Market attention is moving from Anthropic's record IPO to a plainer question: how much AI infrastructure is being built, and who is paying for it. Microsoft, Alphabet, Amazon and Meta are each committing well over $100 billion a year to AI data center spending, and several of them are now reporting negative free cash flow or turning to bond and stock markets to fund it.
This article looks at the spending itself, using the companies' own latest figures. Our earlier report on how chip stocks traded around the listing covers the market side. The numbers below come from third-quarter results reported in late October, and they differ in definitions and fiscal calendars, so they are not strictly comparable.
AI data center spending by company
Amazon has the largest plan. CEO Andy Jassy told investors the company still expects to spend about $220 billion in capital this year, mostly on AI data centers, according to The Associated Press. That is 10% above the $200 billion plan announced in February and well above the $128 billion spent last year. Third-quarter capital spending was $77.3 billion, up from $35.1 billion a year earlier.
Alphabet left its 2026 capital expenditure guidance at $195 billion to $205 billion, after raising it by $15 billion in July, CNBC reported. Capex in the third quarter was $51.4 billion, mostly for technical infrastructure supporting AI. The company said spending will increase significantly in 2027 and gave no figure.
Microsoft reports on a fiscal year that ends in June. It spent about $45.2 billion on capital expenditures, including finance leases, in the quarter ended September 30. CFO Amy Hood said spending should exceed $50 billion in the second quarter and rise toward about $170 billion for fiscal 2027, according to investingLive, a smaller outlet. An earlier outlook had pointed to about $175 billion, so the full-year figure has moved between sources.
Meta narrowed its 2026 capex outlook to $135 billion to $145 billion from $130 billion to $145 billion, according to its earnings release. Third-quarter capex, including finance lease principal, was $34.91 billion. Meta did not give a 2027 figure, saying only that it expects to keep investing aggressively. CNBC cited FactSet analyst estimates that put Meta's 2027 capex above $210 billion and Amazon's above $280 billion.
Cash flow is under strain
The spending is now visible in the cash numbers. CNBC reported negative quarterly free cash flow at Meta ($2.12 billion), Amazon ($42.4 billion) and Alphabet ($9.07 billion). Microsoft stayed positive at $18.63 billion, but that was down 27%, and CNBC noted its cash generation had fallen sharply for a second straight quarter. Meta's free cash flow a year earlier was $10.63 billion.
Investors reacted unevenly. Amazon shares fell about 7% in after-hours trading as attention turned to cash burn tied to AI spending. Forrester analyst Tracy Woo told CNBC that Alphabet's massive capex had been "tolerated for now." Christopher Ballard of Check Capital Management described the approach as "borrowing from the for-now in order to invest in the future."
How the AI buildout is being financed
Debt is carrying a growing share. Goldman Sachs counted five AI-related bond deals worth $67 billion since October 29, which brought hyperscaler debt issuance this year to $296 billion, CNBC reported. Goldman forecasts $440 billion of hyperscaler issuance next year, plus $300 billion of project-level debt. Morgan Stanley said Microsoft, Meta and Oracle sold new U.S. bonds after September 30, and Alphabet issued in Europe.
Equity is part of the mix too. Meta's board approved a new non-voting Class C stock, effective in January 2027, with stated purposes that include raising capital. Hood said Microsoft would consider other ways to raise capital if they were attractive, but that it has not needed to.
The pressure shows most at smaller builders. CoreWeave's 2032 bond, which carries a 9.625% coupon, yielded above 16% early in November, and its five-year credit default swaps spiked above 1,289 basis points on November 5, according to Deutsche Bank figures cited by CNBC. Jefferies strategist Christopher Wood said the most plausible way the AI capex arms race ends is that the credit card gets taken away. The same report notes that Nvidia's and Amazon's default-protection costs also rose, from low starting levels.
Morgan Stanley strategist Fernanda Lima drew a line between issuers. She sees an improving risk-reward for high-quality hyperscalers and expects more spread volatility for data center bonds and neoclouds, meaning the newer AI cloud providers. Spreads have tightened on most hyperscaler names since their recent deals priced, with Oracle the exception.
What could slow data center construction
Money is one limit, and local politics is another. CNBC reported that markets are weighing possible delays under incoming Democratic governors in Ohio, Texas and Georgia, and Goldman Sachs estimates that moratoriums could affect 45% of planned U.S. data center capacity. Lima said project timing may shift because Democrats swept Congress and voters favor local restrictions. National Economic Council Director Kevin Hassett expects data center growth to continue as forecast.
Supply is a constraint as well. Microsoft's first-quarter spending came in below what analysts expected because some third-party capacity deliveries were delayed, according to Hood. Spending is expected to speed up in the second half of the fiscal year as CPUs, GPUs and third-party capacity arrive.
Forecasts still point up. Morgan Stanley expects $1.66 trillion of AI capital expenditures in 2027, 75% more than in 2026, and $1.77 trillion in 2028. Bank of America expects AI-related credit issuance to rise 31% to $685 billion in 2027 and then fall 23% to $531 billion in 2028, CNBC reported.
The return question behind the capex
The companies say demand justifies the outlay. Amazon's cloud unit AWS grew 39.3% to $46 billion in the quarter, its fastest growth in nearly five years. Jassy said the opportunity "remains extraordinary" and that customers are "gobbling it up." Microsoft guided Azure growth to 49% to 50% for the next quarter, and Hood argued that efficiency gains let each capex dollar support more revenue over time.
What markets want is timing. Investors are asking how quickly new capacity turns into revenue, and how long the companies can fund it from operations before debt and equity take over. A strong IPO for one AI developer does not answer that, which is why the focus has returned to the spending. This article is a news summary and not investment advice.
Frequently asked questions
How much are the biggest tech companies spending on AI data centers?
Amazon expects about $220 billion this year, Alphabet $195 billion to $205 billion, and Meta $135 billion to $145 billion. Microsoft is guiding toward about $170 billion for its fiscal year ending June 2027.
Why are free cash flow figures turning negative?
Capital spending is growing faster than operating cash flow at some companies. CNBC reported negative quarterly free cash flow at Meta, Amazon and Alphabet, while Microsoft's stayed positive but fell 27%.
How is the AI buildout being paid for?
Mostly cash flow and a rising amount of debt. Goldman Sachs counts $296 billion of hyperscaler debt issuance this year, and Meta has approved a new share class that can be used to raise capital.
Is AI spending expected to keep rising?
Forecasters say yes. Morgan Stanley projects $1.66 trillion of AI capex in 2027, and Alphabet has said its own spending will rise significantly. Local restrictions and supply delays could shift the timing.
Why did the Anthropic IPO not settle the debate?
Analysts quoted by Bloomberg said the listing leaves open questions about infrastructure spending, circular financing and returns. It adds capital to frontier AI models but does not show how the spending will pay back.



