AI’s Energy Paradox: Can AI Solve the Grid Crisis It Created?
The rapid surge in AI infrastructure’s power demand has created a historic collision between big tech and the U.S. power grid. This could become the watershed moment for American energy policy and grid stability.
The combination of staggering growth, abrupt load changes due to AI workloads, and infrastructure bottlenecks are leading to higher bills for consumers and resource competition in the industry. The consumer backlash against AI datacenters is becoming part of the daily news and the policy makers are drooling over the opportunity to create legislative controls over the nascent industry.
An average home solar installation generates about 28 – 36 kWh of electricity per day. The unused energy gets pumped back to the grid that helps lower your electricity bill based on the rate set by the grid. On top of it, they have the right to curtail – that is, the grid operators can block the absorption (buy back) based on the grid capacity and other factors.
As per statistics available online, the small-scale photovoltaic (PV) systems across the US pumped ~93 billion kWh of electricity back into regional grids and the large-scale systems contributed significantly more. Interestingly, just the state of California curtailed 3.4 million megawatt-hours (MWh) of wind and solar in 2024, with solar accounting for 93% of that waste! See this Reuters article on it.
The problem of curtailment is even worse across Europe. Projections indicate that European grids are on track to waste ~40 terawatt-hours (TWh) of electricity because regional transmission networks cannot handle midday summer surges. This is enough energy to power London for an entire year.
Let that sink in for a moment! This colossal waste while Europeans face sweltering summers without air-conditioning. I am surprised that no environmentalist sellabrities or climate-change advocates are not up in arms against this wastage.
AI to the rescue
What if AI becomes part of the decision-making layer of the energy system itself? Imagine a solar installation generating more electricity than its owner needs. The question is no longer simply, “Where does the surplus go?” but a whole set of decision making AI agentic systems that look at the situation holistically.
- How much solar power is being generated?
- What will the weather produce next day or week?
- How much energy is stored in the battery?
- How much electricity will be needed by the household?
- What are the electricity prices doing?
- Who nearby needs power and is willing to buy?
- Will the grid curtail or absorb?
And then the AI makes a decision: Use it. Store it. Sell it. Or wait.
Another layer can be added to ‘uberize’ your power generation and storage. The AI may decide to wait and sell the generated / stored power to your neighbour, or a distant town resident, during the peak hour at a slightly lower price than the grid. You make money and your neighbor saves some and the grid does not need to curtail. This is a win-win-win situation.
Now scale that from one solar installation to thousands or millions – homes, factories, commercial buildings, and parked EVs. AI could potentially coordinate millions of distributed energy assets and turn them into something resembling a virtual power plant.
This changes the role of AI. It isn’t merely predicting energy demand or optimizing a solar panel. Now, it is beginning to participate in the economics of energy. We may be entering a world where AI doesn’t just consume the world’s electricity. It helps decide where the world’s electricity goes. And perhaps that’s the more interesting AI story.