Microsoft Is Not Overbuilding Data Centers; Flexible Capacity and Limited Forward Hardware Commitments Support a Valuation Upgrade

Overview

The report analyzes the core controversy of whether Microsoft’s large-scale capital expenditures and data center leases are creating excess AI capacity. Bernstein believes the market is simply equating capital expenditures, leases, and purchase commitments with AI investment, while overlooking traditional Commercial Cloud growth, equipment refreshes, contract terms, and facility reusability. After integrating data center footprint, lease payment schedules, and purchase commitment duration, the report concludes that Microsoft’s expansion approach is relatively prudent, and that capacity growth is consistent with historical cloud business growth.

Core views

First, Microsoft’s new capacity serves both AI and non-AI businesses, with AI accounting for only about 16% to 17% of FY26 Commercial Cloud revenue, so not all investment can be attributed to AI. Second, total data center footprint growth is lower than Commercial Cloud revenue growth, with no clear evidence of premature expansion. Third, although total future lease obligations are large, they will come online gradually over seven years and be spread over contracts of up to 20 years, leading to relatively moderate annual expense growth. Fourth, in addition to GPUs, power, cooling, networking, storage, and mixed-use facilities can all serve traditional cloud demand. Fifth, purchase commitments after FY27 are relatively limited, allowing the company to accelerate or slow equipment purchases based on demand.

Analysis framework

The report combines balance sheet disclosures, 10-K lease data, purchase commitments, facility area, and Commercial Cloud revenue. It first distinguishes between existing leases and leases that have not yet commenced, then assumes future leases come online evenly from FY27 to FY33 and conducts sensitivity analysis using 12- to 15-year weighted average lease terms. At the same time, it incorporates AI revenue, estimated gross margin, and cloud business cost structure to assess the portion of capital investment truly related to AI, and finally forms its valuation conclusion through earnings forecasts and target P/E multiple adjustments.

Methodology notes

• Capital expenditure and capacity analysisRevenue–physical capacity growth comparison

  Compare the long-term growth rates of data center facility area and Commercial Cloud revenue

  If physical area growth consistently lags cloud revenue growth, it usually implies higher revenue carried per unit area and does not support a conclusion that capacity is significantly ahead of demand. The report uses total owned and leased area disclosed in the 10-K as an approximate indicator of data center expansion.

• Lease liability analysisAnnualization of lease obligations

  Convert total lease contract amounts into expected payments by year

  The report distinguishes between existing leases and future leases that have not yet commenced, while considering commissioning years and contract terms, to avoid misinterpreting the total amount of multi-year contracts as single-period capital pressure.

• Scenario and sensitivity analysisLease term sensitivity analysis

  Test the impact of different weighted average lease terms on annual lease expenses

  Under the assumption that future leases come online evenly from FY27 to FY33, the report uses 12- to 15-year weighted average lease terms to estimate annual payments and their year-over-year growth.

• Business mix analysisAttribution of AI and non-AI investment

  Estimate AI-related resource consumption based on revenue, gross margin, and cost structure

  Using AI ARR, Commercial Cloud scale, and an AI gross margin of about 27%, the report estimates that AI may account for about 40% of cloud business costs, while also noting that mixed-use facilities and equipment refreshes may make the pure AI investment share lower.

• Valuation analysisP/E target valuation

  Adjust the target P/E multiple based on reassessment of fundamental risks

  Given that overcapacity risk is lower than market concerns suggest, the report raises the target P/E by 0.5x to 27x, thereby increasing the target price from US$647 to US$660.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Microsoft Corp (MSFT.US)

  Core recommended stock, rated Outperform

  Strengths

  Commercial Cloud is large in scale and continues to grow; data centers can support both CPU and GPU workloads; future hardware purchase commitments have shorter duration; AI product gross margins are improving; earnings are expected to maintain relatively rapid growth in FY27 to FY28.

  Weaknesses

  The AI business currently has lower gross margins than the traditional cloud business, capital expenditures and lease obligations are large, and the near-term free cash flow yield is relatively limited.

  Comparison

  The report notes that Microsoft still trades at a significant discount to its historical valuation multiple range; in the peer comparison table, its future revenue growth is about 17.8%, but its valuation is higher than most mature software companies and lower than some high-growth cloud software companies.

  Risks

  AI demand falling short of expectations, slower cloud business growth, front-loaded lease commencement, rising GPU and data center costs, and insufficient improvement in AI gross margins could all weaken investment returns and the target valuation.

• Microsoft Commercial Cloud and Azure

  Main foundation for capacity demand and investment returns

  Strengths

  Non-AI Azure, SaaS, and IaaS/PaaS migrations still have room to grow, and AI inference may also drive CPU and other cloud workloads.

