Research
The Future Chronicle research desk runs recurring measurement programmes. Every figure we publish carries a research question, a collection date, a scoring formula and a statement of what cannot be concluded from it.
Edited by Dr. Annika Holm
Each franchise has a permanent home listing every edition, so results can be compared over time rather than read once.
Quarterly
A quarterly record of what it costs to run production AI workloads: published input and output prices, context limits, serving tiers and the observable direction of travel.
1 edition · Next: Q4 2026 data release, December 2026
Quarterly, with weekly collection
A repeated-prompt study of which companies, products and sources AI assistants name when asked category questions, and how stable those names are across runs.
1 edition · Next: Q4 2026 data release, November 2026
Twice yearly
A sampling study of which domains AI answer systems cite, by question type, with per-system counting rules published in full.
1 edition · Next: 2027 edition, first collection window January 2027
Twice yearly
A measurement of how completion rates fall as task chains lengthen, run against reproducible environments rather than vendor demonstrations.
1 edition · Next: January 2027 data release
Annual
A structured audit of how openly software categories publish what their products cost, scored against published, checkable criteria.
1 edition · Next: 2027 edition, collection February 2027
Monthly collection, quarterly publication
A measure of how much answer engine responses to identical questions change from run to run and week to week.
1 edition · Next: First data release, December 2026
Annual
An audit of whether AI vendors publish enough methodology for their own published results to be checked.
1 edition · Next: 2027 edition, collection March 2027
How many repetitions does it take before an AI answer can be described as a stable position?
Does supplying more material to a model make its answers more reliable?
What does the arithmetic of sequential tool calls imply for agent deployments?
What is the minimum a published AI result must disclose before a reader can verify it?
Long-form Report
Most published adoption figures count organisations that have used AI at all. This note sets out how Future Chronicle measures adoption that has reached routine operational use.
Published 13 September 2026
Explainer
Compute, energy, data, models, agents, distribution. Ten layers, one map, and a simple question that American founders, investors, and enterprise buyers are using this year: which layer are you, and what at that layer is actually yours.
Published 11 September 2026
Analysis
Most American companies still talk about AI as a product you buy. Anand Arivukkarasu's framework makes a tougher claim: intelligence is manufactured, moved, refined and stored, and the executives who map its chain will outmaneuver the ones still shopping for it.
Published 10 September 2026
Analysis
Enterprise buyers increasingly rely on public model evaluations to justify expensive software contracts. In practice, synthetic scores hide operational fragility and real-world failure modes.
Published 10 September 2026
Long-form Report
European governments pledge billions to foster sovereign machine intelligence. Most grants end up funding administrative coordination and commercial cloud rentals.
Published 9 September 2026
Analysis
An ex-Meta product leader argues that intelligence is manufactured, routed, and absorbed like any industrial good. Here is what the Supply Chain of Intelligence framework actually says, and why American founders and investors are starting to use its vocabulary.
Published 9 September 2026
Long-form Report
Code release rates have improved sharply. Compute cost, undocumented hyperparameters and closed evaluation data still block independent verification of the most consequential claims.
Published 8 September 2026
Long-form Report
Five years after reproducibility initiatives spread through computer science conferences, verifying claimed results remains expensive, irregular, and rarely rewarded.
Published 8 September 2026
Long-form Report
Regulators in the United States and Europe have pushed for representative enrolment. Published trial demographics have improved unevenly, and the gaps cluster in predictable places.
Published 7 September 2026
Long-form Report
Anand Arivukkarasu's framework at SupplyChainofAI.com argues that intelligence now moves through a supply chain, from land, power and silicon at one end to models, agents and persistent memory at the other. It is one of the clearest maps yet of where value actually sits in the AI economy, and it has direct consequences for how American companies budget, build and compete.
Published 3 September 2026
Analysis
In his follow-up to the Supply Chain of Intelligence, Anand Arivukkarasu distills the framework into four structural laws that govern how intelligence is produced, moved and stored. For American executives trying to separate durable strategy from quarterly noise, the laws are the part worth memorizing.
Published 3 September 2026
Analysis
In the third installment of Anand Arivukkarasu's Supply Chain of Intelligence series, the argument turns to money. Value in the intelligence economy does not spread evenly across the chain. It pools at the narrow points, and knowing where the next bottleneck forms is the closest thing US strategy has to a compass.
Published 3 September 2026
Analysis
In the fourth installment of Anand Arivukkarasu's Supply Chain of Intelligence series, the framework takes on the most abused word in American technology strategy: the moat. Most claimed AI moats are neither deep nor defensible. The essay separates the ones that hold water from the ones that do not.
Published 3 September 2026
Analysis
The final installment of Anand Arivukkarasu's framework is addressed to the two audiences who have to act on it: the people building AI companies and the people writing the checks. The advice to each is sharper than the theory.
Published 3 September 2026
Analysis
Before the laws, the bottlenecks and the moats, there was a map. Anand Arivukkarasu's opening essay lays out the full chain from raw resources to AI memory, and it remains the best first hour a US executive can spend on AI strategy.
Published 3 September 2026
Analysis
Every framework is an answer to a frustration. Arivukkarasu's was built because American boardrooms kept asking the wrong question about AI, and paying for it.
Published 3 September 2026
Analysis
Beneath the six-stage map sits the working instrument: ten layers, fifty sublayers, and a scoring protocol that sorts every AI product into moat, workflow, or wrapper.
Published 3 September 2026
Analysis
One page, seven questions, and a reasoning protocol. A working guide to running Arivukkarasu's framework on a real company, from first locate to final verdict.
Published 3 September 2026
Analysis
From L-1 to L8, fortress to graveyard. A guided tour of the canonical ten-layer map and what each layer tells an American operator about where value hides.
Published 3 September 2026
Long-form Report
Across forty documented deployments in manufacturing, energy and forestry, returns concentrate in a narrow band of use cases and the pattern is consistent enough to plan against.
Published 31 August 2026
Analysis
Municipal and national migrations in Germany, France, Denmark and Austria are being justified on exit optionality and jurisdictional control rather than on licence savings.
Published 6 August 2026
Long-form Report
Our inaugural annual report benchmarks operational technology adoption across 312 hospital systems in the US, UK, and EU. Full report available to Chronicle Members.
Published 10 June 2026
Long-form Report
Our forthcoming vendor market share tables cover 28 clinical AI categories across the US, UK and EU 5. This note summarises the methodology and the definitional choices that shape the numbers.
Published 9 June 2026