Same Capex Is A Different Decision
The Decision Signal is a curated executive intelligence brief built to sharpen how decision makers see, think, and act. An approved capital plan can change its funding, its duration and its risk without ever returning to the board, and three signals show that happening now.
Capital spending on artificial intelligence infrastructure has grown faster than any corporate investment program in modern memory, and for three years it was paid for out of operating cash. That is changing. On current trends aggregate hyperscaler capital spending crosses operating cash flow this quarter, and the shortfall has to be funded from somewhere other than operations.
A plan approved against surplus cash can grow dependent on external financing without ever returning as a new decision. Whether the approval on file still describes the economics now being executed is the question the board owns.
Capital can bridge the gap for a while, but not indefinitely.
Torsten Slok, Chief Economist, Apollo Global Management, August 2026
The question is whether the approval on file still describes the decision now running.
- Allocation: The plan is crossing out of the funding it was approved on.
- Financing: Creditors now price what retained earnings used to carry.
- Duration: The obligation is contractual before the return is observable.
On current trends, aggregate hyperscaler capital spending crosses operating cash flow in the third quarter of 2026. Capex is growing near 70% a year against operating cash flow near 23%.
A capital plan funded from surplus asks the board one question. A plan funded beyond surplus asks a different one. The crossing does not arrive as a proposal, because no single quarter contains it: each increment is affordable against the cash generated that quarter, and the trajectory is visible only when the two curves are drawn together. The crossover is an extrapolation of fitted trends rather than a reported result. Oracle is past the line and Amazon is approaching it. Alphabet reaches the same point around the first quarter of 2027, Meta around the third, Microsoft not until late 2028. The staggering matters, since a board reviewing its own company sees an ordinary increase while the sector passes a threshold.
↗ Epoch AI — Hyperscaler Capex to Exceed Cash Flow by Q3 2026AI-related debt issuance passed $220 billion this year against $12.5 billion in the same period last year. Demand for that paper is cooling, and recent deals have needed more yield to clear.
The funding line is where the decision quietly changed. Spending once covered by operations is now carried in part by creditors, and creditors price duration and repayment in a way retained earnings never do. The scale of that shift is new rather than gradual: AI borrowing went from $12.5 billion to $220 billion in a year, inside $1.68 trillion of US corporate issuance. The market is absorbing it, and charging for it. Fund managers report fatigue setting in, recent deals have needed more yield to clear, and August supply ran to $153.2 billion against July's $147.7 billion, so the demand is being met at a rising price.
↗ Reuters — US corporate AI debt surge tests investor limits as fatigue emergesGoldman Sachs puts the delay between investment and monetization at three months to two years. Roughly 33% of this year's hyperscaler capital spending is expected to be financed with debt.
The obligation and the evidence arrive on different schedules. A financing commitment is contractual on the day it is signed, while the investment it funds may take as long as two years to monetize. That asymmetry converts a capital decision into a governance problem: the next increment comes up for approval well before the last one has demonstrated anything, so each approval is made against a forecast rather than a result. Goldman expects hyperscaler debt supply near $250 billion this year and as much as 400 billion next, with the debt share peaking around 35% in 2027. The gap between commitment and evidence is scheduled to widen for another year.
↗ Goldman Sachs — How AI Debt Is Reshaping Credit MarketsA spending curve, a bond market and a payback schedule are three views of one capital decision. Each has moved since it was approved.
Capital spending is growing near 70% a year against operating cash flow at 23, so the plan crosses out of surplus funding in the third quarter. Creditors now carry part of what operations used to, and they are repricing it, with recent deals needing more yield to clear. The investment behind the commitment may take as long as two years to monetize, while the debt share of spending is scheduled to rise through 2027.
The same capital expenditure number therefore describes a different decision than it did when it was approved. What changed was the funding, the duration and the risk, none of which appear in the number.
The Decision Signal’s posture for the week of August 31, 2026 is Plan.
All five postures are live. Click on any zone to read this week’s signals from there, and what standing in it would cost.
◇ PLAN: The board re-underwrites the capital plan against its current funding before the next increment is approved, with the CFO preparing the revised terms.
A capital plan underwritten against surplus cash can grow dependent on external financing without ever returning as a new decision. The approved number holds while the economics beneath it move.
Every increment approved on the original terms compounds a commitment nobody re-decided. The distance keeps widening for as long as the debt share of capital spending continues to rise.
The plan carries a cash-funded assumption while the funding has moved toward credit, and nothing in the capex number records the change. Closing it means re-underwriting terms that were never formally revised. What was this plan approved as, and what has it become?
An approval is a moment. The thing it approves keeps moving.
A capital plan underwritten against surplus cash, on a horizon the board could observe, is one kind of commitment. The public evidence now shows internal funding headroom narrowing, a larger share carried by debt, and monetization that can lag investment by as long as two years. A change of that kind rarely arrives as a proposal. It accumulates through treasury decisions, financing structures and quarterly increments, each defensible on its own.
The common response is to treat this as a question about whether AI spending is wise. That question is comfortable, because it can be debated indefinitely without revisiting a decision already taken.
The governing question is narrower. When the funding, duration or risk behind an approved strategy changes materially, the original approval no longer governs it. Two questions settle it. What was this plan approved as, and what has it become? A board that cannot answer the second risks ratifying a trajectory rather than overseeing a strategy.
Respectfully,
PJ Bickett
After reading this briefing, what is your immediate posture?