Unscarcity
Sign in for free: Preamble (PDF, ebook & audiobook) + Forum access + Direct purchases Sign In By signing in, you get our daily podcast email. Unsubscribe any time.

Unscarcity Research

Robots Run on Electrons: Automation Is an Energy Bet

Walmart's robot warehouses tripled the power bill. A humanoid runs on cents an hour. The flip from wages to kilowatt-hours is decided by the grid.

10 min read 2173 words Updated October 2026 /a/robots-run-on-electricity-not-wages

Note: This is a research note supplementing the book Unscarcity, now available for purchase. These notes expand on concepts from the main text. Start here or get the book.

Robots Run on Electrons: Why Automation Is an Energy Bet, Not a Labor Bet

Every automation business case has a line for wages saved and a line for machines bought. Almost none has a line for the substation.

On October 9, 2026, the Wall Street Journal reported what that missing line looks like at Walmart (Supermarket News summarized the report the same day). Electricity at one of the company’s automated grocery warehouses can run to $800,000 a month, against about $250,000 when people did the work. The engineers who keep the equipment running cost more than the warehouse workers they replaced. The wheeled robots that shuttle goods around break down too often, and some cardboard boxes are too big for the machines to handle. Walmart, which employs more than 140,000 warehouse workers and has put close to $1 billion into automated e-commerce fulfillment since 2016, is retrofitting buildings that went up in the 1980s and 1990s, and it is finding that the machine it bought does not plug into a wall. It plugs into a grid.

Read the $800,000 the way a finance chief would and it is a nuisance. Read it the way the book reads it and it is the whole story. A warehouse that automates has converted a payroll into a power bill. Everything that follows about when, where and whether physical work flips to machines sits downstream of that conversion.

The arithmetic nobody runs

Start with the number that makes the energy objection look silly, because it is the number most people reach for.

Figure’s second-generation humanoid carries a 2.25 kilowatt-hour battery in its torso. Suppose it drains that pack in two hours of hard work. That is a draw of about one kilowatt. The United States sold industrial electricity at an average of 9.09 cents per kilowatt-hour through the first eight months of 2026. So the robot’s meal costs around nine cents an hour. Make it a one-hour drain, a brutal assumption, and the figure doubles to eighteen cents. The order of magnitude does not move.

A human doing the same job runs on roughly 100 watts. Two thousand kilocalories a day is 2.3 kilowatt-hours, which would cost 21 cents a day at the same tariff if people could be charged like batteries. They cannot, so the comparison that matters is the wage. The median hand laborer and material mover in the United States earned $19.35 an hour in May 2025, and there are 2.95 million of them. Per robot-hour, electricity is a rounding error on the wage line, a hundred to two hundred times smaller. If the question were whether the robot’s food costs less than the worker’s pay, automation would have won a decade ago.

So why did Walmart’s bill triple?

Because a facility does not automate by buying a robot. It automates by becoming a machine. Storage-and-retrieval towers, conveyor loops, sorters, charging banks and the computers that choreograph them run around the clock, in buildings that now have to be lit, cooled and ventilated for equipment rather than for people on an eight-hour shift. The Journal gave the bill, not the breakdown, but the shape is familiar from every other automated plant: the humanoid’s battery is the smallest electrical appliance in the building. The warehouse that used to be a shed with people in it is now a machine with a meter on it, and the meter is what the grid sees.

That changes the question. Per unit of work, electricity is cheap. Per facility, it is a new fixed cost, a new interconnection request, and a new exposure to a price that is moving the wrong way.

The price is moving the wrong way

The Energy Information Administration’s Table 5.3 is the national price list. In 2020, industrial customers paid an average of 6.67 cents a kilowatt-hour. In 2025 they paid 8.51 cents. Through August 2026 the average is 9.09 cents, and the August figure alone, 9.64 cents, is 7% above August 2025. Commercial customers, which is where most warehouses are billed, went from 10.59 cents in 2020 to 14.04 cents so far in 2026. Households went from 13.15 to 18.19.

That is a 36% rise for industry and a 33% rise for commerce in six years, for the input that automation substitutes for wages. Nothing about the robot got more expensive. The thing it eats did.

