TL;DR
Get business pricing on networking and server gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Amazon, Microsoft, Alphabet and Meta spent about $202 billion on property and equipment in July–September 2026, more than twice the year-earlier total, according to figures compiled from company filings. They are funding the build-out through a mix of cash, borrowing, leases, partnerships and share sales, leaving investors, lenders and potentially electricity customers exposed to some of the risk.
Amazon, Microsoft, Alphabet and Meta spent about $202 billion on property and equipment from July through September 2026, more than double the roughly $97 billion they spent in the same period a year earlier, according to a calculation based on their filings. The spending, much of it directed toward data centers, servers and chips, is being financed through a wider mix of cash, borrowing, leases, partnerships and share sales, spreading some of the cost and risk beyond the companies themselves.
The disclosures show the scale of the spending at individual companies. Amazon reported $77.3 billion in property and equipment purchases for the quarter, up from $35.1 billion a year earlier. It said the increase primarily reflected investment in artificial intelligence, while its spending also includes warehouses. Amazon expects cash capital spending of about $220 billion for 2026. Alphabet reported $51.4 billion, compared with $24 billion a year earlier, and kept its 2026 capital spending forecast at $195 billion to $205 billion.
Funding needs are also showing up in financing activity. Meta said on October 26 that it intends to fund AI investment through operating cash flow, investment-grade debt, strategic partnerships and equity capital. This week, Microsoft issued $25 billion in bonds, including debt extending to 2066. Amazon disclosed a loan facility of up to $80 billion secured by servers it will purchase with the proceeds. Meta also sold $25 billion of common stock and $12.5 billion of preferred shares, according to the source material.
Company cash generation does not cover all the reported spending for three of the four firms in the quarter, based on the source’s calculations and company-reported measures. Amazon’s trailing 12-month free cash flow was negative $42.4 billion; Alphabet reported quarterly free cash flow of negative $9.1 billion, and Meta reported negative $2.1 billion. Microsoft remained positive on the comparison presented. The four firms also had about $1.02 trillion in future lease payments for leases not yet commenced as of September 30, mostly related to data centers, according to a sum of their quarterly filings.
Infrastructure finance · Q3 2026
How AI Data Centers Get Paid For—and Who Foots the Bill
Four technology giants spent about $202 billion on property and equipment in one quarter. Cash, borrowing, leases, partnerships and share sales are spreading the funding—and the exposure—across a wider circle.
Amazon, Microsoft, Alphabet and Meta, July–September 2026
For leases not yet commenced as of September 30, mostly data centers
Approximate prior-year quarterly total
More than double the year-earlier amount
Amazon, Alphabet and Meta, first nine months
July through September; company calendars vary
01 / The spending wave
A sharp rise in capital spending
Filings show property and equipment purchases more than doubled year over year. The total includes data centers, servers and chips, alongside other investments such as Amazon warehouses; it is not a precise measure of AI-only spending.
Approximate combined property and equipment spending reported by the four companies. Totals are compiled from filings with differing fiscal calendars.
$77.3 billion
Up from $35.1 billion a year earlier. Amazon said the increase primarily reflected AI investment. Its 2026 cash capital spending outlook is about $220 billion.
$51.4 billion
Up from $24 billion a year earlier. Alphabet kept its 2026 capital spending forecast at $195–205 billion and expects a significant increase in 2027.
02 / The funding stack
More than one source pays for the build-out
Companies are combining operating cash with financing that brings lenders, investors and partners into the picture. The mix varies by company and project.
$25B bond issue
Microsoft issued $25 billion in bonds, including debt extending to 2066.
Up to $80B
Amazon disclosed a loan facility secured by servers it will purchase with the proceeds.
$37.5B combined
Meta sold $25 billion of common stock and $12.5 billion of preferred shares, according to the source material.
03 / Who carries exposure?
Obligations can outlast the hardware
The useful life of computing equipment may be shorter than the contracts used to build and power the facilities around it. If AI revenue grows more slowly than expected, longer-term payments still come due.
Typical server write-off period described in the source material.
Potential duration of electricity supply commitments.
Long-lived property obligations tied to data center capacity.
Companies
Remain responsible for their own debt, lease and operating commitments.
Lenders & investors
Bondholders, lenders and shareholders provide capital and bear different forms of risk.
Partners & customers
Property owners, pension funds and electricity customers may be affected; the source does not quantify how or by how much.
“Through a balanced mix of operating cash flow, investment-grade debt, strategic partnerships, and equity capital.”
Meta · Statement dated October 26, 2026
The increase in spending “primarily reflects investments in artificial intelligence.”
Amazon · October 29, 2026 results
04 / What the figures can—and cannot—show
Big commitments; uncertain returns
The disclosures reveal how infrastructure is financed, but they do not establish how quickly AI services will generate enough revenue to cover the investment.
Pressure at three firms
Source calculations report negative free cash flow for Amazon (−$42.4B trailing 12 months), Alphabet (−$9.1B quarterly) and Meta (−$2.1B quarterly). Microsoft remained positive on the comparison presented.
