Abundance is a Choice

In 1830, producing 100 bushels of wheat took about 250–300 hours of human labor. Five acres. A walking plow. Seeds scattered by hand. Grain cut with a sickle and threshed with a flail.

150 years later, producing that same quantity of wheat took about three hours with modern machinery. Think about it — that’s a 100x increase in output per hour of direct labor, and that change freed human beings to do an incredible number of things with their lives.

You can see the result in the family budget. In 1901, American households devoted 42.5% of their spending to food. In 2024, that share was 12.9%. Feeding a family went from 2/5ths of your monthly expenses to 1/8th.

Putting that food on the table took 60+ hours of work a week in 1901. In 2024, that had fallen to 31 hours. I think it’s possible it gets to 15-20 hours in five years.

We stand at the precipice of an era of abundance like nothing we’ve ever seen before. It will create a lot of change, and require a lot of adjustment.

But if we don’t mess this up, the world will become a lot more prosperous for all.

Make a better life affordable

For most of human history, getting expert help meant finding a person who knew something you didn’t, and paying for their scarce time.

A patient tutor. Someone who could make sense of a complicated business problem. Someone who could take an idea in your head and turn it into something useful.

AI can make that help available to vastly more people.

Imagine a child who struggles with math having a tutor who can patiently explain the same concept six different ways, as often as needed. Imagine a small business owner getting hours back every week because routine paperwork, scheduling, and follow-up happen automatically.

Multiply those possibilities across millions of families and businesses. More gets produced. Services become cheaper. People get time back. Those gains will show up in lower prices and higher real incomes, giving families more room to live.

There will still be a lot of services driven by humans. Things that are hugely important and complex will continue to require trust, relationship, and care — and those will likely be provided by humans equipped with AI superpowers. (That’s certainly the bet I’m making with Contio.)

Over the next five, six, seven years, I believe we have the opportunity to achieve a world where 15 or 20 hours of work can pay for a decent life. Many (including me) will choose to work more than that, because they love what they do. Others (like Elon Musk) will keep sleeping on the floors of their factories and will gain even bigger rewards.

But think about what that base case means for a parent who needs more time with their kids. For someone caring for an aging spouse. For someone with the talent to start a business who has never had enough time or money to try.

It’s profound — but we can absolutely screw this up.

Look at what we did to nuclear power

In 1954, Atomic Energy Commission chairman Lewis Strauss imagined a future in which electricity would be “too cheap to meter.” The opportunity was enormous: produce vast quantities of reliable electricity with zero carbon emissions.

That future ran headlong into the morass that became the Nuclear Regulatory Commission. An agency that was supposed to license the construction of safe nuclear power plants steadfastly refused to do so, and managed to grind an entire technology’s advance to an all-but-complete halt.

MIT’s account of the industry notes that more than 30 plans for new American reactors have been canceled since 2000, while only three reactors have been built. The two new reactors at Georgia’s Vogtle plant finished seven years later than scheduled.

President Nixon launched Project Independence in 1973, arguing that the US could avoid conflict in the Middle East and have a more abundant future if we built hundreds of new nuclear plants.

Ironically, he also then proposed the creation of the NRC that spent its every moment of existence blocking the future he envisioned.

We could have chosen an abundant and safe clean energy future.

Instead, we now face a looming gap between our per-capita electricity generation and that of China, our biggest strategic competitor.

We should learn something from the last time.

The slices of cost in a token

Behind every AI query or agent, there are specialized chips, data centers, cooling systems, and electricity.

While we’ve learned how to massively decrease a data center’s water consumption through recirculation and recycling (modern ones now consume water at 1% the rate of farming), there is no getting around the need for power to light up data centers, and enable them to do their work.

By now, you’ve heard the term “token,” which is the uniform unit of measure in AI. Every token put into and generated by a large language model has a cost, usually expressed in the price per million tokens.

As of this writing, one of the most expensive general intelligence models is Claude Fable, at $10 per million input tokens and $50 per million output tokens. One of the leading models on the pareto frontier of highest intelligence for lowest cost is SpaceXAI’s Grok, at $2 in, and $6 out.

What’s interesting is how that cost breaks down — a slice of every token’s cost goes to the chips, the data center, the electricity, and finally, to the creators of the model itself. We are seeing a massive proliferation of “free” open source models, but they only eliminate the latter slice — you still need the chips, the server racks, and the power.

We are just scratching the surface of this abundance that AI can deliver, and yet we’re already straining to build the data centers and generate the power necessary to satisfy the demand we already have.

The International Energy Agency projects that global electricity consumption by data centers will roughly double between 2025 and 2030. It projects that electricity consumption by AI-focused data centers will triple over that period.

If demand for intelligence — not to mention Netflix, online banking, and a thousand other critical services Americans rely on — keeps growing at today’s pace, and we refuse to build the data centers required to meet it, there is only one possible outcome.

Token prices will explode.

That’s basic supply and demand economics. More demand chasing insufficient supply always makes that supply more expensive.

Compute inequality is a choice

In that world, wealthy families and big corporations will certainly have the means to buy the intelligence they need to get ahead.

Everyone else will get rationed access. Less intelligent models. Smaller allowances. And tasks that would be valuable to complete, but cost too much to justify.

Consider two children with AI tutors. One can get help for as long as it takes to understand the lesson. The other hits a usage limit halfway through.

Or consider two businesses. One can afford to run AI across its operations all day. The other has to ration each use. The company with more money gets more productive, faster, while its smaller competitor struggles to keep up.

That’s not a world I want to see. Compute inequality would be the opposite of abundance. And refusing to build the infrastructure for intelligence will push us directly toward it.

Let’s be clear — the companies servicing all this demand should pay for the costs they create. No government subsidies or corporate welfare. New data center projects should be paired with onsite power generation that does not spike the cost of electricity for the localities they sit in. The best of these projects will actually improve local power costs, by selling excess generation back into the grid and driving down prices.

The machines that changed agriculture gave us more food with a fraction of the human labor. AI can bring that kind of progress to work we could never afford to do, and capabilities we could never afford to have.

A child gets the help they need. A small business gets the capabilities of a much larger company. A family gets more of its week back.

We should choose the world that gives everyone access to abundant intelligence.

It’s time to build that future.