AI Faces the Price of Uncertainty: Oil, Forecasts and the Market Test

This Sunday, the important question is not how much artificial intelligence can do, but how much it costs to get it wrong. A better weather forecast can save energy, anticipate a risk and change an industrial decision. A badly priced technology promise can tie up capital for years. Between those possibilities lies the story of October 25: expensive oil, demanding financing conditions and AI that needs to prove its usefulness beyond the stage. Today's argument is straightforward: when uncertainty becomes more expensive, technology that helps manage it becomes more valuable—not merely technology that generates headlines. With Wall Street closed for the weekend, our starting points are Friday's closing figures and Saturday's reporting; there is no fresh Sunday stock-market session to describe.

Macro/Energy: oil does not wait for the next model

The starting point remains uncomfortable. Associated Press, in its Friday report published by BNN Bloomberg, puts Brent's settlement at $116.72 a barrel, up 3.1%. It had been below $97 at the beginning of October. This is not today's live price or a forecast for Monday: it is the closing reference used here. The same report places the US ten-year Treasury yield at 4.99%, against 4.96% on Thursday. Expensive energy and financing create two different pressures: the first squeezes households and businesses; the second makes investments with distant payoffs harder to justify. For a European company, the transmission from crude to its bill also depends on exchange rates, contracts and hedging. Not everyone experiences the same shock at the same time. The practical implication is not to freeze projects, but to establish which remain worthwhile under less favourable costs and which work only with unusually generous assumptions. A technology budget deserves the same sensitivity analysis as any other substantial operating commitment.

Geopolitics: digital infrastructure has a geography too

The connection between energy and technology is no longer metaphorical. On Friday, The Guardian reported Tehran's warning that it would attack oil facilities, undersea cables, data centres and AI facilities in the Gulf if a new US-Israeli offensive against Iran took place. This is a conditional threat, not confirmation that all those targets have been attacked. AP, meanwhile, describes a war that has severely reduced global oil supply. Together, the reports explain why digital infrastructure decisions now involve questions once associated mainly with oil companies: where capacity is hosted, what electricity supply it depends on, where communications travel and what alternative exists if a region fails. Our interpretation is that resilience can no longer be a decorative clause in a contract. Having two suppliers does not guarantee diversification if both share a critical dependency. It is worth distinguishing a service's commercial location from the physical concentration of its risks, without assuming that any local incident would necessarily cause a worldwide interruption. Geography matters, but the actual dependencies must be checked.

AI/Tech: the value of accuracy—and the danger of extrapolation

That brings us to Saturday's most interesting signal. Xataka examined AI weather forecasting, highlighting advances in hurricane tracks while also discussing limitations in intensity and extreme-weather performance. That distinction matters more than a story about a machine beating everybody. A particular success does not establish universal superiority, and a benchmark percentage is not an operational guarantee. The strongest cross-check comes from a primary source: on February 25, 2025, ECMWF announced the operational launch of AIFS alongside its physics-based IFS. At that launch it reported improvements of up to 20% in certain measures, including tropical cyclone tracks, and approximately a thousandfold reduction in energy consumption for producing a forecast. Those are figures from the announcement of that version and task, not an audit of AI's entire lifecycle in 2026. The centre also explained that both systems share initial conditions grounded in observations. Innovation does not remove the scientific infrastructure that makes it possible. A forecast can become cheaper to produce without rendering observation networks, physical modelling or professional interpretation unnecessary.

The economic translation is more interesting than the model contest. ECMWF identified potential applications in solar radiation, wind at turbine height, insurance and shipping. If a forecast helps schedule maintenance or anticipate renewable output, its value lies in improving a verifiable decision, not in writing a more persuasive report. That does not mean a better forecast automatically makes electricity cheaper: regulation, networks, generating capacity and market design still matter. It does suggest a useful way to evaluate AI projects: choose a recurring decision, measure the existing error and establish the cost of reducing it. It also forces the question of which error is most expensive. In weather forecasting, getting the track right and underestimating intensity are not interchangeable outcomes; in a business, saving time and causing an operational loss are not interchangeable either. A proper comparison includes supervision, maintenance and consequences, not just the price of running the model. That is less spectacular than a demonstration, but much closer to a credible return on investment. The task matters more than the label attached to the software.

