State of the Triad — Q3 2026

This is the standing quarterly. It covers the intersection rather than the individual domains: developments that required progress in two or three of artificial intelligence, applied security and settlement to happen at all.

Three things changed materially this quarter. Proving costs fell far enough to make verifiable inference economic on small models. Agent-to-agent settlement moved from demonstration to limited production. And institutional allocation behaviour shifted from trading to holding, which alters what price signals mean.

What we got wrong last quarter

• We expected decentralised compute utilisation to stay flat. It rose on three networks, and we underweighted verification as the unlock.
• We expected retail central bank pilots to continue. Most were shelved.
• We expected seat-based enterprise pricing to hold another year. It is visibly breaking.

What we are watching next quarter

Proof format standardisation, collateral eligibility for tokenised treasuries, and whether extortion without encryption keeps growing as a share of incidents. Each would change a conclusion above.

Publishing what we got wrong is the only part of this document anyone quotes back to us.

Method

Figures are drawn from public disclosures, protocol data and a standing survey of forty institutions. Where a number comes from the survey rather than a public source it is marked as such in the tables. Nothing here is investment advice.