Microsoft Is Not Overbuilding Data Centers; Flexible Capacity and Limited Forward Hardware Commitments Support a Higher Target Price

Overview

The report analyzes market concerns that Microsoft may be overbuilding data centers due to the AI boom. Bernstein believes that seemingly large capital expenditures, lease obligations, and purchase commitments cannot be directly equated with AI-dedicated investment: most of Microsoft's commercial cloud revenue still comes from non-AI businesses, data centers support traditional cloud, SaaS, and AI workloads simultaneously, and new leases and equipment purchases are highly spread out over time and adaptable to demand. Based on a weakening bearish thesis and the stock's valuation still below its previous range, the report raises the target valuation multiple and increases the target price.

Core views

First, Microsoft's new data center footprint has long grown more slowly than commercial cloud revenue, so the pace of capacity expansion is not aggressive. Second, US$329.1 billion of future lease obligations will be brought online gradually over seven years and correspond to contract terms of up to 20 years; viewed on an annual expense basis, the burden is far lower than the pressure implied by the nominal total. Third, purchase commitments are mainly concentrated in FY2027 and include power, cooling, infrastructure services, servers, networking, and storage, not all of which are for GPUs. Fourth, aside from GPU servers, most facilities and equipment can be used for traditional commercial cloud; if AI demand slows, Microsoft can shift capacity to CPU cloud workloads, other customers, Copilot, or SaaS features. Fifth, supply remains below demand, AI product gross margins continue to improve, and Microsoft lacks large-scale forward AI hardware purchase obligations, making the risk of excess capacity relatively limited.

Analysis framework

The report breaks total commitments into existing leases, future leases not yet commenced, and purchase commitments, and examines their commencement timing, contract terms, and annual costs separately; it uses Microsoft's disclosed owned and leased property footprint as an approximate indicator of data center expansion and compares it with commercial cloud revenue growth; it then combines AI revenue, gross margin estimates, and cloud business cost share to assess the actual use of capital investment. For the undisclosed future lease rollout schedule, the report assumes even commencement from FY2027 to FY2033 and conducts sensitivity analysis using weighted average lease terms of 12 to 15 years.

Methodology notes

• Capacity Supply-Demand AnalysisComparison of Data Center Footprint and Cloud Revenue Growth

  Use total property footprint growth as an approximate proxy for the pace of data center capacity expansion and match it against commercial cloud revenue growth.

  Microsoft does not directly disclose total data center area, so the report uses owned and leased property area disclosed in the 10-K as a proxy. Since office space and retail stores have not increased rapidly, this metric is considered to reasonably reflect the trend in cloud data center construction, though proxy-variable error remains.

• Contract Liability AnalysisAnnualized Lease Obligation Model

  Convert total long-term leases into annual payments based on commencement year and contract term to avoid judging the burden solely by the nominal total.

  The model assumes US$329.1 billion of future leases commence evenly from FY2027 to FY2033 and uses weighted average lease terms of 12 to 15 years. Because actual commencements may not be even and some contracts may include conditions, the calculation is scenario analysis rather than company guidance.

• Investment Use AnalysisAI and Traditional Cloud Cost Attribution

  Assess AI's share of infrastructure investment based on AI recurring revenue, estimated gross margin, and commercial cloud costs.

  The report estimates that AI will account for about 16% to 17% of commercial cloud revenue in FY2026, but because of its lower gross margin, it may account for about 40% of commercial cloud cost of sales. However, data center refreshes, traditional cloud expansion, and CPU demand from SaaS mean that not all capital expenditures and leases can be attributed to AI.

• Relative ValuationP/E Target Valuation

  Adjust the target P/E multiple based on changes in fundamental risk.

  The report believes concerns about excess capacity are overstated, so it raises the target P/E by 0.5x to 27x and increases the target price from US$647 to US$660.

Asset mapping & comparison

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

• MSFT.US

  Core subject of the report, with a bullish view and a raised target price.

  Strengths

  Commercial cloud revenue growth is steady, and data center capacity can be flexibly configured between CPUs and GPUs; future leases commence in phases, forward hardware purchase commitments are limited; traditional cloud and AI jointly support demand.

