AI Wants Your Job — And Now It Wants Your Electricity Too
Do the math. Actually don’t — it’s too depressing
If you thought the 250-year arc of American innovation was heading toward some sleek Jetsons-style future, we have got some news for you: we’re basically replaying 1905 with GPUs instead of steam engines. Only this time, instead of Edison tinkering with bamboo filaments, it’s Google and NVIDIA fighting over who gets to drain the national grid first.
According to UBS’s new report on US electrification — a glossy ode to “progress” that somehow reads like a utility-sector obituary — America is about to collide head-on with a physics problem: AI needs more electricity than the country can produce. A single NVIDIA GPU now consumes as much power as an entire US household, and hyperscaler datacenters deploy half a million to a million of them at a time.
Do the math. Actually don’t — it’s too depressing.
The new gilded age of electricity
UBS draws a straight line from Edison’s Pearl Street Station in 1882 to today’s hyperscale server farms — and the parallels are uncanny. Back then, cities scrambled to build coal-fired plants just to light a few bulbs. Today, hyperscalers are quietly doing the same thing… except instead of powering lamps, they’re powering LLMs so TikTok teens can ask AI who likes them back.
Source: UBS
Source: US Census Bureau
Just to prevent ChatGPT from lagging...
UBS notes that the US needs 100–200 GW of new capacity within a decade — roughly adding 8–16% on top of total existing generation. For context, that’s like constructing 10–20 new nuclear reactors or 150+ natural gas plants just to prevent ChatGPT from lagging. Meanwhile, the average investor is still busy debating whether AI is “in a bubble.” Spoiler: the bubble isn’t in the market — the bubble is in the transformer substations about to blow.
AI hyperscalers: The new robber barons
The report also points out that hyperscalers — Google, Meta, Microsoft, Amazon — are already:
1. Hoarding land near high-capacity transmission lines.
2. Striking private power deals with utilities.
3. Absorbing the cost of new supply themselves.
4. Shopping for nuclear micro-reactors.
All while Washington pretends the national grid is fine.
Here’s the uncomfortable truth: There is no AI without power. And right now, AI is outbidding everyone else for the last electrons on the grid.
Source: BofA Quant
History repeats — First as innovation, then as breakdown
UBS helpfully reminds us that electrification has cycled through booms and busts before:
1. The first surge (1880–1920) rewired factories.
2. The second (1950–1970) powered appliances and suburban growth.
3. The third (2020–2035) will… keep the GPUs warm.
Except this time, the stakes are existential.
Electricity demand is expected to grow 2–3% annually, faster than GDP.
Source: Economic History Association
Source: EHA
Source: EHA
History rhymes...
And as electricity reshaped every aspect of work in the early 20th century, the same dynamic is re-emerging — except this time, the workload is shifting to the machines. The chart shows estimated average weekly hours worked in manufacturing.
Source: EHA
Innovation...
Back then, electricity gave us refrigerators and washing machines. Today it gives us 1.5 billion-parameter LLMs that can barely summarize a PDF without hallucinating. The chart shows how household electricity adoption spurred wave of innovation in appliances.
Source: OWID
Stall no more
Electricity use stalled after 5% annual growth from 1950-2000.
After decades of flat electricity demand thanks to LEDs and efficient appliances, AI has managed in 18 months what the entire US economy couldn’t: re-ignite a power consumption boom.
Source: US Energy Information Administration
So it begins
Rise in electricity prices exceeds overall inflation, big jump since 2020.
Source: St. Louis Federal Reserv
One heatwave away...
Electricity demand is expected to grow 2–3% annually, faster than GDP.
Energy outages already cost the US $150 billion per year in lost productivity. And that was before AI went exponential.
The report quietly warns that reliability is now the #1 risk.
Translation:
We’re one heatwave away from choosing between air-conditioning and AI inference.
Guess which one the hyperscalers will choose?
Source: UBS
The coming energy wars (You weren’t supposed to notice)
UBS is too polite to say it outright, but it’s all there between the lines:
1. AI will stress the grid harder than any technology in US history.
2. Utilities will not keep up.
3. Government will be forced into emergency pro-AI energy policy.
4. Private actors will increasingly control the power ecosystem.
It’s the 1900s utility-robber-baron era, rebooted. This time controlled by trillion-dollar platforms.
Tesla and Westinghouse fought AC vs. DC.
OpenAI and Google will fight MW vs. MW — megawatts vs. megawatts.
Bottom line
While Wall Street cheers the AI revolution, UBS’s report accidentally reveals the actual storyline:
AI isn’t an innovation problem — it’s an electricity problem.
We don’t have enough. We won’t build enough fast enough.
And the companies training frontier models know it.
If AI is the new gold, electricity is the new gold mine — and hyperscalers already bought the land.
The rest of us?
We’ll just get the bill.
