🔍 Read the full analysis: Who Pays For The Power Behind AI Data Centers? on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Amazon, Microsoft, Alphabet and Meta spent about $202 billion on property and equipment in the third quarter of 2026, more than twice their combined spending a year earlier, according to figures compiled from company filings. The companies are using cash, debt, leases, partnerships and equity to fund the build-out, shifting some exposure to lenders, investors and power customers; the scale of future obligations and returns from AI remain uncertain.
Amazon, Microsoft, Alphabet and Meta spent about $202 billion on property and equipment in the third quarter of 2026, more than twice the roughly $97 billion they spent a year earlier, according to a calculation based on their cash-flow statements. As the AI data-center build-out expands, the companies are drawing not only on operating cash but also on debt, leases, partnerships and equity, leaving lenders, investors and other parties exposed to some of the cost.
Financing moves announced or disclosed this week illustrate the shift. Meta said it sold $25 billion of common stock and $12.5 billion of preferred shares. Microsoft issued $25 billion in bonds, including debt with a maturity reaching 2066. Amazon disclosed a loan facility of up to $80 billion, secured by servers it will purchase with the proceeds, according to the source report. These are company-specific transactions, not evidence that all spending is financed in the same way.
The quarterly spending figures in the report were $77.3 billion for Amazon, $51.4 billion for Alphabet, $39.4 billion for Microsoft and $33.9 billion for Meta. Microsoft’s figure counts cash capital spending; including new finance leases, the company put quarterly capital spending at $45.2 billion. The report says Amazon expects about $220 billion in cash capital spending for 2026, while Alphabet forecast $195 billion to $205 billion, Microsoft about $170 billion for the calendar year, and Meta $135 billion to $145 billion, including finance-lease payments.
These totals do not mean each company had the same cash shortfall. In the quarter, Amazon generated $42.4 billion in cash from operations, Alphabet $42.3 billion, Meta $32.8 billion and Microsoft $58 billion. The report’s calculation finds that only Microsoft’s operating cash exceeded its reported cash property-and-equipment spending during the quarter. It also estimates the four companies had about $1.02 trillion in future lease payments for leases not yet started as of September 30, based on their filings. That estimate is a commitment over time, not a bill due immediately.
Infographic 01 · One quarter, four companies
The bill doubled. The cash did not.
What Amazon, Microsoft, Alphabet and Meta paid for buildings, servers and chips from July to September 2026, set against the cash their businesses brought in.
AmazonQ3 2025: $35.1B spent
Spent $34.9B more than came in
AlphabetQ3 2025: $24.0B spent
Free cash flow −$9.1B
MicrosoftQ3 2025: $19.4B spent
$18.6B left over
MetaQ3 2025: $18.8B spent
Free cash flow −$2.1B
Cash purchases of property and equipment as reported; not all of it is for AI (Amazon also builds warehouses). Gaps for Amazon and Microsoft are my arithmetic; free cash flow for Alphabet and Meta as the companies report it. Microsoft’s quarter is its fiscal Q1 2027 (July–September 2026).
Sources: Amazon results, 29 Oct · Microsoft 10-Q, 28 Oct · Alphabet results, 28 Oct · Meta results, 21 Oct. Debt raised: cash flow statements, nine months 2026.
Powered by Thorsten Meyer AI
How AI Spending Spreads Financial Risk
The financing mix matters because it determines who carries the cost if AI projects earn less, or take longer to earn, than expected. A company that pays from cash bears the immediate expense itself. Bonds place repayment obligations on the company and interest-rate exposure on lenders and investors. Leases and partnerships can put buildings or other assets on a separate owner’s balance sheet, while the technology company commits to payments over time. Selling shares raises funds without borrowing but dilutes existing shareholders’ ownership.
There is also a timing mismatch. The source report says servers are typically written off over five to six years, while data-center leases can extend to 20 years, power agreements can run 15 to 20 years, and some bonds reach 40 years. If computing equipment becomes outdated or demand disappoints before those longer commitments end, companies and their financing partners may still owe money. That possibility is a risk to evaluate, not proof that losses will occur.
Costs can reach beyond the companies and their direct investors. The report identifies newer GPU cloud providers, data-center building owners and electricity customers among parties that may bear exposure. The precise effects on power bills depend on local rules, contracts, grid investment and how costs are allocated; the supplied material does not quantify those impacts. The central question is not simply how much is being spent, but which parties remain responsible for payments if expected AI revenue fails to arrive.
