When you open ChatGPT and ask a question, you aren't talking to an AI. You're talking to the synthesised output of a corporate infrastructure that your electricity bill subsidises whether you know it or not, that your pension savings finance whether you choose it or not, and that has gone in ten years from a non-profit organisation founded to "benefit humanity" to a Public Benefit Corporation with a Microsoft stake valued at around 135 billion dollars. The friendly face that answers you about how to draft an email or how to cook lentils lives on top of a structure no business school has yet modelled well and that is probably changing the cost of your daily life without your having any way to vote on it.
Today we're going to look slowly at that structure. Not in the tone of a user manual, but in the tone of the person who pays the price without knowing it.
How OpenAI affects you even if you don't use it
Before getting into corporate structure, a concrete question is in order. If you aren't a ChatGPT user, if you have no paid subscription, if you don't use the API, if generative AI doesn't enter your work, in what way does OpenAI affect you?
In several ways. Look at your electricity bill. In 2025 and 2026, electricity prices in markets near data centres are rising at abnormal speeds. In the PJM electricity market — the one running from Illinois to North Carolina and covering more than 65 million people — the cost of capacity shot up in the 2025-26 auction, with data centres as the main driver of new demand, according to IEEFA's analyses. That translates into higher residential bills for households in several states of the market. The people paying that increase mostly don't use ChatGPT. They pay because they live nearby.
Spain isn't in the same electricity market, but the underlying pattern is the same. The construction of data centres competes for electrical capacity with residential demand, traditional industrial demand and the electrification of transport. The growth of that industrial demand doesn't show up in a headline, but it's starting to show up in the budget.
Look at the cost of chips. Demand for high-bandwidth memory (HBM, the specific memory that feeds AI accelerators) has grown with the pace of training the large models, and that pressure on the memory supply tends to transmit to the price of every other product that shares the supply chain: laptops, consoles, servers. The consumer who buys hardware pays, without knowing it, part of the bill for the race to the next model.
Look at the cost of talent. Competition for senior AI engineering profiles pushes the sector's salary packages upward, because the big labs and the hyperscalers bid for the same people with budgets a mid-sized company can't match. That pressure runs down the curve: the company that needs to hire that profile raises its cost, and that cost ends up, by some route, in the price of the service it sells.
Look at general inflation. Financial analysts have spent two years arguing how much of the hyperscalers' abnormally high capex — Microsoft, Google, Amazon, Meta, all simultaneously investing hundreds of billions in AI infrastructure — is feeding a sectoral monetary expansion that transmits to other prices. There's no consensus, but the question is legitimate.
And look at financial risk. If OpenAI's bet fails — if revenue growth doesn't come to sustain the valuation Microsoft today presupposes on its stake of some 135 billion dollars — the adjustment will pass through global capital markets. Your pension fund, whatever it is, probably has indirect exposure to Microsoft. Your savings are, in some small percentage, tied to this turning out well.
OpenAI isn't a company used only by whoever uses it. It's a company that touches, via electricity prices, hardware costs, salary distortion, sectoral monetary policy and capital markets, several dimensions of the life of practically any citizen of the developed world. And that's what matters before getting into how it's organised on the inside.
The original trick of 2015
The story OpenAI tells about itself begins in December 2015. Sam Altman, Elon Musk, Greg Brockman, Ilya Sutskever, Wojciech Zaremba, John Schulman and others sign a founding manifesto. The text says they're creating a non-profit organisation "to ensure that artificial general intelligence benefits all of humanity". The declared intention is to free up research, to prevent a dominant player from monopolising the technology, to keep things open. The legal structure goes along: 501(c)(3), tax-exempt, governed by a board, distribution of profits forbidden. Philanthropy.
That structure lasted four years. In 2019, OpenAI announces the creation of OpenAI LP, a capped-profit arm — limited profit — controlled by the original non-profit. The public justification: training models requires private capital the non-profit can't attract. The implicit justification: the non-profit didn't scale at the pace the technical race demanded. Microsoft enters that same year with an initial one-billion-dollar investment and becomes the exclusive cloud-infrastructure provider.
