Big Tech’s AI Capex Just Overtook Its Cash Flow

by

Key Takeaways

AI capex has now overtaken Big Tech’s free cash flow. Amazon, Microsoft, Alphabet and Meta plan to spend a combined $725 billion in 2026, up 77% from a year earlier. According to Epoch AI, which parsed the companies’ SEC filings (10-Q and 10-K) directly, cash capex is growing about 70% a year while operating cash flow grows about 23%, so the two curves cross around Q3 2026 — the point at which the four hyperscalers’ aggregate free cash flow effectively reaches zero. Oracle has already passed that line.

  • Four hyperscalers’ 2026 combined capex: $725 billion (up 77% year over year)
  • Epoch AI: combined free cash flow hits zero around Q3 2026
  • Growth gap — capex ~70% a year vs operating cash flow ~23%
  • The coming bill: Alphabet depreciation up ~270% in four years; funding shifts to debt and leases

$725 billion. That is what Amazon, Microsoft, Alphabet and Meta have committed to spend on infrastructure in 2026. It is up 77% from a year ago, and almost all of it is AI capex — GPUs, custom silicon and data centers.

The harder question is where the money comes from. Epoch AI parsed the SEC filings of five cloud giants — Microsoft, Amazon, Alphabet, Meta and Oracle — and found their capex curve catching the operating-cash-flow curve around Q3 2026, the point where combined free cash flow for the hyperscalers effectively hits zero.

$725 Billion, and the Quarter Free Cash Flow Hits Zero

Start with the term. Free cash flow (FCF) is the cash a company has left after operating costs and capital spending — the money available for dividends, buybacks, debt paydown or fresh investment. When it approaches zero, that is not a solvency signal; it means nearly every dollar earned is being plowed back into AI infrastructure.

The capex guidance sets the scale. Amazon leads at roughly $200 billion, followed by Microsoft at $190 billion, Alphabet at $180–190 billion and Meta at $125–145 billion. Together that is $725 billion, a 77% jump from $410 billion in 2025.

Epoch AI’s data adds the time axis, and the method is what gives it weight — the research group pulled operating cash flow and capex straight from the XBRL tags (the machine-readable codes in standardized filings) in each company’s quarterly (10-Q) and annual (10-K) reports, spanning Q2 2023 through Q1 2026, using raw disclosure data rather than the polished figures companies headline. The result is the combined free-cash-flow line reaching zero around the third quarter of 2026.

AI-generated illustration of a data center corridor lined with glowing server racks as blue and gold streams of light flow along the floor, symbolizing AI capex
As AI capex pours into the racks, the free cash flow drains away — AI-generated illustration (OpenAI gpt-image-1-mini)

The Original Table: When AI Capex Passes Cash Flow

One combined number only tells half the story. Which company hits the wall first and which has room to spare varies case by case. The table below joins each firm’s 2026 capex guidance with Epoch AI’s estimated free-cash-flow crossover.

Company2026 capex guidanceFCF crossover (Epoch AI)
Oracle(in the five-firm total)Already crossed — FCF turned negative
Amazon~$200 billionAround Q2 2026 (imminent)
Alphabet$180–190 billionAround Q1 2027
Meta$125–145 billionAround Q3 2027
Microsoft~$190 billionAround Q3 2028
Four-firm total$725 billion (+77%)Around Q3 2026 → zero

Our read — the table stitches together two different primary sources. The capex column is each company’s own earnings guidance; the crossover column is Epoch AI’s estimate from parsed SEC filings. Put side by side, they reveal the sequence behind the “combined zero” headline.

  1. Oracle and Amazon lead because their capex outran cash generation the most — Oracle’s free cash flow already turned negative in fiscal 2026.
  2. Microsoft comes last, in 2028, because a wider operating-cash-flow base can absorb the same spending surge for longer.
  3. So “Q3 2026 zero” is not an endpoint but the start of a wave — after the aggregate hits zero, the individual crossings keep rolling out through 2028.

Lens 1 — Cash Flow: The 70%-Versus-23% Math

The cause of the crossover is simple arithmetic — Epoch AI puts the group’s capex growth at roughly 70% a year against operating-cash-flow growth of about 23%. A fast-rising line overtaking a slow-rising one is only a matter of time.

The gap shows up in ratios too. The group now spends about 90% of operating cash flow on capex, up from roughly 65% in 2025. PIMCO estimates that ratio climbs to about 94% in 2026–2027, versus about 40% as recently as 2023 — the most profitable companies in tech now reinvest almost every dollar they earn.

The key here is that this spending is not a choice but a race. The other side of that race — where the cost of making AI explodes while the cost of using it collapses — runs through AI Compute Costs: The Great Divergence. That cost divergence and today’s cash-flow squeeze are two faces of one coin.

Key point: With capex growing 70% and operating cash flow just 23%, the capex-to-cash-flow ratio has jumped from around 40% in 2023 to roughly 90% in 2026.

