Research — Golden Section

Four parts · February – September 2026

The AI Arc

One argument, built and then tested in public: what the 2026 software selloff was actually pricing, the thesis that answered it, the earnings season that put the thesis on the record, and the infrastructure buildout paying for all of it.

Part FourSeptember 2026

Shade at the Parade

Three quarters of a trillion dollars of AI capital spending this year, against $150–200B of genuine AI revenue. Four channels carry that gap into the wider economy: financing, power, return on capital and market concentration. Depreciation has to quadruple on arithmetic alone, the leverage has moved off balance sheet, and the whole cycle turns on a race between two growth rates rather than one. An essay by Dougal Cameron, published alongside the research behind it and labelled so a reader can tell which is which.

Read it AI infrastructureCreditMacro

Part OneFebruary 2026

The SaaSpocalypse

The most severe SaaS selloff since COVID: IGV down 30% from its September 2025 peak, $830B of market capitalisation erased in six sessions. Our finding is that AI fear accounts for 60–70% of the severity, but that it gave the market permission to re-rate what the numbers had been saying for three years. The damage is overwhelmingly horizontal; vertical SaaS with a domain moat is structurally insulated.

Read it Read as PDF Market analysisVertical SaaS

Part TwoMarch 2026

The Enhancement Doctrine

The full account of how AI went from existential threat to strategic partner: the inciting event, the bear case, the ZIRP distortion the market ignored, and the doctrine that ended the panic — that AI enhances software with a domain moat and hollows out software without one. It is also one of the firm's named frameworks.

Read it Read as PDF ThesisFramework

Part ThreeSeptember 2026

Where the AI Budget Lands

The earnings test of the doctrine. Forty-three public software companies reported into an identical macro and got outcomes sixty points apart — Atlassian up 35%, IBM down 25% — decided not by growth but by whether AI arrives in your product or in your customer's capex line. Written for both a limited partner and a founder.

Read it Q2 2026 earningsEvidence

Re-testing our own papers · September 2026

The Addenda

Three papers written in 2020 and 2021, re-run claim by claim against the data available six years later. We publish the corrections rather than quietly reprinting: an investor who acted on the original deserves to know which parts survived.

2026 Addendum to the August 2020 paperSeptember 2026

Investing in Software? You Bet Your Assets

All material claims in the 2020 paper compared against 2026 data, with the result shown. Five of six conclusions hold: a 10x revenue multiple and a 45–50% cash-flow margin are both still defensible, on conditions the original left unstated. Two inputs moved far enough to change what a rational investor should pay — the discount rate and the exit multiple.

Read it Original 2020 paper (PDF) ValuationAsset class

Addendum to Nothing Ventured (2020)September 2026

Something Ventured

Eleven propositions from the 2020 defence of venture capital, scored against September 2026 data. The centrepiece exhibit inverted, the reassuring bottom-quartile floor gave way, and every measure the original relied on was a mark rather than cash. The case for venture in general has weakened; the case for a particular kind of venture is sharper.

Read it Original 2020 paper (PDF) Venture returnsDPI

Addendum to the 2021 paperSeptember 2026

To Fee or Not to Fee — Five Years On

The 2021 instinct holds — the fee rate in isolation is still not where the money is — but the paper measured that fee against a ten-year fund that returned capital, and neither condition is true any more. Venture is the only major strategy whose fee did not move, and “fees are returned first” turns out to be a priority of payment rather than a payment. Golden Section charges management fees; every conclusion adverse to managers is adverse to us.

Read it Original 2021 paper (PDF) Fund economicsLP

The work, published in full.