The AI trade is being repriced by how you pay for it, not how fast you grow

For two years the AI trade sorted winners by growth. This week the market started sorting them by the balance sheet that pays for the growth. Same theme, new axis, and it changes what an allocator underwrites.

The self-funded pole is printing numbers that look like a different asset class. Palantir did Q2 2026 revenue of $1.94B, up 93% year over year, its 12th straight quarter of accelerating growth, with US commercial revenue up 149% and net retention at 157%. The part that matters for this argument is not the top line, it is the operating leverage: operating margin expanded from 25% to 60% over three years, a Rule of 40 score of 155%, and the whole thing funded from its own cash. Alphabet sits at the same pole from the other direction, paying for its AI build directly out of high-margin Search cash flow rather than the capital markets. These businesses fund the buildout from the P&L.

The debt-and-dilution pole is where the market is now charging rent. Oracle's credit default swap spreads hit record highs, above their 2008 crisis levels, on anxiety about debt-financed cloud capacity. Celestica had to announce a $3.0B equity raise to fund custom AI-rack capacity, and the stock fell on the dilution even though the hyperscaler backlog is intact. One pays with borrowed money and the credit market noticed; the other pays with new shares and existing holders noticed. In both cases the funding source, not the demand, is what moved the price.

Underneath both is an accounting question that has not been settled. Cumulative AI-infrastructure spend is now estimated near $2.0T, and to make the near-term earnings work, major operators extended server and GPU depreciation schedules from roughly 24 months to 60-plus months. That flatters reported operating expense today and defers the write-down if the hardware goes economically obsolete before the books say it should. It is the same instinct as the Oracle CDS and the Celestica raise, just expressed through the income statement instead of the financing line: the bill for the buildout is real, and the debate is only about when and how it shows up.

Why it matters (allocator lens): a repricing by funding source is healthier than a repricing by sentiment, because it is legible. You can read a CDS spread, a buyback authorization, a share count, and a depreciation footnote. When the market pays 80x forward earnings for Palantir and simultaneously pushes Oracle's default insurance to crisis levels, it is not confused about AI, it is discriminating between businesses that generate the cash they spend and businesses that borrow or dilute to spend it. That distinction survives a drawdown; a growth-rate story does not.

So-what for ABC. This is mostly an Alpha-sleeve instruction. Screen AI exposure by cash-flow self-sufficiency: does the capex come from operating cash flow, from debt, or from new equity, and what is the depreciation assumption underneath the reported margin. Names that fund from the P&L (the Palantir and Alphabet profile) earn a durability premium; names leaning on debt markets (the Oracle profile) or serial equity raises (the Celestica profile) carry a cost-of-capital risk that widens exactly when you least want it to. On Beta, this is why the index can keep making highs while single names get repriced underneath, the rotation is inside the trade, not out of it. Neutral overall, because the same $2.0T buildout genuinely rewards the self-funders and punishes the borrowers at the same time, and your job is to own the right side of that line, not to bet on the theme.

Takeaway: the AI question is no longer how fast a company grows, it is who is paying for the growth. Underwrite the funding source, and read the depreciation footnote before you trust the margin.

Analytics & education, not advice. DYOR.