Lens 2 — Depreciation: The Bill That Hasn’t Arrived

Money spent on GPUs and data centers today does not dent reported profit right away. It is booked as an asset and then charged against earnings over several years as depreciation — the accounting that spreads an asset’s cost across its useful life. The problem is that the bill is only now starting to land.

Alphabet is the clearest case — WSJ estimates its depreciation climbing from $21.1 billion in 2025 (5% of revenue) to about $78 billion in 2029 (11% of revenue), a roughly 270% jump in four years. Morgan Stanley expects the four companies’ combined depreciation to top $520 billion across 2027–2029.

In simpler terms: Capex is the charge you swipe today; depreciation is the installment bill that arrives, split across several years, starting next month. The bigger the swipe, the more of the next few years’ profit is spoken for.

Our read — the lag is the risk. Revenue is recognized today while depreciation lands over years, so if AI revenue does not outgrow the depreciation curve first, margins compress. Morgan Stanley’s forecast that the depreciation load concentrates in 2027–2029 marks exactly the window when “does AI make money” has to be answered.

Lens 3 — Debt: From Cash-Funded to Borrowed

If internal cash cannot keep this pace, the answer is to borrow. And the funding model itself is already shifting. That is the structural change hiding behind the headline number.

BofA estimates 2025 hyperscaler bond issuance at about $121 billion — 4.3 times the $28 billion annual average of 2020–2024. JPMorgan pegs AI-related debt outstanding at roughly $1.2 trillion, about 14% of the US investment-grade index and its single largest sector. Moody’s adds an estimated $662 billion in off-balance-sheet leases on top.

Where all this money ultimately lands — most of a data-center rack’s value is silicon — is something we unpacked in AI Semiconductors Own the Rack. The Bank for International Settlements (BIS) flagged the risk directly in its June 28 annual economic report, warning about the concentration of capex and the reliance on debt.

Key point: With 2025 bond issuance at 4.3 times its historical average and AI-related debt near $1.2 trillion, AI infrastructure has shifted from a cash-funded business to a debt-funded one.

What the Crossover Means — and What It Doesn’t

Start with what it does not mean — free cash flow at zero is not insolvency. All four still hold enormous operating profit and cash, and Alphabet has raised equity while Meta weighs a stock sale to top up the war chest.

The bull and bear cases split here. Jefferies’ Brent Thill says “the AI economy is healthy” and the “bear thesis is garbage,” arguing demand justifies the spend. SLC Management’s Dec Mullarkey counters that Meta’s “once capital-light money machine may be morphing into a capital-intensive incinerator.”

Our read — both sides are really arguing about one variable: the scale of AI revenue. Epoch AI itself cautions that the crossover estimate is sensitive to the fit window, carries seasonality, and is a simple extrapolation that ignores AI return on investment. Framed as conditions, it looks like this.

If AI revenue outruns the depreciation and interest costs that ramp in 2027–2029, today’s zero free cash flow will read as the trough of an investment cycle. If revenue lags that curve instead, a capex-to-cash-flow ratio above 90% and $1.2 trillion in debt become forces that press on margins and valuations at once. Two numbers belong on the watchlist — the actual free cash flow in the late-July Q2 prints from Alphabet, Microsoft and Meta, and whether capex guidance is revised higher again.

Frequently Asked Questions (FAQ)

Q1. Does Big Tech’s free cash flow hitting zero mean bankruptcy risk?

No. Free cash flow near zero means most of the cash earned is being reinvested in AI infrastructure. Amazon, Microsoft, Alphabet and Meta still hold vast operating profit and liquid assets, so this is not a solvency issue. What matters is that the spare cash for dividends and buybacks is shrinking fast, with the shortfall filled by borrowing.

Q2. How reliable is Epoch AI’s Q3 crossover data?

Epoch AI pulled operating cash flow and capex straight from the XBRL tags in each company’s SEC filings (10-Q and 10-K), so it used raw disclosure data rather than company-polished figures. That is its strength. But Epoch itself notes the crossover timing is sensitive to the fit window, carries seasonality, and is a simple extrapolation that does not account for AI return on investment.

Q3. Why does depreciation become a problem later?

Money spent on GPUs and data centers is booked as capex and does not cut profit today, but it flows into the income statement as depreciation over the following years. Alphabet’s depreciation is estimated to rise from 5% of revenue in 2025 to 11% in 2029 (WSJ), and the four companies’ combined figure is projected to exceed $520 billion across 2027–2029 (Morgan Stanley). If AI revenue does not scale enough before then, margins get squeezed.

Q4. Is this investment advice on specific stocks?

No. This is a structural analysis built on public SEC filings and institutional data, and it does not recommend buying, selling or holding any specific stock. Investment decisions should be made independently, based on your own situation and the latest disclosures.

💡 Related Posts
AI Compute Costs: The Great Divergence
💡 Related Posts
AI Semiconductors Own the Rack — the $1.2T Teardown

References

Stay Ahead of the Curve

Get the latest insights on AI, tech, and markets — delivered in both Korean and English.

Explore More Posts