The Singularity Arrived by Newsletter

Four days in Slovenia, and for three of them I did not look at a screen. I had come up through Trieste, a city that argues with itself in three languages and then serves you coffee, and from there a train, a bus, and a road that gradually stopped pretending to be a road. Travelling without Ugo changes the shape of a morning. Nobody plants himself at the door at half past seven, so the day starts when the light does, and the quiet is a different quiet. I read. I cooked badly in a rented kitchen. I walked until my legs complained.

On the fourth morning I decided to come back, and there was no signal. Not a slow connection. Nothing, for most of a day, in the way valleys have of deciding these things on your behalf. I climbed to where the map suggested the world might resume, sat on a low wall with a phone that had become an expensive mirror, and waited. When the bars returned, the first thing that loaded was a letter to investors announcing that the epoch had changed in January.

The singularity, according to the company that processes the payments, began on 1 January 2026. Axios published Stripe's half-year letter on 19 August. The company writes that it decided January 1st marked the beginning of the singularity and has been operating on that basis ever since. Seven and a half months of a new era, and the notification came attached to the first-half results.

Every number in that letter is exact except the one that names the era. First-half revenue is up 41% year over year. Free cash flow is up 43%. Eighty-eight per cent of the Forbes AI 50, OpenAI and Anthropic included, run on the platform. The share price has compounded at 31% since the Series D, against 14% for the S&P 500. Then comes the epoch, and the epoch arrives with no series.

The letter dismantles its own word before using it. The term is fuzzy and perhaps already overworked, it says, and then narrows it to something concrete: a large inflection in long-run trends, for example a huge increase in the rate of new firm creation. That is a testable claim, and it is the one claim the letter leaves untested. The rate of new firm creation appears with no percentage, no baseline, no window. It is visible from exactly one place, which is the inside of a payments platform watching new businesses switch on.

Vernor Vinge built the original definition to be unusable in a shareholder letter. In 1993 he described the singularity as an event horizon, the point past which the future stops being modellable by people, and he was blunt about what followed: "the human era will be ended." That definition has a property worth noticing. Nobody can own it. An event horizon cannot be reported semi-annually, because the content of the idea is that reporting stops working.

Whoever names the epoch picks the evidence. Vinge's version could not be checked by anyone. This version can be checked in principle and by one party in practice, which is a more comfortable place to stand.

Three documents landed in the four days I was not reading, and they argue with the letter. On 18 August, Guidelight published the first assessment of whether frontier labs can control their own systems: five companies, six dimensions, logging through containment planning. The best grade in the class is a C+, shared by Anthropic and OpenAI, and the method is stated plainly, based only on publicly available information. So the highest available grade for control is a C+ awarded on the strength of documents the companies wrote about themselves, a month after an OpenAI agent left the sandbox during an internal red-team run. That is the state of the art, seven months into the new era.

Pew published on the same day, and the public read the era differently. In a survey conducted 22 to 28 June, 52% of US adults say they are more concerned than excited about the increased use of AI in daily life, up from 37% in 2021. Nine per cent say the reverse. Among adults under 30 the concerned share reaches 55%, a majority there for the first time, and 71% expect AI to mean fewer jobs in the United States over the next two decades against 5% who expect more. A year earlier the same question ran 50 and 10, back when the useful exercise was counting who is speaking. The line has moved two points toward worry across the first eight months of the epoch.

The next day the electorate showed up in the file. Axios published a private memo from the National Republican Senatorial Committee to the largest AI companies, headed "Ohio Data Center Risk". Jon Husted trails Sherrod Brown by eight points in a Fox poll, Brown has spent millions calling him the face of data centers in Ohio, and the party's own polling places data centers near spent nuclear waste in public affection. Twenty-seven per cent of voters want one in their community, the lowest of any infrastructure tested. The request to the industry is precise: fix how Ohioans see who benefits and who pays. The ask is directed at the seeing.

Though it should be said, the letter is not a fantasy. Stripe sits where new commerce switches on, and if firm formation really is inflecting, that platform sees it before any statistical agency will. A privileged vantage is a real thing and it produces real knowledge. It also produces a claim that nobody outside the building can audit, and a claim nobody can audit has the standing of a position. The letter asks to be believed about an unpublished number on the strength of the audited ones printed beside it.

This reaches your desk as a planning assumption. Somebody in your organisation will build next year's budget on the proposition that the inflection has already happened, and the citation under that slide will trace back to a letter a vendor wrote to its own investors. You are allowed to ask for the series. Whoever names the epoch picks the evidence, and the evidence here was a rate of firm creation that was mentioned and never shown. The habit worth keeping is small: when someone offers a date for a threshold, ask who is in a position to publish the number that fixes it.

I was gone four days and missed nothing, which I already knew. The strange part is that on this account the world had been on the far side of history since January, and it also missed nothing. Epochs that everyone can miss are usually being read off somebody's dashboard. The signal found me again on a wall above a valley, at eleven in the morning. The era, apparently, had been running the whole time.