A $148 billion bet: Why does Goldman Sachs believe the market misunderstood Microsoft's capital expenditure?

A $148 billion bet: Why does Goldman Sachs believe the market misunderstood Microsoft's capital expenditure?

Since the financial report was released on January 28, Microsoft’s stock price has dropped about 15%. The market’s concerns focus on two core issues: first, the substantial upward revision of capital expenditure without a corresponding increase in Azure growth guidance, putting the investment return logic under scrutiny; second, the Office 365 business faces competition threats from the new generation of AI tools. Goldman Sachs stated in its latest report, short-term market concerns mask the rationality of Microsoft's mid-term strategy.

According to Wind Trading Desk, the report points out, Microsoft is proactively adjusting its computing power allocation structure: the proportion used for internal R&D and Copilot has increased from about 10% to 20%. If all added GPU capacity were invested in Azure, Azure’s Q2 growth rate could reach nearly 40%, which means some growth potential is being strategically deferred.

On capital expenditure, FY2026 is expected to reach $148 billion, a 400% increase over FY2022. The current computing power allocation is about 70% for Azure and 30% for Copilot and internal R&D; the allocation ratio for additional GPUs is more balanced, close to 60/40. Goldman Sachs believes Microsoft is prioritizing long-term strategic layout over short-term Azure revenue, which will translate into a stronger moat and returns in the mid-term.

"Iceberg Model": Visible and Invisible Computing Power Investment

Goldman Sachs introduces the "iceberg analogy" framework in its latest report to interpret Microsoft's current computing power capital expenditure strategy. The so-called "above the waterline" refers to computing power investments that are directly monetized and reflected in the quarterly results of Azure and Office 365; "below the waterline" refers to computing power deployment that has not yet generated direct income but is crucial to Microsoft’s long-term strategy. These investments do not show up in short-term reports but leave room for future monetization.

Microsoft management says that if all added computing capacity were invested in Azure, its Q2 growth rate at fixed exchange rates would exceed 40%. The actual reported growth rate is 38%. According to Goldman’s estimate, if all were allocated, Azure’s growth rate could reach the low 40% range, an increase of about 5 percentage points. This confirms that Microsoft is strategically deferring some computing power for internal R&D and long-term Copilot layout.

Update on Computing Power Allocation: Internal Use Proportion Significantly Increased

Goldman Sachs has raised the proportion of Microsoft’s internal computing power use in its latest report, from about 10% to 20%. This adjustment is based on the Q2 earnings call guidance as well as updated tracking of Copilot and other first-party applications and top engineers’ token usage. Microsoft previously stated its commitment to developing cutting-edge AI models.

For computing power capital expenditures in FY2026, Goldman Sachs established a detailed framework:

  • Azure AI computing power occupies the largest share, applying a 4-5x “computing power/quarterly incremental revenue” multiplier, i.e., each $1 of capital expenditure can generate $0.20 to $0.25 revenue in the following quarter. This portion implies a 6-year lifecycle gross margin below the current Azure core level.
  • Internal R&D computing power covers employee AI adoption, growing developer computing power needs, and development of Microsoft AI (MAI) internal models.
  • Application computing power includes pay-per-use AI computing power and the unmonetized Copilot computing power (such as Bing and free Copilot apps), applying a 3-4x multiplier.

Overall, Goldman Sachs estimates that in FY2026, about 70% of total computing power will go to Azure, while 30% will go to Copilot and internal uses. For newly added GPUs, the allocation is more balanced, close to 60/40.

Four Driving Factors Behind Increased Internal Allocation

Microsoft's computing power demand is being boosted by four structural factors, profoundly changing the capital expenditure structure.

First, Copilot adoption continues to rise. Both Microsoft’s internal employees and external customers have increasing reliance on Copilot, directly driving up computing power consumption.

Second, internal R&D efforts are significantly intensified. In February, Microsoft clarified it is developing absolutely cutting-edge AI models. Through the IP access agreement with OpenAI until 2032, the company will use branch versions of OpenAI models to build dedicated models for its own domain data, focusing on enterprise-level Copilot applications.

Third, the operational expense structure has fundamentally shifted. Microsoft expects future operational expense growth to be driven more by capital expenditure than employee numbers, because engineering workload is becoming increasingly power-intensive. Employees’ use of internal Copilot also requires incremental computing power. This trend is reflected in the 5% layoffs in 2023 and multiple rounds of personnel adjustments in 2025.