More exposed financing
The source cites GPU cloud providers such as CoreWeave, with senior note rates of 8.5%–9.75%, as more exposed borrowers.
This is a calculated sum of future payments for leases that had not started by September 30, mostly related to data centers. It is a long-term commitment figure, not an immediate bill. The supplied material does not provide a comparable full breakdown of power contracts or estimate effects on customer electricity prices.
05 / Questions to watch
What comes next?
How much did the four companies spend?
About $202 billion on property and equipment in Q3 2026. The total includes non-AI investment as well as data centers, servers and chips.
Where does the money come from?
Operating cash, bonds and loans, leases, strategic partnerships and equity financing. The mix differs by company and project.
What does the lease figure mean?
$1.02 trillion in future payments for leases not yet commenced as of September 30, 2026—mostly data center leases, not an immediate payment due.
Who could bear risk if revenue falls short?
Companies remain responsible for their obligations; lenders, bondholders, equity investors and partners may also have exposure. The supplied material does not quantify how losses would be distributed.
Watch upcoming filings for spending pace, funding choices and AI service revenue. Amazon’s 2026 cash capital spending outlook is about $220 billion; Alphabet’s is $195–205 billion.
Who Carries the Infrastructure Costs
The spending matters because the AI infrastructure build-out creates obligations that last longer than some of the equipment it supports. Servers are generally written off over about five to six years, while data center leases can extend up to 20 years, power contracts may run 15 to 20 years, and some bonds mature decades after issuance. If revenue from AI services grows more slowly than expected, companies still have to meet those longer-term payments.
The costs do not all sit with the largest technology firms. Lenders and bondholders finance borrowing; investors provide equity; and partnerships or property-owning companies can hold assets and obligations connected to data centers. The source identifies newer GPU cloud providers such as CoreWeave as more exposed borrowers, citing senior note rates of 8.5% to 9.75%. It also identifies pension funds and electricity customers as parties that could bear indirect exposure, though the size and route of any such impact are not established in the supplied material.
Products Worth Considering
Anker SOLIX C1000 Gen 2 Portable Power Station
Anker SOLIX S2000 Portable Power Station, 2,010Wh, 1,500W
LG 34GX90SA-W 34-inch Ultragear WQHD OLED Curved Gaming Monitor
As an affiliate, we earn on qualifying purchases.
From Cash-Funded Builds to Mixed Financing
For years, major technology companies could fund data center construction from profits generated by advertising, software and online commerce. The supplied figures point to a change in scale during 2026: quarterly property and equipment spending by the four companies reached about $202 billion, and the source calculates that Amazon, Alphabet and Meta raised about $225 billion in long-term debt during the first nine months of the year.
The reported $202 billion is a total for property and equipment, not a precise measure of AI spending. It includes other investments, such as Amazon warehouse construction, and the companies’ reporting periods and fiscal calendars are not identical. The source says Amazon attributed most of its spending increase to AI, but the aggregate figure should not be read as spending exclusively on AI infrastructure.
““through a balanced mix of operating cash flow, investment-grade debt, strategic partnerships, and equity capital.””
— Meta, in a statement dated October 26, 2026
Revenue Against Long-Term Obligations
The filings and financing announcements show how companies are raising money and accepting future obligations, but they do not establish how quickly AI products will generate enough revenue to cover the investment. The supplied material also does not quantify how much of the four-company spending total is specifically for AI, or how much risk is held by partners, property companies and other investors.
The $1.02 trillion lease figure is a calculated sum of future payments for leases that had not started as of September 30; it is not a bill due immediately. The source does not provide a comparable breakdown of all power contract commitments or the potential effect on customer electricity prices. Those outcomes remain uncertain.
Spending Plans and Returns Ahead
The next evidence will come from company filings and results showing whether capital spending continues at the current pace and how it is funded. Amazon’s stated 2026 cash capital spending outlook is about $220 billion, while Alphabet has maintained guidance of $195 billion to $205 billion for 2026 and expects a significant increase in 2027. Investors will also be watching whether revenue from AI services grows alongside infrastructure costs and whether companies take on more debt, leases or partnership commitments.
Key Questions
How much did the four companies spend in the third quarter of 2026?
Amazon, Microsoft, Alphabet and Meta spent about $202 billion on property and equipment from July through September, based on a sum of their company filings. The total includes non-AI investment as well as data centers, servers and chips.
Where does the money to build AI data centers come from?
The companies are using several sources: operating cash, bonds and loans, leases, strategic partnerships and equity financing. The mix differs by company and project.
What does the $1.02 trillion lease figure represent?
It is a calculation of future payments on leases, mostly for data centers, that had not yet commenced as of September 30, 2026. It is a long-term commitment figure, not an immediate payment due.
Who could bear the risk if AI revenues fall short?
The companies remain responsible for their own obligations, while lenders, bondholders, equity investors and partners may also have exposure. The supplied material does not quantify how losses would be distributed or establish the effect on electricity customers.
Source: Thorsten Meyer AI
