Written by: Rita

The market's biggest concerns about Microsoft focus on two points: aside from Azure's growth and Copilot's penetration rate, it is spending hundreds of billions of dollars building data centers. Bernstein directly responded to this bearish sentiment in a research report published on August 10, stating that Microsoft's expansion pace is restrained. The report's title made its stance clear: "Are they overbuilding capacity?" The answer is no.

The report breaks down these numbers from three dimensions: the growth rate of data center square footage, the timing distribution of leasing obligations, and the structure of hardware procurement commitments. The conclusion is that Microsoft's capacity expansion pace is even more restrained than the market expected.

Data Center Expansion Slower than Cloud Revenue Growth

Microsoft's current leasing liabilities have increased by 33% year-on-year to $114.4 billion, while future leasing obligations surged by 255% year-on-year to $329.1 billion. These numbers are indeed expanding.

However, Bernstein pointed out that the market has overlooked a key fact: Microsoft's total square footage (owned + leased) has long grown at a rate lower than that of commercial cloud revenue. Even in the most intensive AI construction year of 2026, this pattern remains unbroken. This means for every additional square foot of data center, there corresponds a greater amount of cloud revenue growth.

The growth rate of leased square footage is indeed faster than owned, but this is precisely Microsoft's strategic choice to quickly acquire capacity through leasing rather than self-building, avoiding large upfront land and construction funds. While the proportion of owned data centers is decreasing, the overseas leasing proportion is increasing.

The existing leasing liabilities of $114.4 billion are actually spread over a 13-year term. The rent due in 2027 is approximately $13.2 billion, which will subsequently decline year by year. This is not a cost shock that needs to be immediately digested.

$329 Billion in Leasing is Not as Terrifying as It Seems

The future leasing obligations of $329.1 billion are the main source of the market's panic, but Bernstein's breakdown makes it seem much more moderate. These leases will gradually start between 2027 and 2033, with lease terms varying from 1 to 20 years. Assuming the leases start evenly over 7 years with an average term of 12 to 15 years, the annualized rent expenditure growth rate would be approximately 12% to 16%.

This growth rate is roughly in line with Microsoft's historical growth rate of commercial cloud revenue. Bernstein believes that this pace is within a manageable range.

Microsoft stated explicitly in its 10-K that some leases "must meet certain contractual conditions." In reality, there are frequent news reports of data centers delayed due to local approvals, power access issues, etc., which means a significant portion of the $329.1 billion might not all convert into actual rent expenditure. Microsoft has also not defined these leases as irrevocable or unconditional, and some contracts may only require a termination fee to cancel.

Bernstein also conducted another sensitivity calculation. If the weighted average contract term for future leases is 12 years, the annualized rent expenditure growth rate is about 16%; if it is 15 years, the growth rate is about 12%. In either case, the growth rate is roughly equivalent to Microsoft's historical growth rate in commercial cloud revenue. The market views the $329.1 billion as "costs already incurred," but Bernstein argues that a more accurate perspective is "the capacity layout pace over the next 7 to 13 years."

Procurement Commitments Concentrated Within 12 Months

The hardware procurement is another number that has been exaggerated by the market. As of the end of fiscal year 2026 (June 2026), Microsoft's procurement commitments for fiscal year 2027 (July 2026 to June 2027) are $169 billion, while for fiscal year 2028 and beyond, it is only $25 billion.

Bernstein pointed out that procurement commitments cover a wide range of categories including power systems, cooling equipment, semiconductors, and networking equipment, of which GPU servers only make up a portion. Considering that Microsoft is still continuously replacing aging CPU servers, the actual proportion of procurement commitments intended for AI hardware is even smaller.

The concentration of procurement commitments within 12 months means Microsoft maintains great flexibility. If AI demand changes, it only needs to adjust the procurement plan for the next 12 months without bearing large hardware obligations for 2028 and beyond.

Bernstein also noted that AI accounts for about 16% to 17% of Microsoft's commercial cloud revenue. However, the gross profit margin of the AI business is much lower than that of traditional cloud business (about 27% vs. the high gross margins of traditional cloud business), thus AI may account for about 40% of the cost of commercial cloud sales. This is a structural detail that the market generally overlooks; the market interprets most of Microsoft's capital expenditure as "AI investment," but the capacity demand from non-AI cloud business also drives data center expansion. Regardless of market discussions about AI, most of Microsoft's cloud revenue still comes from non-AI business, and the newly built data centers can flexibly switch between CPU and GPU.

If AI Demand Slows, Microsoft Can Shift Capacity to Cloud

What if AI is proven false?

Bernstein's answer is that Microsoft can shift its data centers to traditional cloud business. The data centers are designed to support mixed CPU and GPU deployments, and decisions can be made at the last moment. If demand is insufficient, Microsoft can delay the launch of self-built data centers, cancel lease commitments (even if termination fees need to be paid), or even shut down some facilities, which would have relatively limited effects on margins and cash.

The management's response in the Q4 earnings call also pointed in the same direction. Amy Hood explicitly stated in response to Bernstein's question, "The current situation is clearly that demand far exceeds available supply." Within this framework, the premise of discussing overbuilding does not hold.

Bernstein raised Microsoft's target price from $647 to $660, maintaining an outperform rating, and adjusted the price-to-earnings ratio from 26.5 times to 27 times. The market's panic over Microsoft's leasing numbers is based on a misunderstanding of contract structures and timing distribution.

Disclaimer

This article is a compilation and interpretation of the third-party brokerage report (Bernstein, August 10, 2026) by Trend Research, combined with publicly available market information. The ratings, target prices, profit forecasts, and related judgments cited in this article are the views of the analysts of that brokerage and only represent the position of their affiliated institution, do not represent the views of Trend Research, and do not constitute any investment advice.

The market is risky, and decisions should be made independently. This article should not be used as the basis for buying or selling any securities.