The week Sam Altman stopped pretending
OpenAI's CEO told an interviewer the singularity has arrived, even as the data centres behind the boom push electricity bills higher for households who never opted in. Both stories are true at once.
On 31 July 2026, three short pieces of reporting landed within the same news cycle. The first quoted Sam Altman saying the moment he and his friends used to joke about at the lunch table, "in a very not-serious way," has actually arrived. The second carried a related claim from the same OpenAI chief: that as AI drives productivity gains, people are increasingly expecting more of themselves. The third reported that the proliferation of AI data centres has led to a surge in electricity demand, contributing to higher utility bills for consumers.
Both stories are true at once, and both being true is the point. Monexus analysis: the gap between the future being marketed and the present being billed is not a bug. It is the business model. Read those threads together and you get the structure of 2026's AI economy, which is to say extraordinary claims about capability paid for in kilowatt-hours that someone else's meter records.
The messianic register is now the default
Altman has spent years edging toward prophecy, and the early-morning quotes on 31 July are the cleanest example of the register he has settled into. The singularity line, read in isolation, is the kind of remark a host prints as a pull quote. Read in sequence with the productivity line, it becomes a sales pitch: the moment is here, therefore the humans in it must run faster, work longer, and accept that the baseline has moved.
The rhetoric matters because audiences have been trained, over two decades of consumer-tech launches, to parse the difference between a product announcement and a sermon. Altman's public posture now lives almost entirely on the sermon side of that line. The singularity arrives not as a benchmark, with a test set and a published card, but as a feeling: a shared sense, cultivated in interviews and on stages, that the rules are about to change for everyone whether they consented or not. People who do not use these systems are still told to feel their gravity.
The bills, on the other hand, are very specific
The reporting on data-centre electricity demand is not a vibe. The mechanism is mechanical: a hyperscale training or inference campus draws a baseload that a regional grid was never sized for, utilities pass through transmission upgrades and capacity charges, and the cost lands on residential ratepayers after regulators have been persuaded to socialise it. The Unusual Whales report describes the result in plain terms. Higher utility bills for consumers are the empirical signature of a capital-intensive industry externalising one of its inputs.
This is the part the marketing does not touch. The grid is the part. The substation is the part. The transformer order book, which has become a real industrial-policy story in the United States, is the part. None of this is metaphorical, and none of it is solved by a clever prompt. The more compelling the case for AI capability becomes, the harder each of those physical constraints bites, because the same productivity story that justifies the build-out also justifies the bills.
Productivity for whom, exactly
Altman's argument that humans will demand more of themselves is a fair description of what is already happening in white-collar firms. The convenient framing is that AI amplifies output. The less convenient framing is that AI also amplifies the pool of people competing for the same salaried role, and that the marginal worker absorbs the pressure first.
Read alongside the data-centre cost story, a more honest version of the productivity argument emerges. The producers of the models capture extraordinary rents on the model layer. The operators of the infrastructure capture rents on power and cooling. The enterprise end user captures a margin on labour cost reduction. The household ratepayer, who never signed an enterprise contract, captures a higher monthly line item. The distribution is not hidden. It is just rarely drawn on the same chart.
What the moment actually demands
The right question is not whether AI is powerful. The right question is which version of the productivity story gets to be the default. The one Altman told in the early hours of 31 July, in which humans rise to meet the moment, or the one the utility bill tells in the kitchen, in which a constrained resource gets allocated by who can afford the interconnect.
Monexus assessment: a serious AI policy agenda in 2026 would have to deal with three things at once. First, transparency on which data centres are driving which marginal megawatts, and where the cost is being allocated. Second, a siting regime that prevents the cheapest grid customers from subsidising the most expensive private compute. Third, a labour conversation that does not treat headcount reduction as an automatic social good. None of that requires hostility to the technology. It requires honesty about its plumbing.
The singularity, as advertised, is a feel. The data-centre load factor is a number. A credible politics of AI has to begin with the number, and stop borrowing the feel.
Desk note: Monexus has framed this as a politics-of-infrastructure piece rather than a capabilities piece, because the source material on the cost side is firmer than the source material on the rhetoric side. The two Altman quotes are reported remarks from the same news cycle; the consumer-cost reporting describes a documented trend.
Wire provenance
This editorial synthesis draws on the following public wire/social posts:
- https://unusualwhales.com/news/altman-ai-singularity-has-arrived
- https://unusualwhales.com/news/sam-altman-ai-four-hour-workweek
- https://unusualwhales.com/news/ai-increasing-consumer-costs
- https://x.com/unusual_whales/status/2083009249534214311
- https://x.com/unusual_whales/status/2083002454875271470