AI, cloud platforms and the concentration of information power

1/3 The message
2/3 The numbers
3/3 The names

1/3;
A small cluster of companies is quietly inserting itself into the most important layer of modern civilisation: the layer between citizens and knowledge, between institutions and decisions, between governments and operational reality.

Control no longer needs to look like censorship.
It can look like classification, ranking, routing, access management, model policy, safety review, compliance architecture and enterprise risk mitigation.

Information does not need to disappear when it can be delayed, filtered, deprioritised, reframed, priced, monitored or made technically inaccessible to everyone outside the approved institutional class.

The more plausible danger is cold, clean and far more bureaucratic: a privately owned control layer through which research, intelligence, public administration, defence, finance, healthcare, media and political decision making increasingly pass.

Once that layer becomes infrastructure, power changes form.

It no longer needs to announce itself. It becomes embedded in dashboards, APIs, audit logs, permission systems, safety policies and subscription tiers.

It becomes the invisible architecture through which reality is processed before ordinary people are even allowed to see it.

The new ruling class will not need crowns, uniforms or speeches.

It will sell access, call restriction protection, call dependency innovation and tell the public that every locked door was built for their safety.

2/3 The numbers

The evidence is not hidden;

In 2025, more than 90% of notable frontier AI models came from industry, not public science.

Organisational AI adoption reached 88%. The cloud layer underneath this shift is already concentrated too:

Amazon, Microsoft and Google held 63% of enterprise cloud infrastructure spending in Q3 2025, in a market worth 107 billion USD for that quarter alone.

Palantir reported 1.63 billion USD revenue in Q1 2026, including 687 million USD from US government revenue, and closed 206 deals worth at least 1 million USD.

Anthropic’s Fable 5 can route many biology and chemistry requests away from the stronger model.

X can use public data and Grok interactions for training unless users opt out.

Tesla collects vehicle data such as speed, charging history, safety events and camera images, while location services rely on GPS, Bluetooth, IP address, WiFi and mobile towers.

This is a pattern.

The model layer, the cloud layer, the state software layer, the social graph and the physical sensor layer are converging.

Each piece is defended as safety, efficiency, innovation or national security. Together, they create something far more serious: privately owned infrastructure for deciding what institutions see, what citizens can access and what reality is allowed to look like after it has passed through corporate filters.

The new ruling class will not seize power with a speech.
It will inherit it through contracts, APIs, dashboards, data retention policies and model access tiers.

3/3 The Names

Sam Altman; not just the face of a chatbot.

He is the political salesman of frontier AI, the man who turned safety language, consumer addiction, enterprise deals and government proximity into one pipeline.

Dario Amodei sells Anthropic as the safer version of the same future, with Daniela Amodei close to governance and expansion. The language is cleaner, the product is calmer, but the structure remains brutal: someone decides which model you get, which answer you receive and which capability you are considered mature enough to access.

Alex Karp and Peter Thiel represent the institutional side of the machine.

Palantir is not trying to entertain teenagers with a chatbot. It sells operational vision to governments, defence, intelligence, police adjacent systems and large corporations. Karp provides the philosophical theatre. Thiel provides the old PayPal, venture capital and statecraft architecture.

Different costume, same question: who gets to see the whole map?

Demis Hassabis brings scientific prestige.

Google DeepMind can present itself as civilisation level research, not platform capitalism. But behind DeepMind stands Google: search, YouTube, Android, cloud, ads, location, mail and one of the largest information indexing systems humanity has ever normalised.

When that stack talks about intelligence, it is not speaking from a neutral laboratory.

Satya Nadella and Andy Jassy are the infrastructure men. They do not need to win every public AI product. Microsoft and Amazon already sit inside the boring layer everyone depends on: cloud, identity, office workflows, databases, enterprise security, procurement and compute.

The boring layer is where power becomes permanent.

Zuckerberg owns social scale and identity infrastructure.

Musk owns a more disturbing hybrid: X, Tesla, xAI, satellite infrastructure, defence relevance and political proximity through DOGE under Trump.

Arthur Mensch, Aidan Gomez and Aravind Srinivas are often treated as cleaner alternatives.

Mistral sells European sovereignty.

Cohere sells enterprise AI.

Perplexity sells answer infrastructure.

But they operate inside the same gravitational field: models, compute, access tiers, enterprise contracts, investor pressure and the race to become the interface through which humans query reality.

The pattern is that too much civilisational power is being compressed into a tiny executive class of founders, CEOs, investors and state connected operators.

They speak different dialects: safety, openness, sovereignty, productivity, defence, discovery, efficiency.

The destination is increasingly similar: a world where knowledge, compute, institutional visibility and decision making pass through private gates