The world’s largest cloud operators are on course to spend more on artificial intelligence infrastructure than they generate in free cash flow by 2027, a Reuters analysis of LSEG consensus estimates found, placing debt markets at the center of the next stage of the AI buildout.
Microsoft, Alphabet, Amazon, Meta and Oracle are all expanding data-center capacity, buying advanced chips and securing power for AI workloads at a pace that exceeds their internally generated cash. Reuters calculated that each additional $1 of operating cash flow across the five companies corresponds to roughly $1.57 in additional capital expenditure through 2027.
The figures challenge the assumption that the biggest technology companies can fund the AI race largely from profits. Their balance sheets remain far stronger than those of typical corporate borrowers, but the scale of planned spending is beginning to turn financing decisions into a material part of the story.
Spending plans accelerate into 2026
Capital expenditure by Amazon, Alphabet, Microsoft, Meta and Apple totaled about $227 billion in 2024 and rose to $391 billion in 2025, according to the estimates cited by Reuters. Guidance and market estimates indicate the group’s spending could surpass $635 billion in 2026.
Amazon has set a 2026 spending target of $200 billion. Alphabet has guided for $175 billion to $185 billion, while Meta has indicated that its spending could reach as much as $135 billion. Those commitments alone show how the AI infrastructure race is extending beyond a short cycle of chip purchases into years of investment in data centers, networking equipment and electricity supply.
Goldman Sachs projects hyperscaler capital expenditure of between $1.15 trillion and $1.4 trillion across 2025 through 2027. That compares with $477 billion spent during the preceding three-year period, from 2022 through 2024.
Amin Mrini estimated that the five companies’ combined 2026 capital-spending commitments amount to about 3.8 times their combined free cash flow. Free cash flow measures the cash left after a company pays for operating costs and capital expenditure, making it a useful gauge of how much funding remains available without tapping debt or issuing shares.
The gap does not mean all five companies face the same financing pressure. Amazon, Microsoft, Alphabet and Meta operate profitable core businesses that produce substantial operating cash flow. Yet the aggregate figures show that the AI buildout increasingly competes with share buybacks, dividends, acquisitions and other corporate uses of cash.
Oracle offers the clearest example of pressure
Oracle stands out among the major infrastructure buyers because its capital spending has expanded much faster than its operating cash generation. Reuters reported that Oracle’s capex rose from 47% of operating cash flow in fiscal 2022 to 174% in fiscal 2026, which ended in May.
Over that period, Oracle spent $55.7 billion on capital expenditure while generating $32.0 billion in operating cash flow, according to the figures cited in the analysis. Its free cash flow turned negative, and its shares had fallen 36% year to date in the period covered by the report.
Oracle has outlined plans to raise between $45 billion and $50 billion through debt and equity funding for cloud expansion. The company’s position illustrates the financing challenge facing firms pursuing large AI infrastructure contracts: signed or anticipated demand can justify rapid construction, but data centers and computing equipment require funding well before customer revenue is fully collected.
Microsoft’s latest quarterly results also pointed to the scale of investment. The company reported $35.8 billion in operating cash flow, against $37.5 billion in capital expenditure including finance leases. Finance leases allow companies to use equipment while spreading payments over time, but they remain a long-term financial commitment tied to the infrastructure expansion.
Amazon reported trailing-12-month operating cash flow up 30% to $148.5 billion, while free cash flow fell to $1.2 billion, Reuters said. The contrast shows how rapidly high cash generation can be absorbed when a company simultaneously builds cloud capacity and supports an enormous retail and logistics operation.
Revenue expectations face a demanding hurdle
Mrini’s calculations put the commercial test for AI infrastructure into sharper terms. Using a 15% pre-tax return-on-investment benchmark, AI compute revenue would need to reach roughly $2.9 trillion by 2030 to cover the cost of deployed capital.
That would be more than three times the current cloud-computing market, from a base of about $211 billion, and would require approximately 68% annual growth for five consecutive years, according to the estimates. The required revenue level does not mean it cannot be reached, but it leaves little room for sustained overcapacity, slower corporate AI adoption or sharp declines in pricing.
Anthropic’s Economic Index adds another complication. The dataset cited in the analysis suggests that a 10% decline in the cost of an AI task is associated with only about a 3% increase in usage. If that relationship holds at scale, falling prices for AI services may stimulate demand without producing enough extra consumption to offset rapidly rising infrastructure costs.
AI revenue growth is currently estimated at roughly 25% to 30% annually, compared with capital-spending growth near 60% to 70%, according to Mrini’s analysis. The mismatch places pressure on cloud operators to convert AI experimentation into recurring, high-margin enterprise revenue.
Earlier infrastructure cycles offer a cautionary comparison
A Cornell Capital Group study cited by Reuters compared the AI construction boom with earlier periods in which transformative infrastructure delivered uneven returns to equity holders.
During the 19th-century US railroad boom, railroad stocks and bonds reached $10.6 billion, or about $400 billion in current-dollar terms, while US government debt stood at $1.2 billion. The study said 90% of railroad companies failed following the Panic of 1873, and 25% of track mileage entered receivership during the Panic of 1893.
Commercial aviation created vast economic value while producing relatively limited cumulative industry profits. The study estimated global airline net profit over the past 80 years at about $86 billion, before the sector lost roughly $190 billion during the three-year Covid-19 period.
The late-1990s telecommunications cycle offers a closer technological parallel. More than $500 billion flowed into fiber and wireless networks, but more than 23 major telecom companies had filed for bankruptcy by 2002, alongside about $2 trillion in market-value losses. The Cornell study said less than 5% of laid fiber had been lit at the time.
AI infrastructure may prove more commercially productive than those earlier networks, particularly if software and agents become embedded in corporate workflows. Yet the history of railroads, airlines and telecoms shows that lasting economic usefulness does not automatically translate into timely returns for the companies funding the initial buildout.
For cryptocurrency markets, the evidence supports caution rather than a fixed prediction. Large technology debt issuance could affect credit-market conditions and risk appetite, but no supplied data establishes that hyperscaler borrowing is already draining liquidity from digital assets or guarantees a specific market move. The more immediate issue is whether AI revenue begins to catch up with the capital now being committed.
Worried about AI-driven capex risks? Learn how digital assets reshape funding strategies for large-scale infrastructure.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
