Monday, August 31st, 2026 Independent & Reader-Supported
The constraint on the AI buildout has been described as electricity for two years. The nearer one is a procurement queue for hardware whose price and availability are now set by trade policy rather than by fabrication.
Buyers have stopped asking whether a supplier's software is secure, because they cannot find out. They have started asking who owns the supplier, where it builds, and what it would have to disclose.
The pitch for putting data centres in orbit opens with free power and ends there. Power was never the part that could not be solved, and the constraint that replaces it is worse.
Almost every byte that crosses an ocean crosses a cable, and the cables break constantly. The vessels that fix them are a few dozen ageing ships held on shared retainers, and the replacement pipeline is close to empty.
Procurement has learned to ask what open-source components a vendor ships. It still rarely asks what is running on the controller beneath the operating system, where the code is older, the provenance is thinner and the update path frequently does not exist.
Subsidy programmes brought cell manufacturing to the continent on a schedule. The electricity those factories need arrives on a different one, set by transmission queues nobody budgeted political attention for.
Organisations are sending each request to whichever model suits it, which lowers cost and quietly turns model selection into a piece of infrastructure somebody has to own.
Incident communication began as courtesy and has become evidence. The wording of a status update now determines credits, contract remedies and whether anyone believes the next one.
Organisations deployed systems that produce work product and left unresolved who is answerable for it. The question arrives with the first consequential error.
Engineering organisations are building the same abstraction over the same infrastructure, independently, because the alternative is asking every team to learn all of it.
Companies collect complaints diligently and route them to resolution rather than to comprehension. The signal that would have prevented the next hundred is discarded at the moment it is answered.
Employees adopted assistants on personal accounts long before any committee finished a review. Banning it moves the usage somewhere with no logging at all.
Authentication requirements that were optional for years are enforced, and organisations that never thought about their sending domain are discovering their mail has quietly stopped arriving.
Deployments stall on the unglamorous discovery that the organisation cannot say what its own records mean. The model is fine. The inputs were never assembled for this.
Industrial equipment outlives its control software by decades. Patching it means stopping production, and the vendor that wrote it may no longer exist.
Automated support handles more contacts than ever and the phones have not gone quiet. The gap between the two numbers is where the cost went.
Companies used to satisfy data residency with a paragraph and a regional bucket. The requirements have grown teeth, and satisfying them is now a question about how the software is built.
Generated training data started as a substitute for records nobody could license. It is now chosen deliberately, for reasons that have less to do with scarcity than with control.
Work moved into the browser years ago and the security stack did not follow. Enterprises are now treating the tab, rather than the device or the network, as the place controls belong.
The work most exposed to automation is concentrated at the bottom of the org chart, which is also where the profession's training happened. Firms are removing the rung without deciding what replaces it.
Software companies spent a decade giving away a usable product to build a funnel. The economics of that giveaway changed when the marginal cost of a free user stopped rounding to zero.
Software that acts on a company's behalf has to log in somewhere. The identity systems it logs into were designed around a person who can be fired, and the gap between those two facts is now a live governance problem.
Robotics in distribution was a capital project only the largest operators could justify. Leasing, smaller machines and software that runs on existing racking have moved it within reach of companies shipping a few thousand orders a day.
Companies have always known their software sits on components maintained by volunteers. Insurers, auditors and regulators are now asking which ones, who maintains them, and what happens when that person stops.
Enterprises spent a year making sure they could switch between model providers. The switch turns out to run through a small number of clouds, chips and inference platforms that no procurement policy addressed.
Logging, metrics and tracing bills have grown faster than the infrastructure they monitor. Engineering teams are discovering that the instrumentation is now a line item that needs its own budget review.
Training got the headlines and the capital budget. The recurring cost of running models in production is what finance departments are now trying to forecast, mostly badly.
Companies that ignored security consultants for years are now doing exactly what their insurance carriers demand, because the alternative is being uninsurable.
A decade of frictionless departmental purchasing produced portfolios nobody inventoried. The consolidation now underway is being run by procurement, not engineering.
Corporate AI budgets are shifting out of innovation labs and into core operating lines, a change that is quietly redrawing how companies account for technology itself.
The most consequential AI deployments of the year are not chatbots. They are agents reconciling invoices, chasing documents and closing tickets nobody wanted to touch.
Federal statistics determine how hundreds of billions are distributed and how the economy is understood. Budget pressure and response-rate decline are straining them at once.
After a spending boom, employers are auditing what their mental health benefits actually deliver, and the results are separating clinical substance from wellness decoration.
Away from enterprise pilots and keynote demos, small firms are putting generative tools to work on the least glamorous parts of operating a business.
Passkeys work, are broadly supported, and are being adopted at a pace that has frustrated everyone who built them. The obstacles are almost entirely organizational.
Companies that once waited for a crisis to bring in operational leadership are now hiring chief operating officers years earlier, and the founders pushing hardest for it are the experienced ones.
Deployment has run well ahead of measurement. A striking number of production systems have no defensible answer to whether they are performing better than what they replaced.
The stores thriving in the post-e-commerce era have stopped competing on inventory and started competing on gathering, and the run club is the new loyalty program.
Organizations meter hours with industrial precision while treating focus, the input that actually produces the work, as free and infinite. The ledger is exactly backwards.
More generation capacity is waiting for permission to connect to the grid than is currently operating on it. The bottleneck is not construction or capital but a study process nobody designed for this volume.
Companies built production systems on models that vendors retire on their own schedule. The resulting migrations are unplanned, unbudgeted and increasingly frequent.
Repatriation was dismissed for years as a vendor talking point. It is now a line item, driven less by ideology than by workloads whose usage stopped being unpredictable.
Adults are spending real money and real hours on demanding hobbies, and an economy of gear, instruction and community has organized around their seriousness.
After a long drought, public listings are returning in cautious single file. The companies going first are teaching everyone else the new rules.
The pandemic-era video visit flattened. What is growing instead is narrower and more durable: virtual clinics built around single conditions and long-term protocols.
Stored credentials, agent-driven purchasing and pay-by-bank rails are dissolving the traditional checkout page, along with the merchant's last moment of control.