Global technology companies may spend close to $700 billion on artificial intelligence this year, and the strain that outlay places on their cash reserves is already drawing scrutiny from analysts and investors.

The figure, reported by CNBC, describes capital expenditure (capex, the money firms commit to physical and long-term assets such as data centres, chips and networking gear) across the tech sector in 2025. It marks another step up in a multi-year spending escalation. It also marks the point at which the bill has grown large enough that the sector's cash generation no longer covers it comfortably.

What does the number actually cover?

The $700 billion estimate bundles the major categories of AI infrastructure investment: server farms, semiconductors, power capacity and the associated software stack. The hyperscalers — the largest cloud providers such as Microsoft, Amazon, Alphabet and Meta — account for the bulk of it, with smaller players adding data-centre leases and GPU (graphics processing units, the chips that train and run AI models) purchases of their own.

What the aggregate figure does not reveal is the split. It does not say how much spending goes to training frontier models versus serving existing products, how much is debt-financed versus funded from operating cash flow, or how quickly the assets being bought will depreciate. Each of those questions changes the risk calculus materially.

Why does cash flow matter now?

Free cash flow (the cash a company generates after capital spending) is the metric under pressure. When capex rises faster than revenue, free cash flow shrinks even if profits look healthy on paper. That is the mechanism behind the "red flags" CNBC describes.

For companies with hundred-billion-dollar balance sheets, a period of compressed cash flow is survivable. The open question is duration. If AI-related revenue arrives more slowly than the spending schedule assumes, firms face a choice between cutting investment — and conceding competitive ground — or accepting years of weakened cash positions.

Investors have so far tolerated the spending, on the argument that compute capacity determines who wins the next platform shift. That tolerance has a limit, and management teams know it. The sector's credibility now rests on showing revenue that scales with the infrastructure beneath it.

What does this change for decision-makers?

Three practical consequences follow:

• Financing costs rise in importance. Firms funding capex through debt or off-balance-sheet vehicles (structures such as joint ventures that keep spending off the parent's books) become more sensitive to interest-rate moves.
• Supplier revenue concentrates. Chipmakers, data-centre builders and power providers benefit from committed multi-year budgets — until a single large customer pauses.
• Reporting scrutiny sharpens. Expect analysts to press executives on depreciation schedules and return-on-investment timelines for AI assets, not just on aggregate spend.

What is not being said

The report's framing — record spending, warning signs — captures the tension but leaves the decisive question unanswered: no one has yet published a reliable figure for how much revenue the $700 billion in investment will actually generate. Every bullish case rests on that number staying opaque a little longer.

For chief executives allocating budgets, the practical read is straightforward. The spending wave is real and continuing. The margin for error in financing it is what has changed.