Robots, Chips & WattsRobots, Chips & Watts
RCW-0031 · The Daily · Field record 14 OCT 2026 · 08:00 ET

The AI Boom Has Two Different Supply Problems

ASML cannot meet projected lithography demand, China’s solar producers are cutting workers amid excess capacity, and governments and chipmakers are inventing new ways to finance fabs, power and the factory floor.

Record typeDaily brief
Pre-market
Systemssemiconductors-and-lithography
Advanced packaging
power-grid-and-fuel-cells
Compute infrastructure
industrial-AI-and-robotics
Advanced manufacturing
solar-manufacturing
memory-and-competition
infrastructure-financing
Industrial policy
RegionsUnited States · Netherlands · Japan · China · South Korea · Taiwan · India · Sweden · France · Europe
Source count21 attached
High for company filings, regulator actions and primary announcements; Reuters and AP provide corroborating secondary reporting. FT financing reporting is paywalled. CXMT process claims and future project targets are not independently qualified.

00Opening read

The physical AI economy is running into two very different kinds of industrial pressure.

ASML, the Dutch company that makes essential chipmaking equipment, says its planned production will not be enough to meet customer demand in 2027 and 2028. Some buyers are already asking about machines for the end of the decade. That is a genuine capacity bottleneck, far upstream of the AI server rack.

China's solar industry tells the opposite story. Its largest manufacturers have shed tens of thousands of workers while excess factory capacity keeps pressure on prices and profits. The world needs more electricity, but that does not mean every company making energy equipment earns a return.

Meanwhile, governments and technology companies are finding new ways to pay for what they need. Japan is backing new American chip factories. Google is supporting a major fuel-cell power project. Industrial automation and AI hardware manufacturing are attracting new cross-border partnerships.

The buildout is continuing. Its economics are becoming much less uniform.

01What moved

ASML's queue stretches toward 2030

ASML reported a strong quarter and warned that its planned equipment supply will not meet customer forecasts over the next two years.

That matters because a semiconductor factory cannot make more advanced chips simply by pouring concrete. It needs the machines, engineers and qualification work to turn the building into production.

ASML is looking at expanding capacity and improving the productivity of equipment already installed in customer fabs. Tool output may become as important a signal as new fab announcements.

Japan finances another piece of America's chip supply

Japan plans to invest up to $21 billion in two New York chip factories operated by GlobalFoundries.

The arrangement is unusual. Japanese capital would fund factories jointly owned by U.S. and Japanese interests, while an experienced chipmaker would run them. Customers named in the initial plan include Renesas and Apple.

It is a reminder that semiconductor capacity now sits at the intersection of trade policy, equipment supply and public finance.

The figure is a planned investment ceiling, not completed production.

Google finds another way to secure electricity

American Electric Power plans a $4 billion fuel-cell project near Cheyenne, Wyoming, to supply Google.

The facility is designed for 702 MW of power under a long-term agreement, with full operation targeted for the end of 2028.

This is what capacity underwriting looks like in practice. The customer helps give a power project enough revenue certainty to move toward construction.

The remaining question is whether the equipment, fuel supply and commissioning arrive on schedule.

Physical AI moves onto the factory floor

Omron is buying OnLogic's industrial-computer business for $675 million.

The strategic value is not another general-purpose AI model. It is rugged computing that can sit beside sensors, robots and machine controls in a working factory.

AI deployed at the industrial edge must operate reliably, handle real-world data and fit into existing systems. Omron is buying a piece of that infrastructure.

India starts assembling more of the AI stack

Adani and Jabil have signed a non-binding agreement to explore manufacturing AI racks, servers and related systems in India.

Tata Power is also expanding Mumbai's grid infrastructure to serve growing data-center demand.

These are separate projects, but they address connected problems. Local hardware manufacturing needs customers with powered sites. New data centers need equipment, transmission, cooling and qualified operators.

A manufacturing announcement and an electricity plan only become meaningful when both reach deployment.

02The throughline

The next stage of the physical AI economy will be decided by the quality of capacity, not just its quantity.

A chip tool that is booked years ahead has different economics from a solar panel factory selling into an oversupplied market. A proposed gigawatt is different from one with a funded power path. A new industrial computer matters when it survives daily factory use.

This is why the source trail matters. Contracts, forecasts, memoranda of understanding and completed facilities are different forms of evidence. They should not be counted together.

The strongest signals now connect multiple layers: a buyer willing to underwrite supply, a manufacturer capable of delivery, a grid that can carry the load and a customer prepared to use the finished system.

03What to watch

  • TSMC's October 15 earnings and guidance for advanced-chip capacity and capital spending.
  • Whether PJM revises its rejected data-center curtailment proposal while developing its approved load registry.
  • Whether Omron closes its OnLogic acquisition around January 2027 and announces concrete product integration.
  • Whether ASML can increase 2028 tool output while China's new solar manufacturing standards reduce excess capacity.

04Source trail

ASML, October 14 earnings release: https://www.investor.asml.com/news-releases/news-release-details/asml-reports-eu11.7-billion-total-net-sales-and-eu5.0-billion-net-income-q3-2026

Reuters, October 14, U.S.–Japan GlobalFoundries factories: https://www.reuters.com/business/retail-consumer/japan-invest-up-21-billion-two-us-chip-factories-be-run-by-globalfoundries-2026-10-14/

Reuters, October 14, AEP and Google fuel-cell power: https://www.reuters.com/business/energy/american-electric-power-to-invest-4-billion-fuel-cell-project-power-google-2026-10-14/

Omron, October 12 acquisition announcement: https://www.omron.com/global/en/news/2026/10/c1013.html

Jabil, October 13 proposed India joint venture: https://investors.jabil.com/news/news-details/2026/Adani-Enterprises-Jabil-Sign-MoU-for-5050-Data-Center-Hardware-Joint-Venture/default.aspx

Economic Times, October 14, Tata Power grid expansion: https://economictimes.indiatimes.com/industry/energy/power/tata-power-announces-15000-crore-mumbai-grid-expansion-plan-to-meet-rising-ai-demand/articleshow/134934669.cms

Reuters, October 14, Chinese solar manufacturing employment: https://www.reuters.com/business/world-at-work/china-solar-giants-shed-another-40000-jobs-last-year-overcapacity-persists-2026-10-14/

Reuters, October 12, PJM and FERC: https://www.reuters.com/business/energy/us-panel-approves-pjms-registry-data-centers-rejects-plan-curtail-power-them-2026-10-12/

Reuters, October 8, TSMC revenue and earnings date: https://www.reuters.com/world/asia-pacific/tsmcs-third-quarter-revenue-surges-50-yy-beating-market-forecast-2026-10-08/

05Signal references

This record is cited as evidence by the following active signals.