THE FRONTIER SIGNALSWhat turns AI investment into useful economic output?
Capital, electricity, chips and cloud capacity enable models and applications, but supplier revenue and adoption alone do not establish customer returns.
Open interactive Economics hub →
Capital: what comes in and what comes out?
Equity, debt and operating cash → Funding for infrastructure and development
$122B committed in OpenAI’s March 2026 round. Committed capital is not revenue or cash already spent.
Compare cash actually raised and spent with revenue from external customers.
OpenAI · Funding round ↗Power & sites: what comes in and what comes out?
Electricity, land, grid connections and cooling → Energized data-center capacity
485 TWh global data-center electricity in 2025. All data centers, not AI alone. Electricity use is not available capacity.
Track commissioned megawatts and utilization, not just announced gigawatts.
IEA · Energy and AI ↗Chips & memory: what comes in and what comes out?
Silicon, packaging, high-bandwidth memory and networking → Installed accelerator systems
$89B NVIDIA Data Center revenue · Q2 FY2027. One supplier’s segment revenue; not a measure of all chip or memory output.
Look for delivered systems and customer concentration. No comparable memory-capacity series is included yet.
NVIDIA · Q2 FY2027 results ↗Cloud: what comes in and what comes out?
Compute systems and operating infrastructure → Rented training and inference capacity
AWS primary training/cloud relationship with Anthropic. Documented relationship; no comparable cloud utilization figure in this edition.
Compare utilization and cash flow with depreciation, energy costs and capital spending.
Anthropic · Series H ↗Models: what comes in and what comes out?
Compute, data and research → Model weights, APIs and task capabilities
~12-hour Opus 4.6 horizon at 50% success. May 2026 TH1.1 snapshot; selected tasks and substantial failure, not dependable autonomous work.
Keep model, harness, budget and evaluation conditions comparable.
METR · Current TH1.1 ↗Applications: what comes in and what comes out?
Models, tools and workflow integration → Products, paid usage and economic outcomes
88% surveyed organizational adoption · 2025. Adoption does not establish productivity, retention or return on investment.
Track verified time saved, error costs, retention and margins. Comparable application ROI data is not included.
Stanford HAI · 2026 AI Index ↗What else do readers ask?
What are the six layers of the AI value chain?
Capital funds power and sites, chips and memory, cloud capacity, models and applications; these are dependencies, not measured financial flows.
Does AI infrastructure revenue prove AI investment pays off?
No. Supplier revenue can grow before customers establish profitable, durable demand.
Is announced capacity the same as operating capacity?
No. A plan or commitment must be distinguished from delivered equipment and energized, utilized capacity.
Does the electricity KPI measure AI alone?
No. The IEA figure shown covers all data centers, including non-AI workloads.
Are the scenario sliders forecasts?
No. They apply an explicit growth assumption to a historical revenue anchor without modeling demand, margins or valuation.
Which indicators would strengthen the investment case?
Verified productivity, paid retention, utilization and cash flow after costs would be more informative than adoption or announcement counts alone.