The Factor That Could Kill AI

Accelerating inflation and AI data centers can absolutely create uncontrollable risk to the underlying tech narrative.

In a shockingly sneaky way, the cost of electricity is turning into a major sore point for AI data centers.

Electricity prices have been skyrocketing over the past year as the grid is having a hard time keeping up with AI’s insatiable appetite for massive amounts of energy.

Why is this important?

Data center load demand is exploding – Every major AI model run, training cycle, or inference request runs on GPU clusters that guzzle electricity. AI data centers consume multiples of a standard cloud facility’s power – think 20–50 MW per site. For perspective, a single hyperscale AI facility can draw as much power as a small U.S. city.

Unplanned demand shock – The U.S. grid planning model assumes gradual, predictable increases in demand. AI demand is vertical – hyperscalers like Microsoft, Google, and Amazon are committing to massive expansions right now, forcing utilities to scramble for capacity they don’t have.

Feedback loop with electricity CPI:

1. AI drives up power demand.

2. Utilities face infrastructure upgrade costs + fuel price volatility.

3. Those costs are passed to consumers, pushing electricity CPI higher.

4. Higher rates feed political pressure and make AI itself more expensive to operate – forcing higher pricing on AI services.

Meta angle: AI demand is inelastic. Microsoft won’t cancel a 200 MW training run because of a 10% hike in electricity rates – they’ll just pay it, and pass costs downstream. This means AI is a structural floor under electricity demand growth, which means CPI for electricity doesn’t get a breather even if residential usage dips.

AI and Bitcoin are now competing in certain regions for the same cheap electrons. In places like Texas, miners and AI data centers are already on a collision course over grid capacity. If AI eats too much, miners will be forced to migrate, but the broader public will start lumping “AI + crypto” together as reasons power bills are spiking.

And those numbers are expected to rise as tech companies continue to build out their AI-related infrastructure.

This problem is unlikely to be solved by developing more efficient models, either, and not just because software developers historically see improved efficiency as providing additional space to cram features into rather than a benefit in itself. It’s also because the AI tools available today rely upon the constant ingestion of information, which leads the companies that created them to fetch as much content as they possibly can.

This could really do in AI, and if politics get involved in a bad way, then add costs, delays, and more regulation into the mix.

If the bottleneck isn’t resolved, the pitchforks could be coming for AI if consumers are dealing with triple or quadruple of electricity bills.

It could be simply that AI is too expensive to integrate into daily life.

This risk has quietly risen to the fore.

I will watch this carefully.

Until then, the AI trade continues unabated until something can topple it.

I am bullish on AI in the short-term.