We measure infrastructure dependency, bloc alignment, and trajectory across 199 countries. Uniform, neutral criteria applied to every one, continuously updated.
July to September 2026
Countries are struggling to navigate AI sovereignty. Companies are deciding where to invest. A peace deal was acknowledged to include access to chips.
29 countries sign WAICO's founding agreement
Kazakhstan becomes the only country in both WAICO and Pax Silica
Japan's ¥1 trillion Noetra consortium begins work on a foundation model
Poland's AI Systems Act enters into force
Brazil buys AI infrastructure from the US and China, the same day
Armenia calls its chip and AI agreements with the US “peace dividends”
Malaysia weighs Huawei chips for its sovereign AI cloud
Google pledges €13 billion for data centres in Finland
Annual reports tell you where things were. Newsfeeds tell you where things are.
Reports, studies and news take time. By the time analysis is published, events have already moved on. The strategic windows close in the gap between snapshots: in regulatory formation, infrastructure dependencies, and capital flows that don't surface in mainstream analysis until after alignment has already shifted.
The picture you are looking at
Time between the end of the period each picture shows and today.
AI-Atlas measures the motion.
What separates the leaders
The move that matters is often in the country nobody was watching.
A conclusion is worth more early. Ours is on the record before the headline.
Announced is not delivered. The quiet builds count too.
Figures from What Changed, July to September 2026, and the case studies below.
See it first
Each lane starts on the date the question was in our reading and ends on the date it became headline news. Across 5 recent cases, the median lead is 24 days.
Case Studies
By the time a narrative consolidates, the signal that mattered was already visible. These are the entries where our research had already read the situation, on the record, before the headlines arrived.
One reading, start to finish. What Hungary declares, and where its dependencies actually run.
Jensen Huang says Kazakhstan will become an AI factory hub for Europe. Our reports have tracked that ambition since April. What decides it now is delivery: financed facilities, accelerators in use, and partnerships that survive the US–China contest.
Google's €13 billion announcement read like a done deal. Our reports read it as a commitment still conditioned on permits and grid connections. In October, those conditions became a concrete obstacle.
Indonesia does not want to choose between US and Chinese AI ecosystems. It is already benefiting from both.
No single metric can capture AI sovereignty. So we built a framework that embraces the friction.
AI-Atlas tracks the global AI landscape across 199 countries using four distinct lenses: capital, infrastructure, regulation, and public discourse. Because these forces often pull in opposite directions, our model doesn't blend them into a polite average. We surface the disagreements.
By measuring the tension between present-state evidence and forward-looking signals, AI-Atlas gives you a structured, unfiltered view of where a nation stands, and where it's hesitating. It's an evolving instrument for an unpredictable era.
Who it is for
Two ways to work with us
Quoted to your scope
DTJ
Geopolitical Intelligence · AI Systems Architecture
Engineering and international relations, one desk. AI-Atlas pairs machine breadth (every country, every cycle) with human judgment concentrated where it changes the answer. Independence is the operating model: the same neutral read for all 199, with no mandate deciding what gets covered.
Launched in Tokyo, July 2026. Read the announcement →
DTJ · AI-Atlas
The AI-Atlas is designed around disciplined sourcing, cross-validation, and human review. Our assessments draw upon multiple independent sources and perspectives rather than relying on a single dataset or model. Where evidence is uncertain or conflicting, we make that visible rather than forcing artificial certainty.
We do not claim privileged knowledge. Forecasting is inherently uncertain, and conventional analysis can sometimes reproduce the assumptions of those producing it. This is often why traditional analytical products dismiss data points that contradict conventional views as wrong and miss big events. We reduce this risk by relying on multiple data points. The purpose is not to predict a single future, but to identify plausible directions of travel and developments decision-makers may overlook.
The AI-Atlas is not an LLM. Findings are grounded in identifiable sources and checked against multiple datapoints. Where evidence cannot be adequately substantiated, we do not treat it as an established finding. All analysis is also subject to human review.
No, sources are assessed in context rather than treated as equally authoritative. The AI-Atlas does not rely on a single publication or source type. Claims are always checked against multiple sources.
Because a single score can sometimes obscure more than it illuminates. For example, a country may be closely aligned with one bloc politically, while depending on another for cloud infrastructure, semiconductors, investment or AI models. Collapsing these different readings into a single number can create a misleading impression of coherence where significant tension exists. For decision-makers, understanding why a country is moving is more important than knowing whether it ranks 32 or 33 on an index.
AI-Atlas is a research-grade intelligence product. Access is by invitation or subscription.