TL;DR
Thorsten Meyer AI’s Control Series frames recent 2026 AI events as evidence that control over the stack is narrowing around six chokepoints. The report points to power, compute, data, model access, distribution and capital as places where access can be gated, repriced or withdrawn.
Thorsten Meyer AI has published the first installment of its Control Series, identifying six AI chokepoints that it says became active sources of power in 2026: energy, compute, data, model access, distribution and capital. The piece matters because it argues that access to advanced AI is increasingly controlled by a small set of governments, infrastructure owners, platform companies and funders rather than functioning like a neutral public utility.
The report says the “utility” story around AI weakened after several 2026 developments showed that access can be cut off, rented under conditions, or tied to ownership of scarce infrastructure. It cites a government-ordered shutdown of a frontier model worldwide on about 90 minutes’ notice, Ukraine’s licensing of combat data through Avengers Labs, and large compute rental deals involving xAI’s Colossus cluster.
According to the source material, xAI’s Colossus holds about 555,000 GPUs, with Anthropic agreeing to pay roughly $1.25 billion a month for output from Colossus 1 and Google signing a separate deal near $920 million a month. The report frames those figures as evidence that even leading AI labs may depend on infrastructure controlled by direct rivals or a small group of cluster owners.
The analysis also points to power as a base-layer constraint. It says SpaceX’s Memphis complex is moving toward roughly two gigawatts of capacity by using on-site gas generation rather than waiting for slower grid interconnections. That claim supports the report’s view that power permitting and financing are now part of AI market control.
The Six Chokepoints
For a decade AI was sold as a utility — abundant, neutral, always on. In 2026 it became a lever: scarce, controlled, revocable. Here are the six places power actually sits — and who started to squeeze.
Every layer is concentrating into fewer hands, and 2026 is the year the holders stopped treating their leverage as theoretical. A kill switch wasn’t discussed — it was pulled. The utility you’re allowed to forget about; the lever, you have to watch who’s holding. Optionality just became architecture.
Control Moves Upstream
The report’s central claim is that AI access is becoming scarce, controlled and revocable. If that view proves durable, companies building AI products may face risks beyond model quality or pricing: their access to power, chips, datasets, models, app stores, interfaces or financing could change quickly.
For readers outside the AI industry, the issue is not only which chatbot or model performs best. The more lasting question is who can decide whether AI systems remain available, what they may be trained on, where they can be distributed, and how much access costs. That has consequences for startups, enterprise buyers, governments and users who depend on AI tools for daily work.
ASUS Dual GeForce RTX 4060 EVO White OC Edition 8GB GDDR6 (PCIe 4.0, 8GB GDDR6, DLSS 3, HDMI 2.1a, DisplayPort 1.4a, Axial-tech Fan Design, Dual BIOS, Auto-Extreme Tech), 3 Year Warranty
AI Performance: 246 AI TOPS
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Six Layers Of Leverage
The Control Series divides AI power into six layers. At the bottom is power, where companies able to secure gigawatts can set the ceiling for compute. Above that is compute, where large GPU clusters are concentrated among a limited number of owners and suppliers, including Nvidia upstream.
The third layer is data. The report highlights Ukraine’s Avengers Labs as an example of combat data being licensed under conditions rather than sold outright. The fourth layer is model access, where governments and labs can restrict use. The fifth is distribution, where the owner of the user interface or platform can shape demand. The sixth is capital, where the report points to large intra-industry financing flows and sovereign funds as another control point.
“AI does not flow freely like a utility.”
— Thorsten Meyer AI
AI compute cluster hosting
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Open Questions For Buyers
The report does not establish how permanent these chokepoints will be or whether new suppliers, regulation or technical changes could reduce them. It is also not yet clear how many AI customers face direct exposure to revocation clauses, government shutdown orders or data-use restrictions.
Several claims in the source material rely on reported deal values, company statements and outside sourcing cited by the author, including Anthropic statements, Axios, The Wall Street Journal, Reuters, CBS, TechCrunch, Semafor, Ukraine’s Ministry of Defense, Perplexity Research, Challenger Gray and SpaceX SEC filings. The full contractual terms behind some of the cited arrangements are not public.
Data Center Design Guide – Book 1: Power: Specifying the requirements, power generation, power distribution, power efficiency, and fault tolerance for data centers
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Series Turns To Each Chokepoint
The first installment is framed as an index for a longer Control Series. Future pieces are expected to examine each chokepoint separately, including who holds leverage, how it is being used, and what that means for AI companies and customers.
For now, the practical next step for readers is to watch whether access limits become more visible in contracts, regulation, cloud capacity, data licensing and platform rules. The report’s thesis will be tested by how often these levers are used, not merely by whether they exist.
Access Control Systems: Security, Identity Management and Trust Models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is the main claim of the report?
The report claims that advanced AI is no longer best understood as a neutral utility. It argues that control now sits in six chokepoints: power, compute, data, model access, distribution and capital.
What events does it cite as evidence?
It cites a frontier model being switched off worldwide on about 90 minutes’ notice, Ukraine licensing combat data through Avengers Labs, and large compute rental deals involving xAI’s Colossus cluster.
Why does compute matter in this story?
Compute matters because frontier AI systems require large GPU clusters. The report says ownership of those clusters is concentrated, meaning labs may depend on rivals or a small set of infrastructure providers.
What is still unknown?
It remains unclear how widely revocable-access clauses or similar controls appear in AI contracts, and whether new entrants, policy changes or new hardware supply could weaken these chokepoints.
What happens after this first installment?
Thorsten Meyer AI says each chokepoint will be examined in a later installment of the Control Series.
Source: Thorsten Meyer AI
