Samuel Insull and the electric iron

“Energy is now AI’s next global bottleneck.”

— Jensen Huang¹

Ask AI leaders about the barriers to bringing intelligence to every American and the rest of the world, and they'll talk about chips, regulation, and resistance to data centers. But electricity keeps coming up. Can we get the power we need to fuel the AI revolution?

Sam Altman, CEO of OpenAI, told the Senate in May 2025 that "the cost of AI will converge to the cost of energy" and that its abundance would be limited by the available energy. This is the story of how another Sam, Samuel Insull, helped build the utility system we depend on today, and how his story serves as a lesson for the challenges we face with AI.²

The Chicago Story

Chicago got its first streetcars in 1859. Horses drew the cars down the tracks. Cable cars began replacing horse-drawn service in 1882, and electric trolleys arrived in 1890. Chicago was growing, and it needed its infrastructure to keep up.³

To fund it, the City of Chicago granted franchises to private investors to operate lines in particular areas. Operators had to finance cars and track, and many also built their own power plants. Bonds allowed them to spread those costs over years of operating revenue. But a franchise was permission to use the streets, and disputes over how long that permission would last made long-term investment harder.

Charles Tyson Yerkes, who controlled major street railway companies and the Lake Street Elevated, sought longer franchises. His push for 50-year extensions in the late 1890s met fierce political resistance. After his defeat, he sold his Chicago interests and moved on to London.⁴

In 1901, new management took over the financially troubled Lake Street Elevated. The company subsequently bought land for a yard and powerhouse, then recognized that it could not afford to build them. It turned to Samuel Insull and Chicago Edison for electricity. Insull had a different vision. Customers could buy power from his larger stations instead of financing smaller plants of their own. He would balance demand across thousands and eventually millions of customers.⁵

Insull had come to Chicago in 1892 after eleven years as the right-hand man of the father of commercial electricity himself, Thomas Edison. Now, as president of Chicago Edison, he envisioned a far larger market than providing electric lighting for homes and businesses. The streetcars offered a way to build it.⁶

Generating capacity cost money even when it was idle. Each additional use that fitted into those quieter hours could spread the expense across more sales. A large station serving a wider range of customers could make electricity cheaper, bringing still more customers within reach. By aggregating the demand of transport, manufacturing, and lighting, Insull could build and serve at scale and bring down the cost for everyone. But getting the customers was only half the equation; building the machine was the other.

The piston-driven steam engines of Insull’s day had a physical limit. The larger they got, the more their pounding threatened the foundations beneath them. A steam turbine’s rotary motion offered a way around that limit, but Insull wanted a machine on an experimental scale. His old company, General Electric, was reluctant to build it and take the risk of failure.

According to historian Harold Platt’s retelling, Insull’s response was: "Do it anyway. I’ll take the risk." GE built the turbine for his Fisk Street generating station. As Chicago Edison prepared to start it, Insull’s chief engineer reportedly urged him to leave the building, warning that the machine might explode.

Insull refused to leave, reportedly replying: "If it blows up, I’m going to stay here and blow up with it." The turbine entered service in 1903 and helped establish a new scale of electricity generation.⁷

The Price of Scale

Once it was built, Insull went on the hunt for more customers to keep it busy 24/7. That required changing people's habits. Insull offered favorable rates to restaurants that stayed open all night and encouraged factories to add overnight shifts. His salesmen promoted household appliances. Trucks went through neighborhoods offering six months' free use of an electric iron to anyone who signed up for electricity.⁸

The offer made it easier for a household to try electricity. It could try a useful appliance without buying it first. Insull took on that initial expense in the expectation of a continuing electricity customer. As electric appliances became part of the household's work, the relationship could grow: more uses for the connection, more electricity sold through it, more reason to keep the service.

The customers also made his system work better together. In a later speech, Insull described brickyards and quarries that shut down during winter frosts, while department stores were busy at Christmas. The same generating system could serve their different demands. The company that assembled those customers could offer something harder for each to achieve separately.⁹

As Insull bought competitors and combined their systems, that advantage gave him increasing power over the city he supplied. By 1907, Chicago Edison and Commonwealth Electric had become Commonwealth Edison. His ambition to serve an entire market brought an unavoidable question: what would protect customers who depended on him?¹⁰

Insull proposed terms. In a 1910 address, he advocated public oversight of rates and securities issuance as obligations that accompanied monopoly. The arrangement would constrain what his company could charge and how it financed itself. In return, a protected utility franchise (similar to that trolley car franchise) would give investors greater confidence to fund expansion. He was making dependence on his company politically acceptable as well as commercially useful.¹¹

By 1920, Commonwealth Edison served roughly half a million customers. Electricity was becoming an expense that households and businesses expected to pay because they had organized their work around having it. Insull had helped create the market that justified his expanding system.¹²

The Empire

Insull went on to build an electricity empire. He sold discounted shares to employees and encouraged them to sell shares to friends, neighbors, and customers, paying commissions on their sales. He had recognized the appeal of giving employees a financial stake long before Silicon Valley made equity compensation familiar.¹³

Insull expanded through holding companies that owned stakes in other utilities, using borrowing and layers of ownership to extend his control. By 1930, his empire served about four million customers and held properties valued at roughly $3 billion. Its reach extended across 32 states.¹⁴

His financial machinery was less sound than his generating equipment. Widely distributed shares and relatively small stakes in layers of holding companies left his control vulnerable. Cleveland financier Cyrus Eaton accumulated a large position in Insull companies, creating a threat to that control.

