It’s the robber barons, rails, rents and defaults of the next twenty years I worry about, not skynet, the singularity or misaligned paperclip optimizers.
Infrastructure ownership: it’s the rails, not the robots
Infrastructure capture refers to the consolidation of essential systems, such as transport, energy, communications, or now AI tech stacks, under the control of a small number of private actors. Historically, this pattern has produced enduring power asymmetries. When a resource becomes indispensable for competition in the market and cannot be reasonably duplicated, its owner wields structural power. When public infrastructure gets privatized or commodified into a corporate chokepoint, those who own it do not merely extract rent; they set the terms of participation in economic and civic life.
A classic example is the Gilded Age railroad monopolies in the United States. By the late 19th century, a handful of companies controlled the rail network, enabling discriminatory pricing and political leverage while producing the largest wealth inequality between a small elite and masses in poverty.
We have seen the same more recently with the big social media platforms, that for too long were treated as consumer products, when in reality they aimed to become the roads and bridges of digital life for most of us while extracting a heavy rent on our attention. That rent did not only make them filthy rich but also extremely powerful and politically influential, enshrining continued exploitation of their users data as their business model.
AI infrastructure magnifies these stakes, as AI tech stacks are becoming the interface to information creation, dissemination and consumption. If a few companies own the chips, data centers, cloud compute, foundational models, model weights, and distribution channels, they effectively control any future useful AI application build on top of it; and with it, the epistemic and economic arteries of the 21st century.
The gamble to become the next Vanderbilt, Rockefeller or Carnegie’s of the info age are long underway. Current tech giants have already captured chips and cloud computing, that is why Nvidia’s data‑center revenue hit $18.4 billion in a single quarter—up 409% YoY as AI hyperscalers started hoarding H100 chips like potable water before a drought. Amazon plowed $8 billion into Anthropic so Claude will use Amazon Web Services (AWS) as its primary cloud and training partner, and will utilize Amazon’s custom AI chips. OpenAI will purchase an additional $250 billion of Microsoft Azure cloud compute and locked its model distribution to Azure’s cloud for the next decade.
[…]
This brings us to the next point:
Strong-arming society into technological lock‑in
Infrastructure capture is not just about finding one bottleneck to own, because bottlenecks can be remedied with just building more capacity over time through market competition or government investment. Real infrastructure capture today is much more about owning the full value-chain, tech stack or ecosystem. That is why the current AI players aim to cement vertical control—chips → cloud → model → API—where each layer enforces the others and makes competition almost impossible. The current AI web of partnerships and strategic investments have only one goal: to entrench big tech incumbents across multiple layers of the value chain.
[…]
If the tech moguls win their gambit, if they succeed at infrastructure capture, technological lock-in and public dispossession of our shared information ecosystem, we will have ceded more than we have bargained for. Once control over knowledge creation, distribution and consumption (and with it increasingly cognitive agency) is lost for most of us, what awaits is a new dark age of monarchs, myths, manipulation and magical thinking.
A handful will own, a tiny minority will benefit, and the rest of us will fall into exploitation, exclusion or serfdom.
https://www.protagonist-science.com/p/the-ai-race-to-reboot-feudalism