‍

There is a room where the real decisions about the AI economy are being made. Policy is not happening in that room. Regulation is not happening in that room. Redistribution proposals are not happening in that room.

What is happening in that room is capital allocation. Specifically: who will own the infrastructure that everything else runs on.

In May 2025, Sam Altman stood in Riyadh and announced a partnership with Saudi Arabia's sovereign wealth fund, the Public Investment Fund, to build Stargate: the largest single technology infrastructure investment in history. The same week, the UAE announced its own Stargate campus in Abu Dhabi: one gigawatt of AI compute capacity, described as the largest AI infrastructure project outside the United States.

These are governments. Not technology companies. Not venture funds. Governments. What drove them to move at this speed and scale is the same insight that most founders operating in the AI economy have not yet internalized:

Whoever owns the infrastructure captures the compounding value. Everything built on top of it pays rent.

"The question of who controls the AI infrastructure is not a technical question. It is the central political economy question of the next fifty years." — Mariana Mazzucato, Professor of Economics and Innovation, University College London, 2024

Source  Mazzucato, M. — The Big Con (2023); public remarks on AI infrastructure, 2024

The Game Being Played Without You

Let the numbers sit for a moment before the argument.

Anthropic raised its Series G in February 2026 at a $380 billion post-money valuation. By May 2026 it was in talks for a new round close to $950 billion. OpenAI was valued at $852 billion in March. The five largest hyperscalers, Microsoft, Amazon, Alphabet, Meta, and Oracle, spent $448 billion on AI infrastructure capital expenditure in 2025 alone. The projection for 2026 is $562 to $602 billion. Combined hyperscaler capital expenditure projected across 2025 through 2027 exceeds $1.15 trillion.

Saudi Arabia's HUMAIN, a sovereign AI entity owned entirely by the Public Investment Fund and launched in May 2025, has committed to a $10 billion venture fund, a $10 billion joint venture with AMD, a $2 billion partnership with Qualcomm, and a data center buildout targeting 6.6 gigawatts of capacity by 2034. The UAE's MGX, with a $100 billion target AUM, has taken positions in OpenAI, xAI, Databricks, Anthropic, and Binance. It co-launched the BlackRock-Microsoft-GIP AI Infrastructure Partnership with an initial $30 billion commitment scaling to $100 billion.

Now the other side of the ledger.

The MENA startup ecosystem reached $7.5 billion in funding in 2025, a 225% year-on-year increase. But 91% of that capital concentrated in Saudi Arabia and the UAE, flowing to established names. Across sub-Saharan Africa, the top four markets captured 85% of all tech funding. In the United States, Black-founded startups received 0.4% of all venture capital in 2024. In the United Kingdom, a decade of data found that Black and minority ethnic entrepreneurs received less than 1% of venture capital. The most consequential technology buildout in modern economic history is underway. The founders most exposed to its displacement effects hold almost no equity in its upside.

This is not a diversity problem in the way that phrase is usually meant. It is an ownership problem. And ownership, once concentrated, does not redistribute on its own.

Source  Anthropic — Series G at $380B (February 2026)

Source  Wamda — MENA startups record year $7.5B (2025)

Source  Crunchbase — Black founder funding multiyear low (2025)

The Copyright Fight and What It Tells You About the Frame

In December 2023, the New York Times filed suit against OpenAI. The allegation: that decades of journalism, reported, written, edited, and published by human beings, had been used without permission or compensation to train a commercial product now competing directly with the outlet that produced it.

The case survived a motion to dismiss in March 2025. A preservation order was issued covering more than 400 million ChatGPT user conversations. In June 2025, a federal court ruled in favor of Anthropic in Bartz v. Anthropic. Training on copyrighted material was fair use. A similar ruling followed in Kadrey v. Meta days later.

The legal framing is fair use. The economic framing is something older and more precise: enclosure.

In 18th century England, the enclosure movement converted common land into private property. Land that had been collectively used and managed for generations. The people who had worked that land, built lives on it, and developed knowledge about it were not compensated. The land became capital. They became tenants or displaced persons.

What is happening with AI training data is structurally identical. The collective creative and intellectual output of the internet, built by writers, coders, artists, educators, researchers, translators, and millions of ordinary people over three decades, is being enclosed as private intellectual property. The writers who went on strike in 2023 understood this. The visual artists filing class actions against Stability AI understood this. The programmers whose Stack Overflow contributions fed language models understood this. None of them have won. The enclosure is proceeding. And the people whose creative and intellectual labor built the training data, who come from every country, every language, every background, receive nothing from the asset their work created.

Source  NPR — NYT v. OpenAI copyright case survives motion to dismiss (March 2025)

The Policy Problem Is Structural, Not Political

It would be convenient if this were a partisan story, one political party the obstacle and another the solution. It is not. The speed problem, the gap between how fast AI capability is moving and how fast policy can respond, is structural. Not political.

Sam Altman personally funded one of the largest universal basic income studies ever conducted. Three thousand participants across Texas and Illinois received $1,000 per month for three years. The results, published in July 2024: recipients worked 1.3 fewer hours per week. They used more healthcare. They were marginally happier. They were more selective about employment. Nothing transformed.

This is not an argument against cash transfers. It is an argument about the category of problem. A stipend, at that amount, in that timeframe, did not change the structural position of recipients in the labor market. They were still in the same rooms, facing the same walls, with slightly more in their pockets.

The World Economic Forum projects 92 million job displacements by 2030. The legislative process for any of the serious frameworks being discussed, robot taxes, sovereign wealth funds, universal basic services, has barely begun in any major economy. Policy moves in years. AI capability moves in months. There is no rescue arriving at the speed the disruption is moving. The entities that understand this are not waiting for one. They are buying the infrastructure.

The lesson is not cynicism. The lesson is legibility. See the game clearly. The sovereign funds are not building UBI arguments. They are buying equity positions in the infrastructure layer. That is the tell. That is what understanding this transition actually looks like from the inside.

Source  OpenResearch / Eva Vivalt — UBI study results (July 2024)

What Seeing Clearly Demands

The founder this series is built for does not have sovereign capital. They do not have a $10 billion venture mandate. They are building in Lagos, in London, in Karachi, in Toronto, in Dubai, in Atlanta, in Nairobi. They have the capability. What they often lack is the structural positioning that turns capability into durable leverage.

The structural principle driving governments to move at this scale and speed is the same principle available to any founder willing to act on it: own a layer before it is priced in. Establish a position in the AI economy, in a niche, in a community, in a named method, in an audience, before the consolidation makes that position unavailable or prohibitively expensive to build.

The window is not infinite. Markets discover value and price it. Communities form and fill. Categories solidify. The founder who moves in 2026 is not moving at the same cost as the founder who moves in 2029.

Nobody is coming with the cavalry at the speed this requires. The question is not who is going to fix this. The question is what you build while the window is still open and what it means to own something in an economy where everything else is renting.

‍

This is Part 2 of The Machine Doesn't Owe You a Seat.
Part 3 - Own a Layer Before the Window Closes is the operational close: what founders actually build, who is already doing it, and why the window for this specific move has a date on it. Subscribe to get it direct.