What was happening, translated without diplomacy: an organisation founded to prevent the corporate concentration of AI was coming to depend structurally on the world's second hyperscaler. What was going to be open, closed. What was going to be distributed, concentrated. What was going to be non-profit, set up a for-profit arm with capped profit. The founding manifesto stayed on the website. The operational reality changed.
The coup of November 2023
Fast-forward to November 2023. On 17 November, the board of the OpenAI non-profit — which in theory controls the entire conglomerate — fires Sam Altman as CEO. The official reason, in the statement: "lack of candour in his communications with the board". No further specifics.
Four days of chaos. Microsoft, with 13 billion committed and total operational dependence, announces it will hire Altman and anyone who wants to follow him. More than 700 of OpenAI's roughly 770 employees sign a letter saying they'll resign if Altman doesn't return. Investors press. On the night of 21-22 November, Altman's return as CEO is announced. The board changes. The members who had fired him leave. A new board, friendlier to the executive leadership, comes in. The CEO's supremacy over governance is sealed shut.
Here something happens that's interesting operationally. The legal structure was still, on paper, that of a non-profit controlling a capped-profit. In practice, the non-profit board has just demonstrated that it can't exercise its theoretical function — firing the CEO — without the corporate ecosystem neutralising it within a week. The founding governance, the one that supposedly guaranteed OpenAI would serve "humanity" and not its investors, is demonstrably weak. The one who rules is the CEO with Microsoft's backing. The one who signs is the board. The signature counts for less than the command.
And then the end of the non-profit
The erosion of the founding model culminated on 28 October 2025. OpenAI completed its reorganisation: the operating entity became OpenAI Group PBC, a Public Benefit Corporation. It's an intermediate legal figure — a corporation with a declared public-benefit purpose but structured to distribute profits — that definitively abandons the capped-profit cover and opens the door to an eventual IPO.
In that reorganisation, Microsoft's stake came to around 27% (on a diluted basis), according to the count reported by the financial press. On the total committed investment — 13 billion dollars — that stake was valued at some 135 billion, within an OpenAI valuation close to 500 billion dollars. Even with the rounding, the multiple on the committed capital is among the highest recorded in modern corporate investment.
The founding manifesto's line about "benefiting all of humanity" is still somewhere on the website. The operating structure, today, is that of a public-benefit corporation with a majority technological shareholder, an explosive valuation, a declared commercial purpose, and a CEO armoured against his own original governance after the 2023 episode. Anyone who asked, in 2015, what OpenAI was going to be in ten years couldn't have got it more right than by saying: "it'll be a company".
The race to nowhere
Here it's time to ask whose fault this is. Because the process described isn't the result of a malevolent decision: it's the result of the sector's competitive dynamics. And it's worth looking at the players carefully so as not to fall into easy personalisation.
Is it Sam Altman's fault? Partly. His management has taken OpenAI from a non-profit to a PBC with an astronomical valuation, he's been the engine of the funding rounds and he piloted the November 2023 conflict in his favour. But Altman is executing a strategy any CEO in his position — with investors pressing, with Google DeepMind and Anthropic breathing down his neck, with the option of falling behind or accelerating — would probably execute much the same.
Is it the board's fault? Partly. The original board tried to halt the drift. It failed. The current board cooperates. But an isolated board can't oppose a sectoral dynamic that comes from outside the company itself.
Is it the investors' fault? More so. Microsoft, the venture-capital funds, the institutional investors who took part in the later rounds — all have pushed in a consistent direction: maximise value, abandon structural restrictions, move toward an eventual IPO. Their incentives are fiduciary. Each manager answers to their stakeholders. The capital structure pushes the CEO in the only direction profitability allows.
Is it the competition's fault? Yes. The presence of Anthropic, Google DeepMind, xAI, DeepSeek, Meta AI competing simultaneously for the same technical frontier forces OpenAI to invest enormous amounts in infrastructure so as not to fall behind. If OpenAI had stayed a conservative non-profit, one of the competitors would have taken its place. The competitive structure of the sector doesn't allow a single company to put on the brakes.
Is it geopolitics' fault? Quite a bit. The United States-China race has given the American frontier sector political cover that prioritises speed over caution. The US administration, since 2023, has incentivised national leadership in AI. Any attempt by OpenAI to self-regulate against the pace of the frontier would carry a geopolitical cost.
