00Opening read
There is a difference between planning an AI data centre and getting permission to power it. Norway has made that difference unusually clear. Its grid operator has stopped taking new connection reservations for data centres while the government develops rules for deciding which projects deserve access. A country with substantial electricity resources is still finding that the right to connect is not unlimited.
That decision arrives as chipmakers and infrastructure investors confront other practical limits. More advanced memory creates harder cooling problems. More powerful racks require sophisticated electrical components. And long-lead equipment purchases can become costly when a project's design changes before it opens.
The result is a more complicated picture than the familiar race to buy processors. AI infrastructure now depends on several schedules moving together: chip production, memory qualification, electrical equipment, construction, utility approval and financing. Missing one can delay the entire facility. For operators, the real measure of progress is no longer what has been ordered, but what can reliably be switched on.
01What moved
Norway draws a line at the grid connection. Statnett's reported pause on new data-centre reservations is more than a local power story. It raises a question every large computing developer must answer: who gets scarce grid capacity, and who pays for the upgrades? Norway plans to favor some critical services, making the answer partly political rather than purely commercial.
The United States considers a faster permitting route. Senators have advanced a broad energy and infrastructure bill that could shorten approvals for some projects. The legislation still faces negotiations, and faster federal reviews would not erase local opposition or equipment shortages. For now, it is a possible change in the rules, not a completed solution.
The components around the processor gain importance. Infineon has lifted its outlook for sales into data centres, reflecting demand for power electronics. Meanwhile, Samsung and SK hynix are exploring different ways to manage the heat produced by increasingly dense memory. Neither development is as visible as a new AI chip, but both influence how efficiently the finished system operates.
A big infrastructure bet meets accounting reality. SoftBank has reported a large quarterly loss, with the Financial Times attributing part of it to a write-down related to changing requirements at a U.S. data-centre project. Management reportedly says the project is not being downsized. The lesson is not that AI construction has stopped; it is that changing plans after committing to equipment can be expensive.
Industrial robots show measurable demand. North American companies ordered nearly 11,000 robots in the third quarter, according to the Association for Advancing Automation. That is a more tangible sign of factory investment than a promotional video. But an order still has to become an installed machine, connected to production software and maintained through real shifts.
02The throughline
A factory, a port and a data centre share an increasingly important characteristic: their productivity depends on the systems surrounding the headline technology. The robot needs tooling, safety controls and reliable processes. The AI accelerator needs memory, power conversion and cooling. The automated port needs equipment, software and permission to exchange sensitive trade information.
This creates opportunities for companies that rarely dominate the AI conversation. Suppliers of industrial controllers, electrical components, specialist manufacturing equipment and integration services can become essential even without owning a foundation model. It also raises the stakes for governments and utilities, whose decisions increasingly determine where new capacity can be built.
The next phase of physical AI will be measured less by announcements and more by operating evidence: commissioned facilities, qualified components, reliable machines and customers willing to buy again.
03What to watch
- Whether Norway publishes clear criteria for admitting new data-centre projects to its grid queue.
- Whether the U.S. permitting legislation survives negotiation and meaningfully changes project timelines.
- What Nvidia's November 17 earnings reveal about shipments, supply availability and customer deployment.
- Whether growing industrial robot orders translate into sustained factory installations.
04Source trail
- Reuters, Norway grid restrictions: https://www.reuters.com/business/energy/norways-statnett-pauses-new-grid-capacity-data-centres-protect-power-system-2026-11-09/
- Associated Press, U.S. permitting legislation: https://apnews.com/article/c6290598e26a7b01622e5c4eec571516
- Wall Street Journal, Infineon outlook: https://www.wsj.com/business/earnings/chip-maker-infineon-lifts-ai-sales-goal-signaling-confidence-in-demand-80ba33f3
- ETNews, Korean memory packaging: https://en.etnews.com/20261106200006
- Financial Times, SoftBank results: https://www.ft.com/content/5ceed157-0016-4822-a272-8b607911da5e
- Association for Advancing Automation, robot orders: https://www.automate.org/robotics/news/north-american-robot-orders-increase-24-8-in-third-quarter-of-2026
- Reuters, Nvidia earnings watch: https://www.reuters.com/business/finance/wall-st-brokerages-say-ai-selloff-overdone-downplay-democrat-regulation-risks-2026-11-09/