The Collection

Mashrabov’s method, as Gustafson quotes it: “We look at revenue over the last 28 days and multiply it by 13.” Annual subscriptions and enterprise contracts are prorated across twelve months. “It is only live revenue. We are not taking three-year enterprise deals and baking them into a $1BN figure.” Gustafson grants the point. Next to contracted ARR, this is the mild end of the ladder. His objection is the window.

A month that carries a big promotion, a product launch, a whale or a consumption spike gets, in his phrase, thirteen little friends. Suddenly a good month is a great year. Growing 10% month on month, the method keeps dropping older, smaller periods and keeping the newest, biggest one. He prefers the old run rate, last quarter times four, because three months smooth the noise. He notes that public companies already argue about this with usage pricing. Some use a month, some 90 days, and MongoDB has used different windows for enterprise and self-serve customers.

A good month is not a great year.

Then the turn. Stop growing, or fall 10%, and the same arithmetic “will very efficiently annualize your shitty month on the way down.” He bets on 90-day averages arriving the day that happens. He would not run a company off the 28-day figure either: that is how you lock in compute on a promotional month or under-hire support, in a business that already juices demand with credits and free usage. His ladder runs subscription ARR, then CARR, consumption ARR, run-rate revenue and 28-day annualized. “We are now in Westworld.”

He applies the lens to the leaked Anthropic FY25 sheet. Of a $42 billion net loss, he reads $34 billion as a noncash revaluation of financing that may convert into shares. The annualized pace he cites ran from $9 billion at the end of 2025 to $30 billion in April and $65 billion by the end of July, and he expects the 2026 revenue that is eventually reported not to match the $65 billion everyone quotes. Monday keeps those as the letter’s reading of leaked figures, not as audited numbers. The Koyfin benchmark tables in the same letter are images and were not read.

Tobi keeps a list of every layer AWS built on top of ECS, and the letter reads it out like a notice of deaths. ECS CLI, archived after ten years. Fargate CLI, archived. Copilot CLI, end of life since June. AWS Proton, end of life in October. App Runner, in maintenance mode since April, with no new features and no new customers. The letter admits its own line from 2023, that App Runner was “the easy way to run your containers,” has not aged well.

None of them touched ECS itself. Each put a friendlier CLI or console flow in front of containers that ran on the same task definitions and services underneath. ECS launched in 2014 and its core API has not changed since. Fargate added a capacity mode in 2017, not a new API. Tobi has services on it running since 2019 that he has not touched once.

Strip the wrapper. The primitive was there the whole time.

The letter is honest about the exceptions. Elastic Beanstalk and Lightsail are wrappers still running a decade on, and Terraform and CDK are wrappers too, which Tobi uses. Its two questions for telling a safe wrapper from a risky one, and its verdict on ECS Express Mode, sit in the full article, which Monday did not open. The rule it prints is plain enough: check what sits underneath, build on that, and let the layer above prove it survives before anything production-critical depends on it. Trailing revenue is the primitive. The 28-day multiple is a convenience layer on the same money.

Jungheinrich, a German logistics company that makes forklifts, put up all the money for Uplift Ventures’ $110 million first fund, with no pension funds or family offices involved. It wants to double its revenue by 2030, and part of that plan now runs through a venture fund. Darío tagged who backed each of the 89 funds he tracked in September. Thirteen had a corporate as LP or sponsor, worth $2.5 billion together. Leave out the four $1 billion-plus funds and one in every five dollars raised that month sat in a fund with a corporate behind it. Those funds were bigger too, a median of $110 million against $78 million. Fourteen more were backed by public money, from the British Business Bank to the EIF.

Matter Venture Partners’ $450 million Fund II has ASML, TSMC and Quanta Computer as LPs, the same companies its startups will one day sell to. His read: these LPs want an early look at the startups that could change their own business, not only returns. His open question is whether they come back for the next fund if the benefits take longer than expected.

An anchor with a home market brings deal flow with the cheque.

The same sheet carries the week’s price of a story. Instinct raised $1 billion at a $10 billion valuation from Sequoia, Benchmark and Coatue for a consumer agent that, in Darío’s words, has no revenue to justify that price. In New York, Accelevation priced at $18, below its $20 to $24 range, and opened at $17.55. He puts that down to rates and supply, with Treasury yields at 19-year highs and a September Fed hike, as printed. A run rate flatters a private round. A public book prices the year.

Before photography, the letter begins, a realistic image meant years of study in paint or charcoal. Then it became mechanically easy, and artists turned out not to be exceptional in that narrow sense. Sam lines it up with the other blows to human exceptionalism: Darwin on where we came from, Einstein on there being no universal now, quantum physics on determinism, and Gödel on a consistent formal system that cannot also be complete.

His claim is that AI does the same to mind. It can already produce explanations, write programs and generate images that once took real human effort. Whether it understands is still open, he says, but the work is being done. Early daguerreotypes imitated oil portraits. Then the uses exploded into the personal and the professional, and the image stopped being something only the rich could afford.

Easier to make changed what it was for.

He does not forecast. History is only a hint, not a firm guide, he writes, and what he sees is anxiety first and new uses after. Monday files it beside The Run Rate for one reason. A 28-day multiple is a photograph of one month: cheap to take and accurate about its moment, but it is not the painting of the year. Once a number gets that easy to make, people start using it for new things, investor updates among them.