Markets: the index falls, but the story is not uniform

Friday's session shows how much the label “technology sell-off” can simplify. According to AP, the S&P 500 fell 0.5%, the Dow 1.2% and the Nasdaq 0.2%; almost three out of four S&P constituents declined. Yet Micron rose 2.5%, Intel and AMD each gained 1.7%, and Microsoft added 0.7%. On the same day, Chevron advanced 2.9% and ConocoPhillips 3.6%. There is no need to invent a single explanation for every move: the evidence is enough to show broad pressure, technology rebounds and an oil-related sector benefit coexisting. AP describes concern about the upfront costs of AI infrastructure and the timing of its returns. A company can sell more and still disappoint if it needs to invest even more to achieve that growth. The useful examination combines revenue growth, investment and cash conversion instead of confusing a product's success with the profitability of buying its shares at any price. A strong technology and an attractive investment are related questions, not identical ones.

The coming calendar sharpens that examination. AP points to results from Alphabet, Microsoft and Amazon during the coming week, alongside a Federal Reserve meeting and the PCE inflation indicator. These are weekly references: we do not claim that all those events fall within the next 72 hours or assign times that have not been checked here. Nor do we convert market expectations into decisions already taken. Before the announcements, the sensible approach is to prepare questions: whether monetisation is improving, whether investment absorbs that improvement and whether companies retain flexibility under expensive energy and financing. For a business buying technology, the questions are similar: which costs are fixed, which rise with usage and which services can be scaled down without compromising operations. Monetary uncertainty does not require cancelling innovation; it requires avoiding commitments whose attraction depends entirely on cheaper money arriving soon. Product adoption can be encouraging while the financing structure behind it still deserves careful scrutiny. Both can be true without contradiction.

The next 24–72 hours: three checks before drawing conclusions

From Sunday the 25th through Wednesday morning the 28th, the radar should begin by separating new facts from repetitions of Friday's news. First, watch whether infrastructure warnings turn into confirmed incidents and whether evidence of actual disruption appears, rather than declarations alone. Second, observe energy markets reopening and Monday's and Tuesday's sessions: the important question is whether crude's movement comes with renewed financing stress and broader equity weakness, not merely an eye-catching headline. Third, check official company and central-bank schedules before assigning times or interpreting rumours as earnings guidance. For someone running a project, this radar can become a concrete task: choose a process that could improve, establish a cost-and-quality baseline and test an alternative within clear limits. You cannot control oil or geopolitics; you can decide what evidence to demand before expanding a contract. That distinction is especially useful over a weekend, when commentary can multiply much faster than genuinely new information. A prepared decision rule is more useful than an improvised reaction to the next alert.

Conclusion: three scenarios, one shared discipline

In the base case, energy remains expensive and AI investment stays under scrutiny: prioritise measurable projects that can expand in stages and retain a reasonable exit. In the bullish case, verifiable easing of energy risks coincides with results showing monetisation and better efficiency: conditions for expanding capacity improve, but margins and cash generation still need checking before extrapolating. In the bearish case, a confirmed interruption to supply or infrastructure combines with persistent financing costs and disappointing technology returns: operational continuity, alternative suppliers and control over commitments become more important. These are analytical scenarios, not predictions or calculated probabilities. The day's lesson fits into one sentence: reducing uncertainty is valuable; pretending it has disappeared is expensive. AI weather forecasting offers a useful picture of the difference: it helps us look further ahead without abandoning the obligation to compare the map with reality. That is a more durable basis for decisions than either unqualified optimism or blanket rejection of a technology whose applications are still developing.

Sources and reading approach

October 25, 2026 edition. Quoted market figures refer to Friday, October 23; implications and scenarios are editorial analysis.