A manufacturer I spoke to last week mentioned how they deployed AI-powered quality inspection on one production line last year and their defect detection improved 34%.
Their plant manager wanted to expand to 10 more lines but the electrical infrastructure couldn't handle it as the grid connection was maxed out.
And the utility said a transformer upgrade would cost $2 million and take 18 months.
So their AI deployment stalled.
👋🏻 I'm Leonardo Ubbiali. This week we're looking at why energy access limits AI deployment at scale, and why the companies securing power now will be the ones still running AI in 2030.
Molly Beerman (Alcoa's CFO) spoke at the Citi 2025 Basic Materials Conference in December:
"We are now competing with Amazon and Microsoft, who are willing to pay over $100 per megawatt hour for power."
A single aluminum smelter today uses 11 terawatt-hours of electricity per year i.e, the same amount of power as the entire city of Boston or Nashville.
Alcoa is considering selling its smelters to Big Tech companies who value the power generation equipment more than the metal production itself.
Microsoft had agreed to pay an estimated $110-115 per megawatt hour to restart Three Mile Island's Unit 1 nuclear reactor.
The 20-year deal gives them 835 megawatts of dedicated power for their data centers.
Beerman described what happened to Alcoa's West Coast smelters: "They're done. They're gone. They could not get the power."
Smelters shut down because they couldn't compete for electricity against buyers willing to pay triple what aluminum production can afford.
Unplanned disruptions
90% of companies experienced energy disruption in the past year, according to the Prologis survey. Price volatility, extreme weather, and full outages were the most common causes.
70% of executives now fear power outages more than any other disruption.
And only 27% have advanced backup systems in place.
AI data centers plus automated factories means massive power consumption, making reliable power a core strategic asset.
A Deloitte survey of 120 U.S. power companies and data center executives shows a seven-year wait on some requests for connection to the grid.
Most AI adoption projections assume unlimited power. They calculate ROI based on labor savings and efficiency gains but don't account for the grid not being able to deliver the electricity.
Grid strain plus data center competition means slower deployment.
For decades, manufacturers were passive consumers. You paid the utility bill, assumed power was always available and ran it 24/7 regardless of grid conditions.
That model is dead.
Now you're competing with data centers for limited power so you must lock in long-term contracts before deploying AI at scale.
Century Aluminum's Mt. Holly smelter in South Carolina secured power through 2031 before investing $50 million to restart idle capacity.
90% of manufacturers would pay a premium for reliable energy infrastructure today, according to Prologis as it is now the #1 factor in facility location decisions.
Companies investing in on-site generation gain an advantage. They're not vulnerable to grid disruptions or being outbid by data centers for power access.
Solar panels, backup generators, and energy storage aren't just for outages anymore.
Five things you can do this quarter
The problem: You're planning to scale AI across your operations, but you don't know if your electrical infrastructure can handle it.
What you need:
List of current and planned AI deployments
Current facility power capacity (if known)
15 minutes
The prompt (copy this):
I'm a [YOUR ROLE] at a [YOUR FACILITY TYPE] manufacturing plant. We're planning to scale AI deployments, and I need to understand if our electrical infrastructure can handle it.
Current AI systems running:
[List what's deployed]
Planned AI deployments in next 12-24 months:
[List what you want to deploy]
Current facility details:
Total power capacity: [if known]
Current power usage: [if known]
Grid connection limits: [if known]
Analyze this and tell me:
What's the estimated additional power draw from scaling these AI systems?
If our facility is near capacity, what's the typical lead time and cost to upgrade?
Which AI deployments are most power-intensive?
If we can't get more grid capacity for 18+ months, what alternatives should we consider?
What questions should I ask our utility provider right now?
Be specific. I need to know if power constraints will delay our AI roadmap.
This artifact is an interactive dashboard that answers one question: can your plant's electrical infrastructure handle your planned AI deployments?
You get a green/orange verdict badge, visual capacity bars, and a breakdown of how many kilowatts each AI system adds.
It also shows upgrade costs and timelines if you do hit limits, ranks which systems eat the most power, lists fallback strategies if grid capacity is constrained, and gives you ready-made questions for your utility provider.
How to read it: If the badge is green and utilization stays under ~90%, power won't delay your roadmap. If orange, jump to the Alternatives tab.
Aluminum Association: "Powering Up American Aluminum"
The white paper documents how data center power demand is pricing manufacturers out of the electricity market. Includes the calculation showing a single smelter needs the same electricity as Boston, the $40/MWh vs $115/MWh comparison, and why manufacturing can't compete with price-inelastic tech buyers.
Time to value: 25 minutes
How many of your AI deployment plans assume power will always be available when you need it?
Hit reply. I read every email.
Leo