Headlines said stop. Investors said records.

Signal 011 named convertibility — suppliers sold off while orders improved, guarantees stacking, a leveraged fund forced to sell paper wealth into cash. This week’s headlines stayed in that register: models acting on their own in security tests, OpenAI pausing Astra, Texas pausing new data-centre approvals while it checks whether hundreds of gigawatts of power requests on its waiting list are real projects or paper. Easy to read as a build-out hitting walls. The share prices mostly disagreed.

Two audiences watch the same AI spend. Shareholders buy a growth story. They can sell in seconds. They care whether the next quarter looks better and whether someone else will pay more tomorrow. Lenders buy an IOU — a bond. They care whether the company can pay it back with cash it has not yet earned. All week the shareholders cheered: indices at records on weak jobs data, no repeat of the supplier selloff that marked Signal 011, big moves in the names that carry the trade. The lenders were slower to speak. That split is the story.

What shareholders are buying is not one thing. Application revenue that actually clears — Palantir’s week is the clearest instance — and cloud economics still converting at the layer that pays the data-centre bill. Microsoft, Google and Amazon added roughly $56 billion of annualised cloud revenue in the second quarter at an incremental margin above fifty per cent. That is the bet equity is making: not whether AI works, but whether the spend is producing revenue the market will fund at these prices.

SpaceX is the floor the complex needed to find. First post-IPO earnings showed $18.4 billion of quarterly capital expenditure against $7.8 billion of revenue — the kind of headline that usually punishes a stock. Buyers showed up again near a hundred dollars, a price that has held more than once since listing. Lock-up expiry forced some holders to sell. Other buyers absorbed it. That matters beyond SpaceX. If the company building rockets and data centres can keep executing at the pace the market demands, the next twenty-five-billion-dollar bond from Amazon or Nvidia is easier to believe — not because the maths got safer, but because the crowd has shown it will fund the story.

Then came Saturday — the lenders’ turn. Several companies had just sold large new bonds to fund chips and data centres. By the weekend those bonds were trading below the price they were issued at. The people who bought the IOUs on day one could not sell without a loss. The market for the next IOU took notice. Bloomberg reported selective allocation on a Meta data-centre bond — a bouncer at the door, letting only slow, serious money in. Not a funding crisis: a warning that the easy round is over. Credit dissented after equity had already voted.

Read the combination as late in the cycle, not a clean new leg. Caution still warranted. Whether fresh all-time highs follow is not settled. What is settled: share-market interest has not waned and buyers keep appearing at known levels — while lenders are starting to ask harder questions about the IOUs underneath. The lead signal below carries the weekly closes. Texas and governance carry the headlines shareholders brushed past.