US electricity by 2035, according to a forecast Bloomberg published on July 21, 2026. That is a fifth of national generation going to AI servers. A climate conflict is sharpening in parallel: greenhouse gas emissions at Amazon and Alphabet (Google) jumped in 2025 because a significant share of the energy for AI is still produced from fossil fuels, even though both companies have publicly stated climate goals.
What this means
The AI race is running up against a physical limit: electricity. As long as data centers are built faster than grids can add capacity, rising electricity prices and emergency modes in hot weather become a systemic risk, both for operators like PJM and for consumers in 13 states.
FAQ
How much have prices in the PJM grid risen because of data centers?
According to Bloomberg, capacity procurement costs in PJM Interconnection rose by more than 60% and reached a record $16.4 billion. About $6.3 billion of that amount is attributable to data centers.
How much electricity will data centers in the US consume?
According to Bloomberg's July 21, 2026 forecast, data centers will consume roughly 20% of all US electricity by 2035, a fifth of national generation.
Why is heat dangerous for data centers?
Cooling accounts for about 40% of a data center's energy consumption at normal temperatures, and that share rises in the heat. As a result, data centers demand the most power exactly when the overloaded grid can supply the least, Bloomberg notes.
Executive AI Consulting, Custom Agents & Team Workshops
Work directly with Zhemal Khamidun at hamidun.com — enterprise AI architecture, production RAG & multi-agent systems, and executive training.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.