After the summer earnings season concluded, Microsoft demonstrated a cautious spending approach. Unlike competitors Amazon and Alphabet, which are heavily investing in artificial intelligence and consuming substantial cash, Microsoft generated $19.6 billion in free cash flow in the quarter ending June and anticipates stable cash flow generation for at least the next year. This strong performance shifted market sentiment and boosted confidence, driving the company's stock up 29% year-to-date.
External observers have not fully grasped the logic behind Microsoft's sustained positive cash flow. A key reason is that, compared to Amazon and Google, Microsoft has significantly leased data center computing capacity from new AI-focused cloud providers like CoreWeave. This approach lowers short-term capital expenditures but, over the long term, weakens Microsoft's own cost control leverage.
Another reason was disclosed in Microsoft's June quarter earnings report: the company implemented an accounting adjustment that moved some expenses out of the capital expenditure category, directly reducing the expected capital expenditure for the 2026 calendar year by $15 billion, or 8%.
Charles Fitzgerald, a Seattle angel investor and former Microsoft executive, noted, "Many quantitative trading programs only look at earnings headlines. Seeing a decline in capital expenditure numbers, they deem it positive. But the total actual spending hasn't changed; only the accounting category has shifted from capital expenditure to operating expenditure. Microsoft hasn't halted a large number of data center projects or canceled orders for Nvidia servers."
This accounting adjustment vividly illustrates the discretion companies have in polishing financial data. Microsoft CFO Amy Hood revealed during the earnings call that starting July 1 (the beginning of Microsoft's 2027 fiscal year), the company will extend the estimated useful life of data centers and office buildings from 15 to 25 years.
Hood explained, "Due to this adjustment, going forward, the majority of data center lease transactions will be classified as operating leases rather than finance leases, and the corresponding rent expenses will no longer be included in capital expenditure."
For example, under a finance lease, a data center is treated as if Microsoft directly purchased the asset. However, under an operating lease, it is only considered a rental, with rent expenses recorded as operating expenditure rather than capital expenditure. Operating expenditure directly reduces book profit, while capital expenditure lowers free cash flow—a key metric investors closely monitor amid the industry's widespread increase in capital spending.
Extending the estimated useful life of assets means the depreciation expense for a single new data center is spread over a longer period, reducing the impact on per-period profit. However, Hood acknowledged that the impact of this adjustment in the current fiscal year is "very negligible," implying that depreciation pressure will gradually emerge over the coming years.
Microsoft still plans to significantly boost capital expenditure. In the first quarter of the 2027 fiscal year, starting July 1, the company expects capital expenditure to surge 43% year-over-year, exceeding $50 billion total. Hood stated that as demand for cloud computing power rises, capital expenditure will continue to grow throughout the entire 2027 fiscal year ending next June.
The core factor truly suppressing Microsoft's capital expenditure scale is the extensive leasing of external computing power, primarily from new AI cloud providers like CoreWeave and Nebius. In regulatory filings, Microsoft states, "In addition to our owned and operated data centers, we also rely on third-party service providers to support some business operations, including colocation spaces, leased data centers, and cloud infrastructure providers."
In contrast, Amazon Web Services (AWS) operates the vast majority of its computing power through its own data centers, with only a small amount leased externally, such as the computing power lease agreement with Cipher Mining last November.
Google's situation is similar, though it has also gradually implemented external computing power procurement agreements, such as the recently announced $300 billion deal to lease computing power from SpaceX. Such collaborations are expected to increase. Google CFO Anat Ashkenazi said on last month's earnings call that the company will expand third-party computing power procurement in the third quarter, viewing it as a bridging solution before its own computing power comes online. She also noted that third-party computing power procurement costs will put some pressure on the cloud business's operating margin.
Microsoft has been earlier and more extensive in pursuing such collaborations. The company has committed at least $60 billion to multiple AI new cloud providers, including Nscale, Nebius, Iren, and Lambda, for leasing data centers fully equipped with AI chips and supporting hardware, with most contracts lasting five years.
Since external data center leases are generally classified as operating leases, Fitzgerald stated, "There's no doubt that a significant portion of Microsoft's infrastructure spending is reflected in operating expenditure, not under the capital expenditure line item."
Microsoft's extensive leasing of third-party computing power stems from the sudden surge in AI computing demand following the launch of ChatGPT, requiring the company to quickly secure computing capacity. Building its own data centers takes a long time, whereas leasing already-built, connected, and cooled computing power can rapidly fill the gap during the demand surge, serving both OpenAI's self-developed models and Azure customers.
Fitzgerald commented, "The reason Microsoft got a head start on all competitors is that it was the first to anticipate the explosive growth in OpenAI's computing power needs and managed the entire computing infrastructure supply. Amazon took a long time to catch up with the pace of computing expansion."
Currently, the specific proportion of Microsoft's own data centers versus third-party leased computing power is not publicly disclosed. However, the drawbacks of this model are obvious: over the long term, Microsoft will have to bear the risk of its partner cloud providers raising computing power pricing.
Of course, Microsoft has signed five-year long-term agreements with various external cloud providers, which somewhat mitigates the impact of price increases from new cloud providers. D.A. Davidson analyst Gil Luria believes that after Microsoft's own data center capacity comes online in the next three to five years, the company may no longer need to rely on external AI computing power. Additionally, Microsoft resells the leased computing power to Azure customers (the largest being OpenAI), pricing it higher than the rental cost paid to new cloud providers, thereby earning a spread.
However, Fitzgerald offered a different view: "The entire industry is frantically competing for available computing resources. Once computing prices rise, third-party providers who control power resources and can quickly build and deploy computing capacity will have absolute leverage, and customers with real needs will have to passively accept price increases."
Comments