The AI Moat Question Tested

2026-10-18 · The investor's Sunday · AI's moat, tested against credit, A portfolio into the ECB and GDP flash

Daily show10:08Download MP3About the show

How it is made. Aavistus Daily is written and voiced by our own systems, every morning, from data we gather ourselves. The voices are synthetic. Every fact is traced to a source on this page, and every dated call is settled in public on the calibration page. The presenters, Nora and Callum, are named characters, not people.

Aavistus is a new framework, still under development. We will make mistakes, and every dated call is settled here in public so you can see them; the aim is to keep calibrating until the record beats the average forecaster, not to sound right.

Jay Martin argues American AI companies lobby Washington for regulation to defend commodity-like businesses. We test the case against the credit tape and set up the ECB week.

Aavistus briefs. Independent market intelligence — satellites, flows, filings. Register free to follow new briefs.

The AI moat question, tested against the credit tape

Today's investing subject is a business-model question. It is the argument that the big American artificial-intelligence labs are lobbying Washington for strict regulation not because they fear what the technology does, but because they do not have a durable business. The case was made this week by Jay Martin on The Jay Martin Show in an episode published Friday.

Martin's case has four parts. The frontier labs look commodity-like, with capability converging across providers and thin pricing power. Open-weight competition from outside the United States — China's Moonshot AI is his flagship example — is breaking the price. The multi-trillion-dollar valuations and the roughly trillion dollars of hyperscaler compute commitments do not clear in a world where the price of inference keeps falling. And aggressive lobbying for frontier-model licensing and safety pre-clearance reads naturally as an attempt to build the moat — scale and compliance cost — that the technology itself does not provide. Martin frames it as a 1930s-style cartel built around existential-risk language.

Two other voices this week argued the same conclusion with different mechanisms. Lance Roberts, on Thoughtful Money with Adam Taggart, described his firm as having rotated out of mega-cap technology and semiconductors into financials and healthcare, called CoreWeave and the other neo-cloud AI infrastructure plays speculative compared with the foundational hyperscalers, and pointed to the equal-weighted S&P outperforming the cap-weighted index — breadth shifting under the surface. Joseph Wang, on Markets Weekly, argued the AI complex is increasingly fragile — widening cracks across the sector, heavy leverage in the supply chain, and that fragility threatening the broader growth picture.

Where the arguments are right: capability convergence plus free open-weight competition is a real pressure. The gross-versus-net recognition question in AI revenue — how much of a labelled dollar is new economic activity rather than a round-trip between a model builder and its own cloud supplier — tightens Martin's point rather than loosens it. Jack Farley and Quinn Thompson on Forward Guidance this week added the second-order piece: Anthropic's Q3 print showed the rate of change of annual recurring revenue decelerating, and they pointed to the response in neo-cloud credit, CoreWeave in particular. Roberts' breadth point is the mechanism of a slow index-level de-rating under the surface, and Wang's growth point says a hyperscaler capex cut drops into GDP, not just the index.

Where the arguments overshoot: on the index conclusion. The S&P closed at 7659 on Thursday, the Shiller CAPE at 40.5, the VIX at 20. All of that fits a market under tension. But the credit tape is not confirming a break. US high-yield spreads are at 323 basis points, up a touch on the week, still well inside the 400-basis-point mark. The 10-year-to-2-year Treasury spread is at 35 basis points, flatter but positive. The chain this programme has argued from — hyperscaler capex concentration into memory-margin reset into index-level de-rating — has fired at the memory stage. It has not fired at the credit stage. That ordering is what separates a slow de-rating from a sharp break. And Roberts' near-term rotation into financials and healthcare lacks a catalyst inside the next quarter given continued hiking into sticky inflation, the long end near multi-decade highs, and funding-cost compression on the banks.

The read: the business-model question, the breadth reading and the macro channel together argue against adding frontier-AI equity at these valuations with any size — but they do not, on their own, argue for forcing an exit on names whose cash flow is real while the credit tape still prices benign. The position of the house remains short-dated euro and US government bonds, funded in dollars, over US equity.

The read is wrong on the upside if a top-four hyperscaler delivers 2027 capex guidance inside 10 per cent of consensus, Anthropic's next revenue print reaccelerates, and high-yield spreads tighten back under 300 basis points. It is wrong on the downside if the convergence fires inside the next month: a hyperscaler capex cut of more than 10 per cent, spreads through 400 basis points, and equal-weighted underperformance by more than 3 per cent over four weeks. Neither has happened yet.

One line on the regulatory piece: the lobbying is a tell about how operators see their own moat. If frontier-model licensing passes in the form Anthropic's submissions describe, the moat gets built — on cost paid by US customers in a market where free open-weight alternatives are a download away. The medium-term outcome is a two-tier market and lower returns on invested capital for the licensed layer, not higher.

The coming week for a Nordic portfolio

Two scheduled items dominate. On Thursday the 29th, the European Central Bank sets rates. The read here, said on Thursday, Friday and Saturday, is a hold at the 2.5 per cent deposit facility, with the cycle still open. The Harmonised Index of Consumer Prices for the euro area — the HICP — ran at 3.8 per cent in September with core at 2.5 per cent, both still rising, which keeps the hiking posture intact. The read is wrong if the Council surprises with a hike this meeting, or if the statement turns clearly dovish on growth. Neither is the base case.

On Friday the 30th, Eurostat's Q3 GDP flash. A flat print with inflation above target keeps the Council's path tight; a clearly negative print cuts into the hiking posture and the long end of the euro curve richens.

Through the week, US large-cap technology begins Q3 reporting — the capex guidance read is what will fire or not fire against the mechanism described above. On the European side, Nordic Q3 bank reporting continues; Nordea has already printed and did not deliver the stress signal that would turn the defensive read on the sector.

Brent closed at 110 dollars per barrel on Thursday. The Iran-war refining chain fired through to Q3 earnings on the integrated oil complex, and chasing the integrated names at the top of the range with relative strength stretched is not warranted. The standing range on Brent holds. France, outside the calendar, has produced the worst street unrest in years around cost-of-living protests, which keeps the French political-risk tail alive rather than quiet.

This has been a discussion, not investment advice.

This episode discusses investing in general. It is not investment advice, it does not describe anyone's positions, and nothing in it is a recommendation to buy or sell anything.

Sources