By 1930, Insull faced a choice: buy Eaton out or risk seeing that large block of shares sold elsewhere. In the shaky market after the 1929 crash, the prospect put him under pressure. Borrowing to defend his control would add financial obligations just as the economy was deteriorating.

Insull borrowed heavily for a buyout that cost $56 million. The timing could hardly have been worse. When securities prices fell sharply in September 1931 as the Depression intensified, banks refused to renew his notes. His investment companies entered receivership the following April, taking his family’s fortune and many small investors’ savings with them.¹⁵

But Commonwealth Edison survived, and the utility services continued. Electricity demand fell during the Depression, but people kept buying power. Insull’s financial obligations had become more than he could sustain, even after decades of building the business underneath them.¹⁶

The Lessons for Today

Sam Altman is now proposing to build another metered utility. In March 2026, he described customers buying intelligence much as they buy electricity or water. There are lessons we can take from Insull’s empire for today’s AI buildout, but we need to be careful. Some analogies are lazy, particularly the idea that intelligence will be a utility.¹⁷

Here are the four lessons:

The Utility Parallel

Tokens are often compared to other utilities because they are a base input that makes all the applications on top of them work or work more effectively. Intelligence is to your email what electricity is to your toaster. To be honest, the comparison isn't that compelling.

Electricity, gas, and water are standardized outputs that change little over time. The technology supplying them evolves, but customers can generally treat the product as interchangeable. AI is different. Not every token has the same value, and the way we produce and use them shifts constantly. Treating intelligence as an undifferentiated product is a disaster.

The comparison is also supposed to extend to regulation, but here again it has limits. Utilities require expensive local distribution networks. Duplicating those networks can be wasteful, and customers may have little choice of supplier. AI models reach you through the internet. Changing providers can be as simple as putting a credit card into another website, although moving an established business workflow can be much harder.

Many calls to regulate AI concern harmful uses, autonomous systems, and the fear that it will do its job too well and outcompete humans. Those are different questions from how to govern a local distribution monopoly. Insull didn’t call for electrical utility regulation to keep John Henry in a job. I’m not sure he has much to say to current AI leaders from this perspective.

Funding the Ecosystem

This analogy to Insull’s efforts is much more accurate. Nvidia is probably the standout example, trying to fund an entire ecosystem around its AI chips. It has committed capital to frontier labs including OpenAI and Anthropic, backs robotics companies, and develops open Nemotron models. Huang’s reach, like Insull’s, extends throughout the industry.¹⁸

Nvidia has also agreed to buy residual unsold capacity from CoreWeave under a deal initially valued at $6.3 billion, running through April 2032 subject to contractual conditions. That supports demand for a customer’s infrastructure; it is not a blanket promise to assume its debt or compensate it for obsolete chips.¹⁹

Jensen is everywhere, and that invites comparisons with Insull’s empire. I still see a stronger financial foundation at Nvidia. At July 26, 2026, it held $56.6 billion in cash, cash equivalents, and marketable debt securities against $33.5 billion in principal debt. But its commitments deserve attention: its latest filing also discloses conditional guarantees capped at $105 billion for an OpenAI-related campus. Those obligations phase in with completed construction and cover defined lease and power payments if the tenant defaults. They are substantial potential exposure, even though they are not current funded debt.²⁰

You Get an Electric Iron

I also see lessons in how Insull built the utility business that OpenAI and Anthropic will probably adopt. Their incentives differ from Nvidia’s. Huang wants as many providers of intelligence as possible building on Nvidia hardware and systems. OpenAI and Anthropic want to differentiate what they supply, intelligence, and make the hardware and applications around it cheaper.

For them, it’s about adding value to their tokens and integrating other parts of the AI stack with them. We’ve already seen them begin to do this. OpenAI moved down the stack with Jalapeño, an inference chip developed with Broadcom, with initial deployment planned for the end of 2026. I would expect Anthropic to explore more control of its hardware too. Both companies have also moved up the stack with coding applications that compete with products such as Cursor and GitHub Copilot.²¹

Anthropic’s Claude Design competes for work done in Figma and Canva, while also supporting export to Canva. I expect the expansion into applications to continue. It’s a little like Insull’s free electric iron trials. I expect both companies to offer free or heavily subsidized applications in additional broad categories such as CRM and ERP, using them to establish recurring paid demand for intelligence tightly integrated with their models and other products. The application could be free while the work performed through it remains paid. Customer records, integrations, and approval processes would make that relationship harder to replace.²²

Beyond that, I expect them to extend further into knowledge-heavy fields such as biotech and medicine. I’ll be curious to see how far they can go in manufacturing, robotics, and transportation.

Peak Pricing

Here I see a strong electricity analogy. I expect pricing to shift to better match data-center capacity, just as it does for electricity. Software factories running with little human supervision could become the equivalent of Insull’s nighttime factories, taking advantage of cheaper compute during quieter periods. We can fit tasks that don’t demand immediate answers into windows where suitable capacity would otherwise go unused.

There is already a version of that trade: Anthropic’s batch service offers lower prices in exchange for accepting delayed results. It does not yet amount to universal time-of-day pricing, but it shows why the ability to wait has commercial value.²³

Signals to Watch

• Do the labs move further up and down the AI stack?
• How much debt and contingent exposure do the labs and Nvidia take on?
• Who is winning the contest to make complementary products cheaper: Nvidia or the frontier labs?