  Weaknesses

  The AI business currently has a low gross margin, while capital expenditures, lease obligations, and infrastructure costs are large in scale, potentially pressuring near-term free cash flow and margins.

  Comparison

  Microsoft's expected FY2027 P/E is about 25.1x; the report raises the target P/E to 27x, arguing that the current valuation remains below its previous valuation range.

  Risks

  AI demand below expectations, slowing cloud growth, faster-than-expected project commencements, rising lease and equipment costs, and AI gross margin improvement falling short of expectations.

Key data

• Target PriceUS$660.00Prior target price was US$647.00; potential upside versus the US$499.99 closing price is about 32%.
• Future Lease ObligationsUS$329.1 billionUp 255% year over year; expected to commence from FY2027 to FY2033, with contract terms of 1 to 20 years.
• Existing Lease LiabilitiesUS$114.4 billionSpread over 13 years; expected FY2027 payments of US$13.2 billion, declining gradually thereafter.
• FY2027 Purchase CommitmentsUS$169.0 billionIncludes power, cooling, infrastructure services, and data center hardware, not all of which is AI computing equipment.
• Purchase Commitments After FY2027US$25.1 billionForward commitments are relatively small, indicating Microsoft has not locked in large amounts of AI hardware purchases over the long term.
• FY2026 Commercial Cloud RevenueUS$214.4 billionFourth fiscal quarter grew 9% sequentially; AI accounts for only part of commercial cloud revenue.
• AI Recurring RevenueUS$37.0 billionAs of the third fiscal quarter of FY2026; the report estimates AI accounts for about 16% to 17% of commercial cloud revenue.
• AI Cost of Sales ShareAbout 40%Estimated based on an AI gross margin of about 27%, reflecting the currently lower gross margin of the AI business.
• FY2027 Adjusted EPS ForecastUS$19.96FY2026 actual was US$17.28, and FY2028 forecast is US$24.01.
• FY2027 Revenue ForecastUS$392.630 billionFY2026 actual revenue was US$331.839 billion, and FY2028 forecast is US$466.335 billion.

Impact & implications

If the report's view is correct, the market discount related to returns on AI capital expenditure and idle data center capacity is likely to continue narrowing, and Microsoft's valuation can benefit from cloud demand growth, supply release, and AI gross margin improvement. New capacity serves not only AI but can also support the migration of traditional enterprise workloads to Azure, so it retains high reuse value even in a downside scenario. However, near-term capital investment and lease payments will still rise, and investors need to distinguish between accounting classification changes, annual cash burden, and long-term nominal commitments rather than assessing risk solely based on total commitments.

Risks

• If AI demand is significantly below expectations, utilization of dedicated equipment such as GPUs may be insufficient, putting pressure on investment returns and margins.
• Disclosure on the actual commencement schedule, contract terms, and use of future lease obligations is limited, and the model assumptions of even commencement and a 12- to 15-year average lease term may deviate from reality.
• Rising power, cooling, construction, and hardware costs may cause investment amounts to continue growing faster than data center footprint.
• Although AI product gross margins are improving, they remain below those of traditional software and cloud businesses, and changes in revenue mix may weigh on overall gross margin.
• If traditional commercial cloud growth slows, the buffering ability to shift data centers from AI workloads to CPU workloads will also weaken.
• Some data center projects may be delayed due to legal, approval, or construction issues, affecting capacity go-live and revenue realization timing.
• Large capital expenditures and lease payments may continue to suppress free cash flow and increase investor concerns about capital allocation efficiency.

What to watch

• Whether revenue growth in Azure, commercial cloud, and non-AI cloud businesses can continue to exceed growth in data center footprint and lease expenses.
• The actual commencement pace, annual payments, and changes in contract terms for future leases from FY2027 to FY2033.
• The specific composition of GPUs, power, cooling, and traditional servers within capital expenditures, new leases, and purchase commitments.
• Quarterly improvement in AI recurring revenue, inference demand, and gross margins of various AI products.
• After new data center capacity comes online, the supply-demand gap, utilization, and realization of customer demand.
• Whether purchase commitments after FY2027 remain at a relatively low level or new long-term hardware lock-ins emerge.
• Realization of forecasts for FY2027 revenue of US$392.630 billion, adjusted EPS of US$19.96, and operating margin of 47.7%.