What emerges is a familiar situation in economic history: a race to nowhere in particular, in which each player optimises rationally within its constraints, and where the aggregate result is a brutal concentration of power in very few companies, a massive infrastructural dependence among them, a diffuse social cost spread across the whole of society — in electricity bills, in hardware, in salaries, in financial markets — and a founding promise to "benefit humanity" that evaporates because no corporate structure was designed to withstand that pressure.
What ChatGPT gives back to you
When you ask ChatGPT, you aren't asking an AI. You're asking an infrastructure organised by OpenAI Group PBC on Microsoft Azure servers. The text box is the only piece of that whole machine designed for you to see.
The answer you get is legitimately useful in many cases. This isn't a criticism of the product. ChatGPT solves real tasks for you. That's true. What's also true is that behind the dialogue box there's a structure whose real configuration no user would accept if it were put before them explicitly as a contract: "do you accept that the company you're going to ask things of tomorrow has gone from philanthropy to a corporation with a trillion-scale valuation in ten years, with a CEO armoured after a failed governance episode, dependent on a single infrastructure provider?". If the question appeared on the first screen before accepting the terms, a lot of people would think twice.
It doesn't appear. What appears, instead, is a friendly text field asking what you want to know.
It's interesting that the 2015 founding manifesto is still online. It's called Introducing OpenAI. It talks about creating "a non-profit research organisation" focused on "the safe development of artificial general intelligence". Read it today with the 2026 structure in your head. It reads like a historical document of an organisation that no longer exists. But the name is the same. The brand is the same. The URL is the same. Everything else just changed.
Quick definitions
- Non-profit (501(c)(3)): a US non-profit organisation exempt from federal taxes. It's forbidden from distributing profits among members, officers or directors. OpenAI Inc. was founded under this figure in 2015.
- Capped-profit: a hybrid structure with profits limited to a predefined multiple over the investment. Designed to attract private capital without formally abandoning the charitable mission. OpenAI LP operated this way between 2019 and 2025.
- Public Benefit Corporation (PBC): a US legal figure combining a commercial purpose with a declared public benefit. It allows profits to be distributed and capital attracted without structural restrictions. OpenAI Group PBC adopted this figure on 28 October 2025.
- AGI clause: a contractual clause between OpenAI and Microsoft that limited Microsoft's access to future models classified as AGI by OpenAI itself. The 2025 reorganisation redefined this clause. Its exact content remains under negotiation.
Referencias
- OpenAI. Introducing OpenAI (2015). Founding manifesto.
- CNBC (28 October 2025). OpenAI completes for-profit restructuring; Microsoft takes ~27% stake. https://www.cnbc.com/2025/10/28/open-ai-for-profit.html
- Fortune (28 October 2025). OpenAI for-profit restructuring and Microsoft's stake. Valuation of Microsoft's stake (~27%, around 135 billion dollars on an OpenAI valuation close to 500 billion). https://fortune.com/2025/10/28/openai-for-profit-restructuring-microsoft-stake/
- IEEFA (2025). Analysis of the PJM electricity market's 2025-26 capacity auction and the weight of data centres in new demand. Source for the rise in electrical capacity in zones near data centres. https://ieefa.org/
- Microsoft / OpenAI (2019). Announcement of Microsoft's initial one-billion-dollar investment in OpenAI and the exclusive cloud-infrastructure provision agreement. Reported in the Wikipedia chronology OpenAI. https://en.wikipedia.org/wiki/OpenAI
- The New York Times (November 2023). Coverage of Sam Altman's firing and reinstatement.
- Wikipedia. OpenAI and Removal of Sam Altman from OpenAI. Founding (December 2015), names of the co-founders, Microsoft's 2019 investment and the chronology of the governance episode (Altman's return announced on the night of 21-22 November). https://en.wikipedia.org/wiki/OpenAI
- Hao, K. (2025). Empire of AI. Penguin.
- Lehdonvirta, V. (2022). Cloud Empires. MIT Press.
- Stanford HAI (2026). AI Index Report 2026. Section on corporate concentration.
También te interesa
- Sam Altman's real credibility
- Anthropic and OpenAI, two ways of understanding AI
- DeepSeek and the end of the American monopoly
No comments yet
No comments yet. Be the first.
Leave a comment