TL;DR
Thorsten Meyer AI’s new Control Series argues that recent 2026 events show AI no longer behaves like a neutral utility. The report identifies six chokepoints: power, compute, data, model access, distribution and capital.
Thorsten Meyer AI has published the opening article in its Control Series, arguing that events in 2026 exposed six AI chokepoints where access can be granted, priced, limited or revoked, a shift that matters for companies and governments relying on frontier AI as stable infrastructure.
The article frames the shift through several examples it attributes to recent reporting and institutional statements: a frontier model allegedly switched off worldwide on about 90 minutes’ notice; Ukraine’s Ministry of Defense treating combat footage and annotations as licensed training material; and large AI firms renting supercomputing capacity from direct rivals under contracts that may allow the owner to reclaim capacity under certain conditions.
The six chokepoints identified are power, compute, data, model access, distribution and capital. On power, the piece points to SpaceX’s Memphis complex moving toward roughly two gigawatts with on-site gas generation, framing fast permitting and private generation as a gate on who can scale AI systems. On compute, it cites xAI’s Colossus at roughly 555,000 GPUs and states that Anthropic and Google agreed to payments near $1.25 billion and $920 million a month, respectively, for capacity from the cluster.
For data, the article highlights Ukraine’s Avengers Labs and says combat data can be licensed for training while Ukraine keeps the improved model. For distribution, it points to a $60 billion valuation around the coding interface Cursor as part of its argument that user-facing software can be more strategically valuable than a model alone. For capital, it describes roughly $26 billion a year in intra-industry financing tied to AI infrastructure, based on its cited sources.
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.
AI Access Becomes A Bargaining Tool
The article’s claim matters because many companies are building products as if frontier AI access will remain broadly available on predictable terms. If access depends on power permits, GPU clusters, unique datasets, model policy decisions, app distribution or financing relationships, customers may face outages, price changes or limits driven by contracts and governments rather than model quality alone.
That has consequences for buyers, developers and public agencies. A company using AI for coding, customer support or research may need backup providers, local models, data rights clauses and clearer service commitments. For governments, the same chokepoints raise policy questions about energy use, national security, procurement and who gets access to frontier capability.
high capacity GPU server for AI training
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The 2026 Cases Cited
The Control Series is presented as an occasional set of articles on where power sits in the AI stack. This first installment is an index rather than a full case study; the source says each chokepoint will receive a separate article later.
The cited sourcing spans March through June 2026 and includes 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 excerpt does not reproduce those underlying documents, so the specific numeric claims remain attributed to the article and its cited sources.
“AI does not flow freely like a utility.”
— Thorsten Meyer AI, on the article’s central premise
industrial power generators for data centers
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Contract Details Still Missing
Several details are still not established in the excerpt provided. It does not name the government that switched off the frontier model, identify the model, give the legal basis for the reported worldwide shutdown or provide the full contract language for compute repossession clauses. It also does not show how much of the reported $26 billion in annualized payments is binding versus projected.
The larger question is whether these events are one-time cases from a fast-moving market or the start of a durable structure in which access to AI systems is routinely gated by a small group of asset owners, states and platforms.
As an affiliate, we earn on qualifying purchases.
Future Installments Test The Thesis
Thorsten Meyer AI says this is Part 1 of The Control Series, with later installments planned for each chokepoint. The next test will be whether future pieces document the contracts, policy decisions and ownership patterns behind the six layers in enough detail for readers to separate hard constraints from market positioning.
Readers should watch for three markers: whether governments use model access limits again, whether major labs keep renting core compute from rivals, and whether data owners such as defense agencies, publishers or platforms retain rights to model improvements made from their material.
enterprise AI distribution platform
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What are the six AI chokepoints in the article?
The article lists power, compute, data, model access, distribution and capital. It argues each layer can be used to permit, price, slow or stop access to AI systems.
Is this a breaking news story?
No. It is best treated as analysis built on reported examples. The existence of the article and its framework are confirmed by the supplied source; individual claims about contracts, payments and shutdowns are attributed to the article and the outlets or institutions it cites.
What figures does the source cite?
The article gives specific reported figures, including about two gigawatts for SpaceX’s Memphis power buildout, about 555,000 GPUs for xAI’s Colossus, and monthly compute payments attributed to Anthropic and Google.
Why could this matter for AI customers?
Customers may face dependency risk if critical AI functions rely on suppliers that can lose power access, compute capacity, data rights or model permissions. The article’s practical point is that redundancy and contract terms may matter as much as benchmark scores.
What happens after this first installment?
The source says later articles will examine each chokepoint on its own. Those pieces would need to add names, dates, documents and counterparty responses for the broad thesis to hold up as reporting, not only market analysis.
Source: Thorsten Meyer AI
