ARTIFICIAL INTELLIGENCE / REAL ECONOMY
AI: who captures value as investment grows.
Rising AI usage does not guarantee attractive returns for every supplier. The opportunity requires separating installed capacity, effective adoption and profits remaining after the cost of capital.
The three perspectives
Wei Li · BlackRock
Her midyear framework identifies opportunities in physical AI constraints: power, grids, chips and data centers. It is used here as a dated structural framework, not a September update. [1]
Blackstone · operating evidence
Its September observations show AI experiments progressing toward deployment among portfolio companies. This is evidence from a particular business population, not a representative sample of the economy. [2]
Eric Sheridan · Goldman Sachs
Following Communacopia, he highlights implementation and consumer agents. Lower token costs and better utility are important conditions for broader adoption. [3]
Three links in the chain
Infrastructure sells capacity; platforms distribute tools; applications complete tasks. A technical improvement may help one link and compress another’s margins. Cheaper inference can improve an application’s economics while creating price pressure for capacity suppliers. A sector thesis should specify where economic profit is expected to remain.
Our interpretation: follow value to the customer
For an application, tokens consumed are less informative than value per completed task and its total production cost. Include human review, errors, retries and support. A time-saving pilot that requires permanent supervision may still be viable, but must prove that pricing covers the supervision.
Infrastructure: announced versus profitable capacity
For a data center, distinguish planned megawatts, grid connection, operational capacity and contracts with creditworthy customers. A project pipeline is not contracted revenue. Construction timing, financing costs and equipment replacement also determine returns. This is an evaluation framework; it does not imply that ADVANS owns data centers.
Applications: renewal tests the competitive advantage
Early customer cohorts reveal whether clients return, expand spending and recommend a product without increasing subsidies. Integration in critical workflows, data permissions and switching costs can matter more than a compelling demonstration. If a new model reproduces the core feature, the business must find its advantage elsewhere.
How the thesis could fail
Cost-disciplined adoption would favor solutions with measurable savings. Capacity growth ahead of demand could pressure prices and utilization. Frequent errors or unclear responsibilities could increase selling and operating costs. All three scenarios allow AI usage to rise, but produce different financial outcomes. Technological growth alone therefore cannot justify blanket investment conclusions.
Monitoring framework
Indicators to assess at the next review; this table does not display live values.
| Variable | What to check | What changes the interpretation |
|---|---|---|
| Conversion into revenue | Quarterly results and company-specific customer evidence. | Compare investment and backlog with collected revenue and operating margins. |
| Task economics | Pilots: total cost, success rate, human review and renewal. | Savings must survive support, quality control and lower commercial subsidies. |
| Capital and execution | Contracts, permits, grid connections and debt maturities. | Operating delays can consume the advantage suggested by strong demand. |
A defensible perspective identifies measurable value and an advantage that survives model improvements. For early opportunities, the bridge from technology to cash deserves more scrutiny than headline market size.