TL;DR
Recent 2026 events, led by the U.S.-ordered suspension of Anthropic’s Fable 5 and Mythos 5 models, show that access to advanced AI can be restricted quickly. Thorsten Meyer AI frames the shift as six chokepoints: power, compute, data, model access, distribution and capital.
A series of 2026 AI deals, shutdowns and state-linked data arrangements has sharpened a new contest over control of the AI stack: power, compute, data, model access, distribution and capital. The clearest confirmed flashpoint came June 13, when Anthropic disabled access to its Fable 5 and Mythos 5 models worldwide after a U.S. export-control directive, according to company statements cited by Tom’s Hardware and TechRadar.
The reported order barred access by foreign nationals, including some people inside the United States, and Anthropic said it could not apply that restriction selectively at launch speed. Axios reported, citing an Anthropic source, that the company was given about 90 minutes to take down the models after a government call warning of a national-security threat.
Thorsten Meyer AI’s Control Series uses that episode as the lead example in a broader claim: advanced AI no longer behaves like a neutral utility. The piece identifies six pressure points where access can be limited or repriced: electricity supply, GPU clusters, scarce training data, model access, user-facing distribution and capital.
Other examples in the source material are reported rather than fully public. They include SpaceX/xAI’s Memphis AI infrastructure, large monthly compute-rental arrangements involving Anthropic and Google, Ukraine’s use of combat data as a licensed AI asset, and a reported $60 billion Cursor-related distribution play. The common thread is that ownership or control at one layer can shape who gets to build, run or sell AI systems.
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 Leverage
The immediate lesson for companies and developers is operational: access to a model can change faster than procurement plans, product roadmaps or customer contracts. The Anthropic shutdown affected users because the legal order targeted a class of access that the company said it could not separate cleanly in real time.
The broader concern is market power. If only a small group can permit gigawatt-scale power, assemble frontier GPU clusters, license rare data, route users through dominant apps or finance multibillion-dollar compute deals, AI access becomes less like metered infrastructure and more like a set of private and state-controlled gates.
GPU cloud compute rental
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Utility Metaphor Breaks
For much of the generative-AI boom, companies described AI as infrastructure: always available, broadly neutral and paid for through subscriptions or API calls. Meyer AI argues that 2026 has exposed a different structure, where each layer has an owner able to slow, redirect or revoke access.
The power layer is tied to the ability to build or permit energy supply faster than local grids can serve data centers. The compute layer is tied to scarce GPUs and the companies that can finance clusters. The data layer is tied to material that cannot be scraped or bought easily, such as military sensor data. The model-access layer is now also a state-policy question, as the Anthropic order showed.
“AI does not flow freely like a utility.”
— Thorsten Meyer AI
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 Details Still Missing
Several facts remain incomplete. The full legal basis and technical evidence behind the U.S. directive have not been fully disclosed. Reports differ on the timing sequence between the government call and written order, and the public record does not yet show exactly how officials judged the alleged jailbreak risk.
Some other chokepoint examples rely on reported deal terms or source synthesis. The exact contract language for compute rentals, any reclamation clauses, the verified scale of some GPU and power figures, and the terms of data-licensing arrangements are not all public.
ASUS Dual AMD EPYC 9004 Series 4U NVMe 8X Dual Slot PCIe Gen 5.0 GPU Server (ESC8000A-E12P), 8X Trays, 2X H200 NVL Tensor Core 141GB HBM3e PCIe 5 Accelerator, Rails (Renewed)
No Processor Installed; Supports 2x AMD EPYC 9004 Series Processors
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Policy And Contracts Decide
The next test is whether Anthropic restores Fable 5 and Mythos 5 access, and whether U.S. officials create a clearer process for frontier-model restrictions. Companies using advanced AI will also be watching whether compute owners, cloud providers and app platforms write more explicit control rights into contracts.
For customers, the practical response is likely to be redundancy: backup models, portable workflows, open-weight options where acceptable, and closer review of vendor terms. The policy question is whether governments can set rules before the next emergency-style intervention.
The Azure Cloud Native Architecture Mapbook: Design and build Azure architectures for infrastructure, applications, data, AI, and security
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is the main news development?
The main development is the June 2026 restriction that led Anthropic to disable Fable 5 and Mythos 5 worldwide, combined with other 2026 examples showing control points across the AI stack.
What are the six AI chokepoints?
The six named chokepoints are power, compute, data, model access, distribution and capital. Each can limit who can train, deploy or monetize advanced AI.
Is every claim in the source confirmed?
No. The Anthropic shutdown is reported across several outlets. Some figures on compute rentals, data licensing and deal terms remain reported claims or source-synthesis items pending fuller public documentation.
Why should businesses care?
Businesses that rely on one AI vendor or one model can face sudden service loss, price changes or access limits. The Anthropic case showed that policy decisions can affect even domestic users when a provider cannot quickly segment access.
Source: Thorsten Meyer AI