Fourth, overall baseline computing power for software stack is rising. Clients now expect AI features embedded in existing applications without additional fees. For vertically integrated hyperscale firms like Microsoft, this change directly shows as increased capital expenditure on computing power; for other software firms, it translates to token payments to third-party LLMs or hyperscale cloud providers.

Paths to Improve Azure Market Sentiment

In the next few quarters, as new capacity is brought online in phases, the computing power balancing between Azure growth, first-party apps, and internal R&D will no longer be a zero-sum game. This means Microsoft can advance Copilot and cutting-edge model development without sacrificing Azure growth.

The core obstacle for current investors is insufficient information disclosure. With capital expenditure intensity persistently revised upward while Azure expectations are repeatedly adjusted, the market urgently needs an analysis framework that distinguishes Azure from other business capital expenditures, to form a clearer judgment on mid-to-long-term growth and investment returns.

Goldman Sachs estimates that in Q2, about $4 billion in GPU expenditures by Microsoft were not directly related to Azure or monetized Copilot business. If this computing power were invested in Azure and given a 5x “computing power/quarterly incremental revenue” multiplier (implying a three-year payback period), Azure would gain about 3% upside, corresponding to a 5 percentage point increase in growth, raising the reported Q2 growth rate to 44% (43% at fixed exchange rates).

Compared with other hyperscale cloud providers, Microsoft relies more on internally developed software solutions, which, to some extent, occupy capacity that could otherwise be used for Azure. For example, demand for computing power driven by adoption of Microsoft AI coding tools is counted as capital expenditure, while other firms buying third-party software count it as operational expenditure. This structural difference also affects market assessment of Azure growth potential.

Recently, some funds have flowed to Google, mainly based on three factors: Gemini’s competitive positioning, TPU’s potential opportunities, and a relatively smaller application business exposure. Notably, after Google released 4Q25 earnings on February 4, its share price fell; its 2026 capital expenditure was about 50% higher than market expectations, but the earnings call provided a clear disclosure framework—60% of capital expenditure goes to servers, with a little more than half allocated to cloud business. Goldman Sachs believes if Microsoft could offer similar disclosure, it would help investors rebuild confidence.

Strategic Rationale for Copilot Investment

Microsoft is building Copilot as an enterprise-grade LLM solution balancing performance and cost, via differentiated delivery mechanisms. Copilot 365 is priced at about $30 per user per month and integrates multiple models; in contrast, ChatGPT Enterprise is about $60 and offers unlimited access to GPT-4. Individual users may still prefer the latter’s simplicity, but enterprises must balance performance and cost and assess the long-term strategic value of abstraction layers.

In contextual intelligence, WorkIQ is becoming Microsoft’s key channel for displaying differentiation. Released at last November’s Ignite conference, this tool analyzes anonymized, aggregated Microsoft 365 data to reveal collaboration patterns and work trends inside enterprises, supporting AI integration into business processes.

Microsoft also demonstrates distinctive strategic advantages as a "fast follower." Leveraging strong distribution channels, enterprise security capabilities, and service support, Microsoft can quickly introduce consumer-proven technologies to enterprise scenarios. Over the past three years, experience has shown the enterprise software stack iteration cycle is long, but management’s urgent demand for AI implementation is driving enterprises to seek incremental value beyond the current stack, which is Microsoft’s opportunity.

More importantly, Copilot has begun to contribute substantially to M365 commercial cloud revenue growth. This business is projected to reach $89 billion in FY2026. Excluding one-off project impacts, M365 commercial cloud revenue has maintained steady growth over the past three quarters, reversing previous slowing trends. Goldman Sachs believes Copilot monetization effectively offsets the pressure from the E5 upgrade cycle winding down and new seats from low-ARPU groups (such as regions, SMEs, frontline workers). ARPU growth over the past three quarters increased to 10-11%, well above the prior year’s 5-7%, also confirming Copilot adoption is accelerating.

~~~~~~~~~~~~~~~~~~~~~~~~

The above highlights are from Wind Trading Desk.

For more detailed interpretations, including real-time analysis and frontline research, please join [Wind Trading Desk · Annual Membership]

Risk Warning and DisclaimerMarkets are risky, investments require caution. This article does not constitute personal investment advice, nor does it take into account the specific investment goals, financial situation, or needs of any particular user. Users should consider whether any opinions, views, or conclusions herein are suitable for their specific circumstances. Invest at your own risk.