Overview
The current read still looks expansionary: labs are releasing product and finding paying demand, cloud demand remains firm, physical buildout is still moving even though the data-center evidence is narrow, and power remains the hardest bottleneck.
Application - Consumer
Core metrics
- Consumer AI usage (blended proxy): 850 million users in April 2026
What changed
- No major named change logged.
AI system impact
- Consumer AI usage (blended proxy) is 850 million users in April 2026. Even without a fresh daily change, that still points to a very large installed demand base for inference and distribution.
Human system impact
- This is where AI becomes ordinary behavior: searching, writing, studying, and asking for help. Today's read says that habit is already large, even if the daily social effects are hard to measure cleanly.
Application - Enterprise
Core metrics
- Enterprise AI adoption (blended proxy): 8 million seats in Q2 2026
What changed
- No major named change logged.
AI system impact
- Enterprise AI adoption (blended proxy) is 8 million seats in Q2 2026. That still matters because enterprise rollouts are what turn model capability into recurring software and cloud spend.
Human system impact
- This is the route by which AI moves from demo to workflow. When enterprise adoption rises, the real questions become budgeting, job design, training, and which teams are expected to do more with the same headcount.
Foundation Labs
Core metrics
- Model launches / upgrades: Mistral released another model, showing that frontier development outside the US is still active.
- Platform / app launches: No major named lab product launch was logged in the last 30 days.
- Enterprise / channel partnerships: No major named lab distribution or enterprise deal was logged in the last 30 days.
- Reach / adoption / ecosystem scale: Anthropic expanded its Google Cloud TPU usage and pointed to continued business-customer growth. This is directional rather than precise, because the disclosure does not tell us by how much business demand increased.
- OpenAI revenue estimate (secondary source): about $2.0bn a month, reported on March 19, 2026
What changed
- Commercial scale still looks real: Anthropic pointed to continued business demand, and secondary reporting still put OpenAI near $2.0bn of monthly revenue.
- Frontier product competition is still active outside the US: Mistral released another model.
AI system impact
- The lab layer is still doing two things at once: shipping new models and proving customers will pay. That combination keeps demand for cloud compute and advanced chips from easing.
Human system impact
- Today's lab read matters because commercialization is no longer hypothetical. If frontier products keep improving and still attract paying demand, more managers have a reason to test them against work currently done by people.
Cloud
Core metrics
- Demand / growth: Microsoft said Azure and other cloud services revenue grew 40% in latest reported quarter. Treat this as the clearest named demand signal in the current data, not as a full cloud-sector measure.
- Product / platform launches: No major named item was logged in the last 30 days.
- Customer / partnership wins: No major named item was logged in the last 30 days.
- Capacity / network expansion: No major named item was logged in the last 30 days.
What changed
- The clearest cloud change in the current data remains Microsoft's 40% Azure growth print; the rest of the cloud layer has not yet produced an equally strong named update.
AI system impact
- The cloud read is positive but narrow: one major provider is still growing very fast. That is enough to keep the infrastructure story expansionary, but not enough to say the whole cloud layer is accelerating in sync.
Human system impact
- If AI is mostly delivered through a few cloud platforms, then most firms adopt it as renters rather than owners. That affects pricing power, dependency, and how much smaller firms can build for themselves.
Data Centers
Core metrics
- Capacity under development: a named operator reported about 1.0 GW in February 2026. For one operator this is meaningful, but it is not enough to tell us the size of the full market.
- Named expansion signal: No additional large named data-center expansion item was logged in the last 30 days.
What changed
- No comparably large new buildout disclosure arrived in the last 30 days, so this layer still rests mainly on the latest stored operator update.
AI system impact
- The data-center layer still says physical expansion is continuing, but the evidence base here is narrow. We can say construction is still moving; we cannot yet say by how much at the full market level.
Human system impact
- This is where AI stops being abstract and becomes local: land deals, construction, utility negotiations, water, and permitting fights all start to matter to real communities.
Power
Core metrics
- Power-ready queue capacity: LBNL's latest queue data shows 195.1 GW of active capacity with signed interconnection agreements across the main tracked regions (ERCOT 90.3 GW, PJM 30.6 GW, CAISO 74.3 GW). That is a large amount of already-advanced queue capacity competing for real buildout.
- Grid interconnection timing backdrop: LBNL's completed-project data says request to operation took about 4.5 years in 2024 completed projects. This is backdrop, not a daily spot metric, but it shows that big power-linked projects still move on multi-year timelines.
- Power cost: the short-run wholesale proxy built from ERCOT North and California NP15 was $16/MWh reported on April 01, 2026. This is a volatile market-price signal, not the long-term contracted power price a data center would actually try to lock in.
- Contracted power and generation mix: the latest large disclosed AI-linked contract in the dataset was 0.8 GW, and that newly contracted supply was 100% low-carbon. This is useful as a read on what big AI buyers are trying to secure, not as a full market total.
What changed
- Near-term wholesale power costs moved down 45.9% versus March 31, 2026, though this is a noisy short-run market signal rather than a full long-term power-access answer.
- Microsoft's Constellation deal still anchors the contracted-power read: 835 MW of clean baseload under a 20-year agreement.
- The queue data still says ERCOT is far ahead of PJM on already-advanced capacity, which helps explain why Texas looks easier to build in than Northern Virginia on this measure.
AI system impact
- Power still looks like the binding physical constraint. Money and demand are not enough on their own; projects need signed agreements, real queue progress, and actual electricity.
Human system impact
- This is where AI expansion runs into the wider human system most directly, because grid access and electricity prices can spill into utility politics, industrial policy, and local conflict over scarce infrastructure.
Semiconductors
Core metrics
- Advanced chip manufacturing demand (TSMC revenue proxy): NT$317.7bn in February 2026
What changed
- SK hynix said HBM demand remained strong in its latest results, showing that AI memory demand is still tight.
AI system impact
- Advanced chip manufacturing demand (TSMC revenue proxy) is NT$317.7bn in February 2026. Chip supply still sets the pace for how quickly the rest of the AI system can expand.
Human system impact
- A system this dependent on a handful of chipmakers leaves profit, leverage, and strategic risk concentrated in a very small number of firms and places.
Suppliers / Inputs
Core metrics
- Chip-equipment demand (ASML bookings proxy): EUR 13.2bn in Q1 2026
What changed
- ASML reported EUR 13.2bn of net bookings in Q1 2026.
AI system impact
- Chip-equipment demand (ASML bookings proxy) is EUR 13.2bn in Q1 2026. Upstream equipment demand still supports further chip-capacity expansion.
Human system impact
- When tool supply stays concentrated, countries and firms without direct access fall further behind. That turns a technical bottleneck into a strategic one.
30d named / stored event count: 8