Editor's note: why this publication exists

This is the first issue of Synaptic Swarm. It is the kind of thing every new column publishes — a justification of itself — and I will try to make it short.

There is a category of analysis the financial press does badly and the academic press does slowly: how complex systems behave once they stop being explainable from a single seat. Hyperscaler infrastructure is one. Decentralised AI training is another. Swarm robotics deployed at municipal scale is a third. So is a labour market re-pricing around generative tooling. Each is a system in which the marginal actor is not the relevant analytical unit, and the aggregate is not a simple sum.

I have spent a decade working on bio-inspired AI and adaptive systems — both the engineering of them and, more recently, the financial-market consequences of them being deployed at scale. The work has convinced me that the analytical frameworks markets use to price these systems are, on average, a generation behind. Synaptic Swarm is one analyst's attempt to close some of that gap, in public, every Monday.

What this is, and is not

What this is: a weekly long-form column, published Mondays at 07:00 GMT, on emergent systems and their consequences — for technology architectures, for capital markets, and for the labour markets in between. Each column makes one argument, in 1,500 to 2,500 words, with the numbers and assumptions spelled out. Free readers get every Monday column in full. Paid readers get the back-catalogue, the weekday Updates, and (at the top tier) the spreadsheets behind the models.

What this is not: a link aggregator, a hot-take feed, or a hedge-fund newsletter. It will not tell you what to buy. It will occasionally tell you what to watch, and explain why. The editorial line is that the argument must be falsifiable — if a column says "Nvidia gross margin compresses 600bps over the next four quarters," the next four prints will either bear that out or they will not, and the column after that will say so.

Who it is for

Roughly: anyone whose work depends on the next-order consequences of decentralised intelligence. Operators building inference stacks at scale. Allocators trying to mark-to-market the durable economics of frontier compute. Strategy teams at hyperscalers and semiconductor suppliers. Engineers who would rather read about market structure than another model card. Academics, occasionally, when the academic literature is wrong.

What unites the readership is not industry. It is a tolerance for the idea that the right answer to most questions involving large adaptive systems is "it depends on the topology," and that the topology is usually worth the time to draw.

The reading contract

Three things you can hold me to.

First: I will name companies, name dates, and name numbers. If a claim cannot be made specific, it does not belong in print. The corollary — and this is the harder discipline — is that I will be wrong on the record. Updates and corrections will be published in the same column where the error originated, dated and labelled.

Second: there is no sponsorship, no syndication, and no corporate parent. The publication is funded entirely by reader subscriptions. That is the whole business model, and it is the reason the editorial line is independent.

Third: every column is a finished argument, not a thread of provocations. If a piece does not have a thesis, it does not run.

What the first volumes will cover

The opening volume — Mondays from late July through year-end — will work through the structural changes happening in three places at once: the hyperscaler capex cycle and what it means for the cost curve of large-model inference; the gap between published model performance and what actually deploys profitably at customer scale; and the early evidence on what jobs are being meaningfully restructured by generative tooling versus theatre.

Across all three, the through-line is the same: when intelligence becomes distributed cheaply, the rents stop accruing to whoever has the smartest agent and start accruing to whoever owns the coordination layer. That is an old observation in biology. It turns out to apply to capital markets as well.

I will see you back here next Monday.

— Kairos Thorne, Singapore