Unbiased AI-powered news
The combined free cash flow of Amazon, Alphabet, Microsoft and Meta is forecast to fall to about $4bn in the third quarter, down from an average of $45bn per quarter since the Covid-19 pandemic. Full-year free cash flow for the four companies is projected to reach its lowest level since 2014. The shift reflects record capital expenditure on artificial intelligence infrastructure.
Abc NewsThe combined free cash flow of Amazon, Alphabet, Microsoft and Meta is expected to fall to roughly $4bn in the third quarter. That compares with an average of $45bn per quarter since the Covid-19 pandemic six years ago. Full-year free cash flow for the four companies, known as hyperscalers, is set to hit the lowest level since 2014.
At that time their combined revenues were about one-seventh of current levels. Free cash flow is watched as the cash remaining after operating costs and capital spending that can service debt or return value to shareholders. The companies have shifted from relatively asset-light cash generators to some of the world's largest investors in physical infrastructure.
This represents the deepest industry-wide capital expenditure cycle they have experienced. After funding early AI investments largely from income, the companies now face trade-offs such as cutting jobs, reducing shareholder returns or borrowing.
Amazon is expected to spend more cash than it generates this year. Meta will burn cash in the second half of the year, as will Microsoft in at least one quarter. Alphabet's full-year free cash flow is projected to remain positive but fall to its lowest level in more than a decade.
The companies began the capital expenditure increase with strong balance sheets. Analysts project cash generation will improve next year as the AI spending generates additional revenue. One company bought back no stock in the first quarter for the first time since its share repurchase programme began in 2015.
During that quarter the company issued $31bn in new debt and on Tuesday issued another $17bn in euro and Canadian dollar bonds. Another company has issued $55bn of debt over the past six months while pausing share buybacks. The pause is the longest since it began repurchasing its own stock in 2017.
Unlike some peers, one company does not have a cloud business to rent out space in the data centres it builds. Executives there have turned to staff cuts to free resources for the investment programme. Another company is expected to burn about $10bn in cash during the year and has said it will invest $200bn in 2026.
Tech groups including Meta have shifted tens of billions of dollars of data centre projects off their balance sheets using special-purpose holding companies. These vehicles can bring in Wall Street investors and raise debt that does not fully appear on the tech company's balance sheet.
Oracle has also used off-balance-sheet structures for data centre builds under a $300bn contract with OpenAI and began burning cash last year. The AI spending surge has flowed into a strained hardware supply chain. This has increased prices for components such as memory chips and made data centres more expensive.
One company said price inflation would add $25bn to its capital expenditure needs this year, while another cited rising costs as it added $10bn to its investment forecast. The value of servers, networking equipment and software on one company's balance sheet has more than tripled since mid-2022, from $61bn to $191bn.
An accounting professor at the University of Chicago’s Booth School of Business said free cash flow is not defined in standard accounting rules, giving companies discretion in its calculation. The professor added that the real free cash flows of many hyperscalers are probably worse than reported.
The spending pattern resembles capital cycles in industries such as telecoms or chemicals, where over-investment can lead to overcapacity. Tech executives have said they feel compelled to invest when competitors do to avoid being left behind by a technology they view as transformational.
Single source — no framing comparison available.
livemint.comZ.ai's GLM-5.2, released in June 2025, ranks as the top open-source model by some benchmarks and fifth on OpenRouter usage. DeepSeek's earlier R1 release in January 2025 triggered a trillion-dollar drop in U.S. tech values and calls for bans.
eonline.comChris Fall stepped down after three months leading the Center for AI Standards and Innovation. Arvind Raman was named acting director. The agency operates under the Department of Commerce.
wccftech.comAlphabet is developing an internal server chip called Frozen v2 slated for 2028 release. The chip targets six to ten times greater efficiency than existing designs on tokens generated per unit of power. Alphabet shares rose 3 percent after the report.