TL;DR
Thorsten Meyer AI’s Control Series Part 1 says 2026 events have exposed six AI chokepoints: power, compute, data, model access, distribution and capital. The confirmed development is the publication of that framework; the examples cited in it depend on attributed reports from companies, governments and news outlets. The takeaway for readers is that AI access may be gated or withdrawn by a small set of infrastructure, platform and finance owners.
Thorsten Meyer AI has published the first installment of its Control Series, arguing that 2026 events show artificial intelligence is no longer best understood as a neutral utility but as a set of controlled chokepoints spanning power, compute, data, model access, distribution and capital, a shift that matters because access to advanced systems can be throttled, repriced or withdrawn by a small set of actors.
The article, titled The Six Chokepoints of AI – The Control Series, Part 1, identifies six control points: power, compute, data, model access, distribution and capital. What is confirmed from the source is the series’ framework and its six named layers. The broader conclusion that those layers have become levers is the article’s analysis; individual examples should be read through the reports and documents cited for each case.
The source material points to several 2026 examples. It says a government switched off a frontier model worldwide on roughly 90 minutes’ notice; Ukraine’s defense ministry treated annotated combat footage as a licensed asset while retaining rights to improved models; and xAI’s Colossus cluster, described as holding about 555,000 GPUs, rented capacity to rivals including Anthropic and Google.
Some figures are attributed to the series’ cited sourcing from outlets and entities including Anthropic statements, Axios, The Wall Street Journal, Reuters, CBS, TechCrunch, Semafor, Ukraine’s Ministry of Defence, Perplexity Research, Challenger Gray and SpaceX SEC filings. Separate tech press coverage, including a Times of India report, has tied Cursor to a reported $60 billion SpaceX deal. The numbers are historical reports and should not be read as forecasts.
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 Above Models
The thesis matters because it moves the AI power debate away from model quality alone. If scarce electricity, owned clusters, exclusive datasets, platform control and financing decide access, then companies using AI face supplier risk even when the underlying models appear interchangeable.
For developers and enterprise buyers, the practical issue is revocability. A model endpoint can be shut off by a lab or government; compute can be reclaimed under contract; data access can be licensed only under conditions; and an application layer such as Cursor can shape which models users encounter first.
INFINIBAND FOR HIGH-PERFORMANCE COMPUTING AND AI CLUSTERS: Configure RDMA networking, optimize GPU interconnects, and build low-latency infrastructure for distributed training and HPC workload
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
From Utility Pitch To Chokepoints
For much of the past decade, AI companies described their products as always-available infrastructure, closer to electricity than to a scarce strategic asset. The Control Series argues that events in 2026 exposed the limits of that pitch.
At the base of the stack, the article cites power as the first limit, saying SpaceX’s Memphis complex moved toward roughly two gigawatts using on-site gas generation. Above that, compute ownership is described as concentrated among a few cluster builders, with Nvidia upstream as the dominant chip supplier named in the source material.
The data example centers on Ukraine’s Avengers Labs, which the source says converts annotated battlefield footage into training material under terms that let Ukraine keep improved models. The distribution example centers on Cursor, whose reported $60 billion value is used as evidence that the interface can be as contested as the model beneath it.
“AI does not flow freely like a utility.”
— Thorsten Meyer AI Control Series
Hewlett Packard Enterprise ProLiant Compute DL360 Gen12 w/one Intel Xeon 6530P Processor, 1P 2x32GB-R 8SFF NS204i-u v2 MR408i-o 2x1000W PS (HPE Smart Choice P89997-005)
HPE SMART CHOICE MODEL – P89997‑005 – ENTERPRISE 1U RACK SERVER Preconfigured and factory‑tested, this Smart Choice DL360…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Evidence Still Needs Detail
The article presents several events as evidence of a single shift, but some details remain dependent on the underlying reports it cites. It is not yet clear how long the reported model-access restrictions lasted, what contractual language governs compute recall rights, or how much of the cited spending will recur.
The source also links public-sector power to private-sector AI access, but the permit, grid and environmental implications of large on-site generation projects are still developing. The balance of control between governments, labs and infrastructure owners may vary by jurisdiction and by contract.
AI Data Center Infrastructure Engineering: Power Distribution, Liquid Cooling, High-Density Networking, and Energy Efficiency for GPU Training … Hardware & Compiler Engineering Series)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Series Turns To Each Lever
Thorsten Meyer AI says each of the six chokepoints will receive a separate installment. Readers should expect the next reports to test the thesis layer by layer, starting with the actors that can permit power, finance clusters, license rare data, approve model access, own user interfaces or supply capital.
The near-term test is whether the 2026 examples remain exceptional cases or become standard practice. Watch for new compute rental terms, government review processes for frontier models, sovereign data licensing deals and platform acquisitions that place more of the AI stack under fewer owners.
Models, Methods and Tools for Product Service Design: The Manutelligence Project (SpringerBriefs in Applied Sciences and Technology)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What are the six chokepoints named in the report?
Power, compute, data, model access, distribution and capital.
Is this a breaking news story?
No. It is an analysis pegged to a June 2026 Control Series installment and to events the source says occurred from March to June 2026.
What has been confirmed?
The confirmed development is the publication of the series installment and its stated framework. The individual examples are attributed to the source material and the outlets or entities it cites.
Why does the utility metaphor matter?
A utility implies broad, stable access. The report argues the current AI stack can be gated, repriced or shut off by actors that control scarce inputs.
What should readers watch next?
Watch for contracts and policy actions that define who can rent compute, keep improved models from licensed data, or suspend access to frontier systems.
Source: Thorsten Meyer AI