The reason is the same one behind the book’s electron gap: demand is back. The EIA’s October 2026 Short-Term Energy Outlook has US electricity consumption up 4% in the third quarter of 2026 over a year earlier, with commercial demand, the category that includes data centers, forecast to grow another 2.8% in 2027 and industrial demand 2.7%. In the PJM grid, which serves 65 million people from Chicago to the Atlantic coast, the price of simply having generating capacity on standby went from $28.92 per megawatt-day for the 2024/2025 delivery year to $269.92 for 2025/2026. It then hit the regulator-imposed cap, $329.17 for 2026/2027 and $333.44 for 2027/2028, with forecast peak load up more than 5,400 megawatts in a single year, “driven largely by data center expansion.” Who absorbs those costs is a fight of its own. The relevant fact here is simpler. An automated warehouse is now a large load competing for the same substation, the same transformer and the same capacity auction as a data center, and it is a much smaller customer.

Capacity is the gate, price is the toll

Here the phrase “energy bet” needs sharpening. The bet is less about the price of a kilowatt-hour than about whether you can get one at all.

A robot vendor can promise a customer a machine in six months. No vendor can promise a utility interconnection in six months. Texas’s queue of large-load requests reached 474 gigawatts this summer, more than five times the grid’s record peak, and 116 towns and one state have now said no to big new loads outright. A grocery chain that wants to automate a 1990s distribution center in a constrained zone discovers that the binding constraint is not the vendor’s delivery schedule or the dexterity of a gripper. It is a transformer the utility will install in 2029.

This is why the right picture of the next decade of physical automation is a map, not a curve. Dexterity demos tell you what is technically possible everywhere. Electricity tells you where it will actually happen first: where power is cheap, where a connection can be had, and where the local politics will tolerate a building that draws like a small town and employs like a small shop. The Substitution Threshold is a total-cost threshold, kilowatt-hours and maintenance included, and total cost has a postcode. Amazon, which passed its millionth robot in 2025, is in effect a utility customer the size of a city, and it sites accordingly.

It also means the flip can run backwards. A facility that penciled out at 6.67 cents may not pencil out at 9.64, especially when the engineers cost more than the workers did. Walmart has not stopped automating. But the honest version of its story is that the company is learning the price of electrons at the same time as the price of robots, and only one of those prices is falling.

What the book does with this

Unscarcity argues that the Labor Cliff arrives as the serial removal of the economic reason to employ a human for a task. The electricity ledger adds a condition that is easy to miss when you look only at robot prices: each of those removals is also an addition to the load on a grid that is not keeping up. Three consequences follow.

The Labor Cliff has a kilowatt-hour clock. The rate at which physical work flips is bounded by the rate at which cheap, connectable power comes online. That is bad news for anyone hoping the transition will be slow enough to manage by inertia, because the constraint is being attacked with hundreds of billions of dollars of capital. It is good news for anyone who wants a lever: power pricing and interconnection policy are, in effect, automation policy. Nobody is voting on them as such.

The Foundation is an energy budget before it is anything else. The book’s Free Zones have to deliver housing, food, energy, healthcare, education and connectivity for roughly $8,000 per resident per year. Every one of those lines is produced by machines that run on electricity, which is why the site’s research treats the kilowatt-hour as the candidate unit of account and why the book’s timeline leans on fusion and cheap solar. A world where the kilowatt-hour keeps rising 5% or 6% a year is a world where abundance arrives late and unevenly, and where whoever owns the reactor owns the warehouse too.

Cheaper robots do not make cheaper goods on their own. The vendor prices the machine against the worker’s cost, not against its own electricity bill. Agility rents its warehouse humanoid for $8,500 a month, which its prospectus describes as “a discount to a customer’s fully burdened labor cost.” If the savings from a nine-cent-an-hour worker are captured by the robot’s owner and the grid’s owner, wage-anchored pricing holds and the consumer sees none of it. The electricity line is one more place the abundance dividend can be quietly kept.

The meter is the message

The twentieth century’s automation debates were about machines versus hands. The twenty-first’s will be about machines versus the grid, with hands as the swing variable. A humanoid’s electricity costs pennies an hour, and that fact is close to irrelevant, because the building it works in has become the customer, and the customer is standing in a queue behind a data center that asked for a gigawatt.

Walmart’s $800,000 a month is not a scandal. It is a receipt, the first big one in public, for what it costs to run the physical half of abundance on today’s grid. The book’s wager is that this bill falls the way the price of compute did. The evidence of 2026 is that it is rising instead, and that the race deciding when your job becomes a machine’s is being run not in a robotics lab but in a capacity auction in Valley Forge, Pennsylvania.

Sources

Share this article:

Share Share Share