My last hydro bill was around $90. I live in roughly 400 square feet. I use lights, heating, and air conditioning moderately, cook daily, wash clothing and linens a few times a week. These are not extravagances. For those of us in the developed world, they are the baseline conditions of a functioning life. If I live to 80, over the course of my adult years, I will have spent close to $70,000 on this utility alone. That's not accounting for inflation or the upgrading of my home to a larger space. That money, gone. Spent on a recurring bill I never agreed to, for a service I never chose, built on infrastructure that was decided, long before I was born, would belong to someone else.
We are all permanent renters of electricity. Not tenants who might one day have the option to buy it. By the time shares in utility companies became available to ordinary people, the terms had already been set, and the compounding returns already captured by the institutions that built the infrastructure. The grid exists. We pay for access to it. We can buy a small piece of the company that runs the wire, but only after the people who built it have already collected the majority of the pot.
This is what Sam Altman's essay "Built to Benefit Everyone," published on June 8, 2026, doesn't quite reckon with. Its central metaphor is rural electrification in 1920s America—towns shaped by physical limits, transformed by access to a new technology, lives measurably improved. Altman's argument is that AI is today's version of that infrastructure, and the design challenge is the same: get it to as many people as possible. It's a generous metaphor, and it's also one that proves his point wrong—because access, as the history of electrification shows, is not the same thing as benefit. And the gap between the two is where the real design problem lives.
It's true electrification delivered real gains. Living standards rose, lifespans extended, and median incomes increased. But these all arrived inside a structure that was never designed to distribute the underlying value of what it was delivering. As access spread, the infrastructure that made it possible consolidated. Grids, utilities, and generation capacity moved into the hands of a small number of institutions with remarkable speed. By the mid-twentieth century, vertically integrated monopolies controlled most of the electricity Americans consumed. Regulators eventually stepped in, but the ownership question had already been answered. Access was broad. Equity was not.
OpenAI's current structure follows this pattern in parallel—and the timing of this manifesto-disguised-as-an-essay is no accident. On the same day Altman published "Built to Benefit Everyone," the company filed its confidential IPO prospectus. The OpenAI Foundation holds 26% of OpenAI Group. Microsoft holds roughly 27%. The remaining 47% sits with employees and investors. The company began as a nonprofit, restructured as a capped-profit entity in 2019 to attract venture capital—a structure whose profit limitations were, according to a March 2026 financial analysis, increasingly vestigial as it pursued full for-profit conversion. Return dispersion across the cap table tells the rest of the story: early investors are looking at returns of 140x; late-round ones are looking at about 1x.
None of this is hidden. But it sits in sharp contrast to the essay's central claim that "transformative technologies can concentrate power, or they can broaden it." Altman states that OpenAI believes in the latter. What the essay does not provide is a mechanism. There is no structural commitment, no ownership model, no explanation of how a company valued at nearly a trillion dollars produces broad-based value rather than concentrated returns. For an essay with ambitions this large, that omission undermines everything that follows.
The farmer got the light. He became a permanent renter of something he would never have the option to own. A century later, we are about to make the same trade at an unimaginable scale—and we are being asked to trust that this time the outcome will be different. Altman may well believe that. But belief, however sincere, is not a design. And the history of infrastructure suggests that the time to ask for the roadmap is before the wire goes in—not after we've already agreed to pay the bill.
The standard defense of transformative technology is that productivity gains eventually lift everything. And the version of this argument applied to AI is that it will generate so much value that the ownership question becomes irrelevant. That argument has been made before—about the internet, about financial innovation in the early 2000s. The abundance arrived. The coordination problem did not resolve itself. We are not living in a moment of scarcity. We are living in a moment where access to more has consistently outpaced our ability to distribute what we already have.
Which means the deeper problem isn't who owns the wire. It's what the wire is built to do.
Electricity does not require anything of the person using it. You run a line to a house; the house is ready. The farmer did not need to understand alternating current to benefit from it. He simply needed to flip a switch. Intelligence is different. Altman writes that "a key long-term role for people will be deciding what is worth doing"—that humans will bring "values, taste, care, and responsibility to the work." He is right that these things matter. What he does not address is where they come from, or how they have been strategically shaped by systems that have been optimizing against them for decades.
A system trained on behavioral signals gets better at predicting what you will do next. And while this is useful, the same mechanism that makes it useful makes it dangerous in a way that is easy to miss: it learns to meet you where your habits are rather than where your needs are. Those habits are not neutral starting points. They have already been shaped by prior systems running the same optimization—a reinforcing loop that becomes harder to exit the longer you are inside it.
This is the foundation the attention economy was built on. It never asked what you needed. It asked what would keep you engaged, and it got very good at the answer. The behavioral data that AI now trains on is not a record of human preference—it is the output of that process. And if we don't address that, the new system will continue to learn from the residue of the old one.
The compounding effect of this is structural, not incidental. Exterior structures built by disoriented people tend to reproduce that same disorientation at scale. Not because anyone intends it—but because orientation is a prerequisite for building systems that serve human needs, and it is precisely what the current architecture does not support.
Altman proposes that an international coordinating body should eventually govern leading AI efforts to reduce catastrophic risk. It is a reasonable suggestion, and one other serious voices in the field have made. But institutions are built by people. And people shaped by extractive systems build extractive institutions—regardless of intention. That is not a cynical read of the individuals involved. It is just what disoriented systems do when they scale.
The car needed seatbelts. Altman uses this analogy in a footnote, and it is apt. But seatbelts manage the consequences of impact from the outside. They don't change anything about the conditions that produced the crash. What the analogy misses is the possibility of an intervention that works from the other direction—one that changes something inside the driver before anything goes wrong.
The alternative to a system that learns your patterns is one that reads your orientation—your present state, not your behavioral history. Not what you have done, but where you are. Patterns are retrospective. Orientation is live. A system trained on patterns serves the version of you that has already been shaped by everything that came before it. A system that reads orientation can meet you where you actually are—not where your history suggests you should be.
That requires a different kind of language. Not words, which carry the freight of existing systems. Not numbers, which reduce everything to what can be measured. Something closer to a universal code of need—simple enough to cross cultures and literacies, precise enough to be technically useful, and owned by the person generating it rather than the institution receiving it. The data stays with the individual. The system responds to where you are, not where you've been.
This is not a wellness product. It is a technical architecture built on a different premise—that the most important signal a system can read is not what you have done, but what you need right now. That orientation, not optimization, is the ground from which genuine human judgment grows. Build for that, and the values Altman wants humans to bring to the work have somewhere real to come from.
Altman is right that the future should be shaped by people rather than only by the companies building the most capable systems. He is right that human judgment should remain central. What the electricity metaphor misses is that judgment has to come from somewhere. It does not emerge automatically from access. It has to be cultivated—and the systems we build either support that or undermine it.
My hydro bill arrives every month. I pay it. I do not own the grid and I never will. That arrangement was made before I was born and has never been renegotiated. We are at the moment—right now, before the infrastructure of AI hardens into the same shape—where negotiation is still on the table. Not just in terms of who owns the wire. But in terms of how the wire behaves.
We do not need smarter infrastructure. We need infrastructure that makes us more ourselves. That is the question frontier AI has not yet asked. And until it does, all the access in the world will not get us where we are trying to go.