Products Worth Considering
Bose QuietComfort Wireless Bluetooth Headphones with Active Noise Cancellation, Mic, USB‑C Charging, Deep Bass, 24‑Hour Battery – White Smok
EcoFlow DELTA 3 Classic Portable Power Station, 1024Wh, 1800W
Anker SOLIX C1000 Gen 2 Portable Power Station
As an affiliate, we earn on qualifying purchases.
From Cash-Funded Campuses to Commitments
For years, large technology companies could finance data centers from profits generated by advertising, software, online shopping and cloud services. The source report says the scale of investment in 2026 has changed that pattern. Its comparison of third-quarter filings puts combined property-and-equipment spending by the four companies at about $202 billion, versus $97 billion a year earlier. The comparison covers one quarter and reflects the report author’s sum of company cash-flow statements.
Companies have several ways to fund construction and equipment: operating cash, borrowing, leases, outside ownership of facilities and share issuance. Long-term electricity contracts add another layer of commitment, though the source material does not provide a combined dollar figure for power agreements. Meta described its funding approach on October 26 as a “balanced mix” of operating cash flow, investment-grade debt, strategic partnerships and equity capital. The approach signals that no single source is expected to cover all of its AI investment.
The report also points to different financing conditions among companies. It says Amazon, Alphabet and Meta raised about $225 billion in long-term debt during the first nine months of 2026, while Microsoft’s new bond sale extends to 2066. It cites senior notes issued by GPU cloud company CoreWeave carrying rates of 8.5% to 9.75%. Those figures describe particular borrowing and do not establish the rates or risks faced by every company in the sector.
Returns and Long-Term Bills Remain Unclear
The spending and financing figures do not show how much revenue the new data centers will ultimately generate or when they will become profitable. Demand, utilization, future AI prices and operating costs could all affect whether the investments produce returns sufficient to cover their costs. The source material provides company forecasts for capital spending, but not a comparable forecast of AI revenue attributable to these projects.
The estimated $1.02 trillion in future lease payments is a sum of commitments reported in four filings for leases that had not started by September 30; the source does not provide a breakdown by company, facility or payment year. It should not be treated as an amount due at once. The extent to which lease owners, lenders or technology companies absorb losses if projects underperform also depends on contract terms that are not supplied.
Power costs and their distribution are another open issue. The source identifies electricity customers as potential parties exposed to the expansion, but does not document a specific rate increase, utility decision or local cost allocation. The scale of commitments alone does not establish that household bills will rise or that any company is unable to pay.
Filings Will Show the Next Funding Round
Investors and customers can track upcoming quarterly filings for updated capital-spending forecasts, cash generation, borrowing and lease commitments. The companies’ guidance already points to spending continuing into 2026 and, for Alphabet, a “significant” increase in 2027, according to the source report. Actual spending may differ from forecasts, and subsequent filings will show how much is funded internally or through outside capital.
Further disclosures about data-center partnerships, loan terms and long-term power agreements would help clarify which parties take on the costs and for how long. The next test is whether AI-related demand and revenue grow enough to support the infrastructure commitments—not just whether companies can raise funds to build it.
Key Questions
How much did the four companies spend in the third quarter?
Amazon, Microsoft, Alphabet and Meta spent a combined about $202 billion on property and equipment from July through September 2026, according to the source report’s calculation from company cash-flow statements. It compares with about $97 billion in the same quarter a year earlier.
Who is paying for the AI data-center build-out?
The companies are using several sources: operating cash, bonds and loans, leases, partnerships and equity sales. The funding mix differs by company, and some financing arrangements shift exposure to lenders, shareholders or the owners of data-center assets.
Does the $1.02 trillion lease figure mean the companies owe that amount now?
No. The report estimates about $1.02 trillion in future payments on leases that had not started as of September 30. It is a total of commitments over time, not a single payment due immediately.
Will data-center investment raise household electricity bills?
The source report raises electricity customers as a possible group exposed to costs, but it does not document a specific bill increase. The effect depends on local power markets, utility investment and cost-allocation rules, which are not detailed in the supplied material.
What remains unknown about the spending?
The available figures do not establish how quickly the data centers will generate revenue, whether demand will meet expectations, or how losses would be divided if projects underperform. Further company filings and details of financing and power contracts will clarify the commitments.
Source: ThorstenMeyerAI.com
