Where value migrates.
The billions do not flow evenly. They gather at bottlenecks and where AI is first translated into productivity. Here we show the themes that emerge from this and what role companies play in them.
We classify every company by one question: where does AI work in its business? In sales to the AI industry, in its own product or in its existing business.
AIThree roles, three waves.
First to benefit are the enablers, who supply what AI requires. Then the innovators, who sell AI as a product. The third wave is the broadest and most enduring: the adopters, who deploy AI within their own operations.
Supplying what AI needs
Chips, energy, cooling, raw materials. Their customers are AI laboratories and hyperscalers, directly or one step removed. They earn first, because every expansion buys from them.
Sell AI as a product
Software, robotics, diagnostics, compute. Without AI, their offering would not exist. Who the customer is makes no difference.
Deploying AI within their own business
Banks, industry, retail. AI lowers their costs or makes their product more valuable. This is the broad, lasting part of the story.
The roles are not a ranking. A bank with documented AI savings counts as much as a chipmaker. Click a theme to open the dossier.
14 investment themes.
EnablersSupplying what AI needs3 themes · first wave
Chips, memory, manufacturing equipment: without them, no AI model runs. The largest share of the first wave flowed here.
Power, cooling, grids: the hardest bottleneck. Data centres are growing faster than the grids that supply them.
Copper, lithium, rare earths, silicon. Every data centre and every power grid begins in a mine.
InnovatorSell AI as a product4 themes · second wave
Models, cloud, data and agent platforms. This is where computing power becomes billable work — services that companies pay for.
Robots, machine vision, sensors. AI leaves the screen and works in factories, warehouses and hospitals.
AI-assisted drug discovery and diagnostics. Medicines and findings that could not exist without algorithms.
The more machines decide, the more valuable protection becomes. AI-assisted defence for an economy that runs on software.
AdoptersDeploying AI within their own business7 themes · third, broadest wave
Payment networks, exchanges, data providers. AI cuts fraud and process costs at the core of the financial system.
Mechanical engineering, vehicles, aviation, rail: AI reduces waste, downtime and maintenance costs.
Assortment, pricing and logistics are becoming algorithmic. Whoever masters this first wins margin.
Device and technology groups making AI productive within their own operations.
Banks and insurers, using AI in analysis, advisory and claims processing. Measurable in lower cost per case.
Media and network operators: AI in recommendation, production and network management.
Pharma and healthcare providers. AI shortens research timelines, sharpens diagnostics and reduces the administrative burden.
How we classify.
Two questions for every companyDoes it fit its role? How reliable are the figures?
The role says where AI operates within a business. Two further judgements say how well that is evidenced. We keep them separate deliberately: does the company fit its role? And how robust are the underlying figures?
Does it fit the role?
For innovators, what counts is the share of the AI product. Core: at least half of revenue. Segment: a reported AI line representing 10 to 50% of revenue, at least half of growth, or more than $1bn per year. For enablers and adopters: full, limited or Borderline.
How robust are the figures?
High, where the company reports the figure itself. Medium, where it is derived from segment data, order backlog or management statements. Low, when only third-party estimates are available. A threshold narrowly missed does not make the data worse.
A borderline case is not a verdict but a deadline. Within one quarter it is confirmed, assigned a different role, or removed.
Announcement is not impactFour steps from use to impact in the figures
For adopters, AI cannot be read from a revenue share. The effect lies in cost, price or revenue, and the evidence is thin: data from McKinsey more than 5% of their operating profit to AI. Some 95% of productivity claims made on US corporate earnings calls relate, according to the Federal Reserve Bank of St. Louis Profits that are yet to come. That is why we assess each stage in turn.
Expenditure, partnerships or launches are documented.
Management names a target figure: saving, premium, conversion.
One achieved effect is named. At Walmart, customers using the AI assistant spend about 40% more per order.
The effect shows up in margin, revenue per customer, or cost ratio. An initial read-across to the 2026 annual reports will provide a first test.
AI makes internal processes faster and cheaper. UPS saves around 100m miles a year through its route-planning system.
AI makes the existing product more valuable: higher revenue per customer, higher prices, more volume.
AI replaces the business itself, as in tutoring or translation. We monitor these companies; they do not enter the index.
As at 16 October 2026: of 75 adopters in screening, 22 reach level 3, 38 stand at level 1 or 2, and 15 are borderline cases. Methodology: AI Screening v1.3.
Four cases that show how the rules workChipmaker but adopter, compute as a product, consultant, riser
It makes chips, but for AI in cars and factories, not data centres. AI makes its products more valuable. That is why it is an adopter, not an enabler.
Sells AI computing power as a service, at close to 100%. Hence Innovator, Core.
Reported AI revenue was just under 4% and has not been disclosed since spring 2026. The business sells consulting. Classified as Adopter, borderline case.
The agent platform generates over $1.5bn in annual revenue. Reclassified on 16 October 2026 from Adopter to Innovator, Segment.
Where expansion stalls, pricing power emerges.
The highest margins arise where expansion cannot keep pace. Today that means power and grid connections, copper and strategic metals, advanced chip manufacturing, and cooling and water. Bottlenecks shift. That is why we review them every quarter.
What doesn't make the cut
A well-known company with no verifiable AI connection is not included. For Adopters, the evidence ladder shows how well the effect is substantiated. Weak evidence does not disqualify a company, but it reduces its weight. Any company whose business AI replaces rather than strengthens stays out. These companies remain in the screening process; they never enter the index.