00Opening read
The AI supply chain delivered two very different messages this week. At the manufacturing end, the orders look real. Taiwan's TSMC reported another powerful quarter, and Dutch equipment maker ASML showed that chip factories are still paying to increase output. These are realized sales.
Farther downstream, the picture is less comfortable. A major proposed British data center has lost its expected anchor customer after problems securing power. In the United States, lenders are demanding much higher compensation to finance a prominent AI-compute company. Grid regulators want more reliable forecasts of future demand.
This does not mean demand for AI infrastructure has disappeared. It means the difference between ordering chips and operating a profitable computing facility is becoming harder to ignore. Projects must demonstrate electricity, customers and credible economics.
01What moved
Chip manufacturing remains strong. TSMC's third-quarter revenue rose about 51% from a year earlier. ASML, which makes essential chip-production equipment, reported €11.7 billion in quarterly sales. Both benefit from manufacturers expanding output. Additional production still needs qualification.
A British data-center setback makes power risk tangible. Microsoft has withdrawn from its anticipated role at Nscale's planned Essex site, according to the Guardian. Nscale says the project remains alive and it is examining other power options. A site can remain planned even as its economics deteriorate.
AI debt is being priced more skeptically. Reuters reported that some CoreWeave bonds were yielding roughly 15–16% as concerns grew about leverage and the risks attached to AI infrastructure. Investors are scrutinizing completion dates and customer concentration. Financing terms are diverging.
China is drawing a boundary around AI know-how. Beijing has proposed new controls on the transfer of sensitive AI technologies, including certain models and agents. Ordinary access through websites and APIs would be treated differently under the proposal. The draft could reshape cross-border collaboration. The rules are not final.
Robotics needs better observations of the physical world. China's Hesai introduced a combined lidar-and-camera device designed to collect spatial data for robot training. Robots can perform impressive demonstrations yet fail in unfamiliar environments. Reliable data matters on the factory floor.
02The throughline
The physical AI economy is developing a more demanding definition of capacity. A chip that has shipped is different from a server that has been installed. An installed server is different from a data center with a firm power connection. And a working data center is different from an industrial system that earns its keep on a production line.
That explains how strong semiconductor earnings coexist with nervous lenders. Delays accumulate across handoffs. They also explain why power equipment, sensing hardware, manufacturing expertise and industrial data deserve attention alongside GPUs.
For the next several months, watch the projects that move from announcements into service. The strongest evidence will come from operating sites, qualified components, repeat customers and cash flows that survive ordinary financing costs.
03What to watch
- October 20–22: Nvidia's Berlin conference may bring clearer evidence of industrial AI and robotics deployments. Source: https://www.nvidia.com/en-eu/events/ .
- October 28: GE Vernova reports results, offering a read on power-equipment demand and the conversion of orders into deliveries. Source: https://www.gevernova.com/investors/events/3rd-quarter-2026-earnings-webcast .
- November 12: The European Union's AI gigafactory tender closes. Watch whether proposed sites come with credible power and customer plans. Source: https://digital-skills-jobs.europa.eu/en/latest/news/eu-ai-gigafactories-call-open-until-12-november-targeting-over-eu30-billion-investment .
- November 15: Consultation closes on China's proposed technology-export controls. Source: https://www.reuters.com/world/asia-pacific/china-proposes-adding-40-ai-biotech-solar-technologies-export-control-list-2026-10-16/ .
04Source trail
- TSMC, Q3 earnings release (primary): https://pr.tsmc.com/japanese/news/3346
- ASML, Q3 earnings release (primary): 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
- The Guardian, Nscale and Microsoft (original reporting): https://www.theguardian.com/uk-news/2026/oct/18/microsoft-drops-huge-ai-datacentre-project-in-essex-over-power-supply-concerns
- Reuters, CoreWeave debt and private credit: https://www.reuters.com/legal/transactional/private-credit-roundup-coreweaves-debt-stress-tests-private-credits-ai-financing-2026-10-16/
- Reuters, China draft technology controls: https://www.reuters.com/world/asia-pacific/china-proposes-adding-40-ai-biotech-solar-technologies-export-control-list-2026-10-16/
- Reuters, Hesai robotics camera: https://www.reuters.com/technology/chinas-hesai-launches-spatial-camera-boost-ai-powered-robotics-adoption-2026-10-14/
- Reuters, PJM large-load registry: https://www.reuters.com/business/energy/us-panel-approves-pjms-registry-data-centers-rejects-plan-curtail-power-them-2026-10-12/