When the Money Runs Out Before the Servers Do

Alphabet spent $44.9 billion of capital expenditure against $39.1 billion of operating cash flow in Q2 2026, producing free cash flow of −$5.9 billion, the first negative free-cash-flow quarter in Alphabet's history as a public company. Alphabet's capital expenditure rose 111% to $98 billion during the first six months of 2026. Shares of Google's parent slid 7% on the day after the announcement, and Amazon, Meta and Microsoft all fell as well. That sell-off was not about the number. It was about the arithmetic underneath it, and what that arithmetic means for the person who has to fund the next six quarters.

If you are the CFO or treasurer at a hyperscaler in August 2026, your job description changed sometime in the second quarter. You used to allocate free cash flow, buyback authorisations, and dividend policy. Now you structure debt offerings, negotiate credit lines, and explain to ratings agencies why leverage ratios that would have triggered covenants three years ago are compatible with an investment-grade rating today. Morgan Stanley and J.P. Morgan project the technology sector will need to issue approximately $1.5 trillion in new debt over the next three years to fund the AI infrastructure build-out. Alphabet raised $49.6 billion by issuing stock in June and brought in $20.3 billion from senior unsecured notes in the second quarter. You are raising capital to spend on data centres the way a utility raises capital to build power plants, except the payback period is a board-level argument, not a regulated tariff.

1. The capex-to-operating-cash-flow ratio inverts

In 2026, capex is reaching levels that look untenable: 86% of sales for Oracle, 54% for Meta, 47% for Microsoft, 46% for Alphabet, and 25% for Amazon. Compare those figures to historical norms. A software company reinvesting 15% of revenue was considered capital-intensive. A cloud provider at 25% was understood to be in an investment phase. The hyperscalers now spend 31-83% of revenue on capex, ratios previously seen only in capital-intensive industrial utilities and telecommunications companies.

The inversion matters because it changes the primary constraint. For two decades, the limiting factor at a large technology company was product-market fit, not the balance sheet. You built what the market would pay for, you collected cash, and you allocated the surplus. The five hyperscalers plan to spend roughly $660-690 billion on infrastructure in 2026. Operating cash flow will not cover it. That makes capital structure the binding constraint, which is a different kind of problem and a different kind of meeting.

2. Debt issuance becomes a quarterly operational necessity, not an episodic event

Q2 2026 was the quarter self-funding ended, and the July 2026 reporting cycle turned this from a projection into a disclosure. Before this quarter, debt issuance at a company like Alphabet was opportunistic. You issued when rates were attractive or when you wanted to repatriate offshore cash without a tax event. Now the debt is structural. JPMorgan raised its estimate for global AI-related capital expenditures through 2030 to $5.5 trillion, and projects AI-related debt financing will reach $4.1 trillion as loan-to-cost ratios rise.

If you run treasury at one of these firms, the work has moved from episodic transactions to a permanent function. You are in the market multiple times per year. You have a forward calendar of maturities, a view on where the curve sits relative to your marginal cost of equity, and a story you tell fixed-income investors about why the capex you are funding today will generate the EBITDA to service the paper five years out. That story used to be implicit. Now it is the pitch.

3. The return-on-invested-capital thesis becomes a board-level input, not a CFO footnote

In recent quarters, investors cheered capital spending hikes, interpreting them as proof of healthy demand and a maturing revenue backlog. That interpretation reversed in July. Alphabet announced plans to boost its 2026 capex forecast, and shares slid 7%, underscoring increased scrutiny of infrastructure investments that are resulting in dwindling cash piles with uncertain returns. The question investors are asking is no longer whether the company can afford to spend. The question is what the return on that spend will be, when it will arrive, and how confident management is in the denominator.

If you are preparing materials for a board meeting in the second half of 2026, you are being asked to quantify the incremental revenue per dollar of capex, the payback period by workload type (training versus inference), and the sensitivity of those assumptions to customer retention rates and utilisation. Alphabet CEO Sundar Pichai acknowledged the scale is significant enough to cause concern internally, but pointed to a cloud backlog that surged 55% sequentially to over $240 billion. The backlog is evidence, but backlog is contracted future revenue, not realised margin. Translating contract value into IRR is the work, and that work now sits upstream of the capital-allocation decision.

4. The ratings agencies start asking different questions

Investment-grade ratings are not automatic. They rest on coverage ratios, leverage multiples, and a qualitative assessment of business risk. The hyperscalers now spend 31-83% of revenue on capex, and the technology sector will need to issue approximately $1.5 trillion in new debt over the next three years. Moody's and S&P are not going to downgrade a company with $240 billion of backlog and operating margins above 30%, but the conversation in the analyst call has shifted. The question is no longer whether you have access to capital. The question is whether the capital you deploy today de-risks the balance sheet in 2028 or leaves you with stranded assets if demand softens.