  Weaknesses

  The business requires continued investment in facilities, power, cooling, and servers, and a rising AI share may temporarily depress overall gross margins.

  Comparison

  Traditional cloud business revenue is significantly larger than AI revenue, which is key evidence that new data centers are not being built solely for AI.

  Risks

  A slowdown in enterprise cloud migration, intensified competition, or customer optimization of cloud spending could reduce utilization of new capacity.

Key data

• Rating and target priceOutperform; target price US$660Previous target price was US$647; target P/E raised by 0.5x to 27x.
• Closing price and upsideUS$499.99; 32%Closing date was August 7, 2026.
• Existing lease obligationsUS$114.4 billionAs of FY26 Q4, obligations are distributed over about 13 years; FY27 annual payments are about US$13.2 billion, then gradually decline.
• Future lease obligationsUS$329.1 billion, up 255% year over yearExpected to commence gradually from FY27 to FY33, with contract terms of 1 to 20 years; some arrangements remain subject to contractual conditions.
• Purchase commitmentsUS$169.0 billion in FY27; US$25.1 billion after FY27Scope includes power, cooling, infrastructure services, servers, networking, and storage equipment, and is not entirely GPUs or AI hardware.
• FY26 Commercial Cloud revenueUS$214.4 billionFY26 Q4 grew 9% quarter over quarter.
• AI business scaleFY26 Q3 AI ARR was US$37.0 billionThe report estimates AI accounts for about 16% to 17% of FY26 Commercial Cloud revenue.
• AI cost contributionAbout 40% of Commercial Cloud costsEstimated based on an AI gross margin of about 27%; management stated that gross margins for each AI product continue to improve sequentially.
• Growth in future lease paymentsAverage of about 12% to 16% from FY27 to FY31Depends on the weighted average term of future leases being 12 to 15 years and assumes even commencement from FY27 to FY33.
• Adjusted EPSFY26A US$17.28; FY27E US$19.96; FY28E US$24.01Corresponding FY27E and FY28E adjusted P/E ratios are about 25.1x and 20.8x, respectively.
• Revenue forecastFY26A US$331.839 billion; FY27E US$392.630 billion; FY28E US$466.335 billionExpected compound annual growth rate from FY26 to FY28 is 18.5%.

Impact & implications

If the report’s judgment proves correct, investors need to reduce the risk discount applied to Microsoft for uncontrolled AI capital expenditures and idle data centers. The mixed-use nature of data centers, long-term growth in traditional cloud businesses, and shorter visibility on hardware purchases allow Microsoft to adjust GPU and CPU deployment according to demand. With improved Q4 results and guidance, continued tight supply-demand conditions, and the market’s renewed understanding of the annual impact of lease obligations, Microsoft’s valuation is expected to recover toward its historical range. However, total lease obligations remain large, and investment returns ultimately depend on demand realization and gross margin improvement for Azure, Copilot, and other AI services.

Risks

• AI demand growth may fall short of expectations, leading to lower GPU server utilization or a longer investment payback period.
• AI business gross margin improvement may be slower than expected, continuing to drag down the overall gross margin of Commercial Cloud.
• Future leases may not commence evenly; if commissioning is concentrated in earlier years, annual cash outflows could exceed model estimates.
• The report lacks detailed information on the use, commencement timing, and weighted average term of future leases, and the sensitivity analysis relies on relatively strong assumptions.
• Power, cooling, chip, networking, and construction costs may continue to rise, causing investment per unit area to grow faster than physical area.
• Legal approvals, construction delays, or unmet contractual conditions may change data center delivery timing and capacity planning.
• If growth in traditional Azure, SaaS, and IaaS/PaaS slows at the same time, the buffering ability to shift facilities from AI to CPU workloads will weaken.
• The current valuation still reflects high earnings growth expectations, and results or guidance below expectations could trigger valuation compression.

What to watch

• The match between Azure and Commercial Cloud revenue growth and the pace of new data center commissioning.
• Quarterly changes in AI ARR, Copilot adoption, inference demand, and gross margins for each AI product.
• Trends in capital expenditures, finance leases, operating leases, and free cash flow.
• The actual commencement schedule, contract terms, and cancellation conditions of the US$329.1 billion in future lease obligations.
• The specific composition of GPUs, servers, power, cooling, and other infrastructure within FY27 purchase commitments.
• Whether purchase commitments after FY27 continue to remain at a low level or rise again due to demand growth.
• Growth in owned and leased data center area, as well as changes in cost and revenue output per unit area.
• Management’s latest comments on the AI supply-demand gap, capacity utilization, and the ability to convert facilities between CPU and GPU use.