As the summer earnings season concludes, Microsoft Executive Vice President and Chief Financial Officer Amy Hood has left the market with an impression of disciplined spending. Unlike cloud competitors Amazon and Alphabet (Google's parent company), which have burned significant cash by heavily investing in AI, Microsoft not only generated $19.6 billion in free cash flow in the quarter ending June but also predicted it would continue to generate positive cash flow for at least the next year. This performance shifted market sentiment, boosting the stock price by a cumulative 29%.
One aspect the market has paid less attention to: how Microsoft maintains positive cash flow. The primary reason: compared to Amazon and Google, Microsoft leases data center resources from third parties on a larger scale, including partners like CoreWeave, a new cloud computing service provider (Neocloud). This approach lowers short-term capital expenditures but may weaken the company's control over costs in the long run.
The second key factor: in its June quarterly report, Microsoft disclosed that it adjusted its accounting methods, moving some expenses out of the capital expenditure category, thereby reducing its capital expenditure guidance for the 2026 calendar year by $15 billion, an 8% decline. Charles Fitzgerald, a Seattle-based angel investor and former Microsoft executive, said, "Many traders are simply looking at the headline numbers. Seeing lower CapEx, they interpret it as a positive. But the actual scale of investment hasn't been reduced; it's just a change in accounting classification, shifting from capital expenditure to operating expenditure. Microsoft hasn't canceled a large number of data center construction plans, nor has it revoked server orders from Nvidia."
This accounting adjustment highlights the operational leeway companies have in reporting financial data. Microsoft CFO Amy Hood explained on the earnings call that, effective July 1 (the start of Microsoft's fiscal year 2027), the company is extending the estimated useful life of data centers and office buildings from 15 years to 25 years. Hood stated that this adjustment "will result in more data center leases being classified as operating leases rather than finance leases," meaning the associated lease payments will no longer be counted as capital expenditures.
To simplify the distinction between the two lease accounting methods: under a finance lease, the data center is treated as if Microsoft directly purchased the asset. Under an operating lease, it is treated as an asset rental, with related expenses recorded as operating expenses rather than capital expenditures. Operating expenses reduce book profit, while capital expenditures reduce free cash flow—a key metric highly watched by investors amid high industry capital spending. Extending the estimated useful life of data centers also means that the depreciation expense for each new data center will be spread over a longer period, reducing the impact on profitability in a single period. However, Hood acknowledged that the impact of this adjustment this fiscal year is "limited," suggesting that the depreciation pressure will gradually emerge in subsequent years.
Microsoft still plans to significantly increase capital expenditures. In the first quarter of fiscal year 2027 (starting July 1), the company expects capital expenditures to rise 43% year-over-year, exceeding $50 billion. Hood said that driven by rising demand for cloud computing power, capital spending is expected to continue expanding throughout the entire fiscal year 2027, ending next June.
Where the real brake on capital spending growth lies
The more important factor in curbing capital expenditure growth is Microsoft's continued reliance on renting computing power from external vendors, especially new cloud providers like CoreWeave and Nebius. In its securities filings, Microsoft wrote, "In addition to our owned and operated data centers, we rely on third-party providers (including colocation facilities, leased data centers, and cloud infrastructure providers) to host some of our workloads."
In comparison, Amazon Web Services primarily operates its own data centers and only leases computing power on a small scale, such as a deal with Cipher Mining last November. Google's situation is similar, though it has recently been securing external computing power, most notably a reported $30 billion deal to lease capacity from SpaceX. Similar collaborations are expected to increase. Google CFO Anat Ashkenazi said on the company's earnings call last month that it plans to "expand third-party computing procurement" in the third quarter as a bridge before its own capacity comes online. At the same time, she acknowledged that the cost of externally purchasing computing power is high and "will put pressure on the cloud business's operating margin."
Microsoft started such collaborations earlier and has a broader partner ecosystem. Bloomberg first reported that Microsoft has committed at least $60 billion to multiple new cloud computing companies, including Nscale, Nebius, Iren, and Lambda, to lease data centers equipped with AI chips and supporting hardware, with most contracts lasting five years. Fitzgerald commented that external data center leases are generally classified as operating leases. "There is no doubt that a significant portion of Microsoft's infrastructure investment is reflected in operating expenses, not capital expenditures."
Microsoft's large-scale procurement of third-party computing power stems from the need to quickly secure AI computing resources after the launch of ChatGPT. Building data centers takes time; leasing capacity from providers that have already completed construction, connected to power grids, and have cooling systems allows Microsoft to rapidly fill capacity during a demand surge, supporting both OpenAI's large model training and Azure customer needs. Fitzgerald commented, "Microsoft got off to a faster start because it felt the explosion in demand related to OpenAI first and locked in infrastructure early. Amazon took longer to advance its computing infrastructure buildout."
Currently, the outside world cannot know the exact ratio of Microsoft's self-owned data centers versus third-party hosted computing power. The weaknesses of this model are clear: over the long term, Microsoft will face the risk of rising rental costs from partner cloud providers. Of course, Microsoft's five-year contracts with third-party computing providers partially buffer the impact of price increases from these new cloud providers.
D.A. Davidson analyst Gil Luria pointed out a possibility: in the next three to five years, once Microsoft's own data center capacity is built, it may "no longer need to purchase new cloud computing power from external providers." Luria added that Microsoft could sublease this computing power to Azure customers (a prime example being OpenAI), charging end customers a price higher than what it pays to the new cloud providers, thus earning a spread. However, Fitzgerald issued a warning: "The entire industry is currently scrambling for available computing power. As long as third-party operators hold power permits and have the ability to expand capacity, and as long as market demand remains hot, companies in need of computing power will have no choice but to accept the quoted prices."
Comments