If you are the IR lead or the treasurer on those calls, you are now defending a capital structure that looks more like a telco or a pipeline company than a SaaS business. The ratings agencies want to see contracted revenue that supports the debt service, a credible plan to moderate capex growth after 2027, and some evidence that utilisation is tracking ahead of capacity additions. They will not say it directly, but the subtext is clear: show us the revenue ramp or show us the capex taper.

5. Compensation structures tilt toward equity because the cash is spoken for

Addison Group's 2026 Workforce Planning Guide predicts tech salaries to jump 8-10% this year, and the technology sector shows even more dramatic increases, with software engineers commanding 12-15% salary premiums compared to 2025. But aggregate salary budgets are rising while free cash flow is negative or marginal. The arithmetic does not close without a shift in how people are paid.

Equity as a percentage of total compensation has increased for roles outside the executive layer. Salary increases for in-demand positions such as mid-level AI engineers have increased sharply, and specialised roles like LLM developers reached averages of $209,000 in base-level compensation, but the increment is showing up as RSUs, not cash. If you run compensation planning, the policy decision you are making is not whether to pay competitively. The decision is how much of that competitiveness you fund with shares rather than payroll, because payroll is a use of operating cash flow and operating cash flow is already allocated.

The second-order effect is dilution. Companies that historically ran buyback programmes to offset option exercises are now issuing equity to fund compensation while simultaneously raising debt to fund capex. Alphabet produced free cash flow of −$5.9 billion in Q2 2026 and repurchased no stock. Share count is rising, which affects EPS, which affects valuation multiples. That dynamic used to be confined to high-growth, pre-profitable companies. Now it is a feature of the largest technology businesses in the world.

6. The treasury function starts to look like infrastructure finance

Amazon's 2026 capex guidance of $200 billion alone exceeds the combined annual capex of the entire publicly traded US energy sector. Hyperscaler capex in 2026 is expected to be at approximately $646 billion, or about 2% of US GDP. These are not software-company numbers. These are the capital-formation figures you see in sectors where the asset base has a 20-year life, a regulated return, and debt that gets refinanced rather than retired.

If your role is to manage the capital structure for that kind of spending, the skillset required is no longer the skillset of a technology CFO. You need people who understand project finance, who can model cash flows at the facility level, who know how to term out construction debt, and who can negotiate covenants with lenders who expect a different risk profile than a corporate revolver. Some of the hyperscalers are hiring from infrastructure funds, from utilities, from project developers. That is not a coincidence. The work has converged.

7. The person holding the budget is now holding the strategy

Capital allocation used to be downstream of product decisions. The executive team decided what to build, and finance structured the capital to fund it. All the hyperscalers report that their markets are supply-constrained, rather than demand-constrained. That statement reverses the usual sequence. Demand is not the question; capacity is. Which means the person deciding how much capacity to build, and when, and in what configuration, is making the strategic call.

If you are the CFO, you are now in the room when the infrastructure roadmap is set, not after. You are the one saying whether the company can carry another $50 billion of capex in 2027, what that does to the debt stack, and what return threshold makes the spend defensible to the board. Product, engineering, and finance used to operate in sequence. Now they operate in parallel, because the constraint is not what you can build. The constraint is what you can afford to build before the rating gets cut or the equity re-rates.

Spending could surpass $700 billion in 2026, up sharply from about $410 billion last year. Analysts forecast capital expenditures reaching $650 billion in 2026 and surpassing $1.1 trillion in 2027. If those numbers hold, the person managing treasury at a hyperscaler in 2027 will be managing more capital than most sovereign wealth funds deploy in a year. The job is no longer about optimising tax efficiency or timing buybacks. The job is about financing a multi-year build-out at a scale the technology sector has never attempted, with a cost of capital that assumes the revenue will arrive on schedule. The risk is not that the infrastructure will fail. The risk is that it works exactly as designed, but the workload it was built for migrates, or the margin compresses, or the replacement cycle comes faster than the depreciation schedule assumed. Then you are left with an asset that cannot be re-purposed and a debt stack that has to be serviced anyway.

That is a different kind of risk than a product launch that underperforms. You can kill a product. You cannot kill a data centre.

Tarry Singh is the founder and CEO of Real AI (realai.eu), an enterprise AI advisory and deployment firm working with global enterprises on production agent systems, model risk, and AI sovereignty strategy. He also leads Earthscan (earthscan.io) for Energy AI, and is a founding contributor to the EU-funded HCAIM and PANORAIMA programmes for responsible AI education across European universities. He writes at tarrysingh.com.