Since its earnings report on January 28, Microsoft's stock price has declined by approximately 15%. Market concerns have centered on two core issues: first, a significant upward revision in capital expenditures without a corresponding increase in Azure growth guidance, raising questions about the investment return logic; second, the competitive threat posed by new-generation AI tools to the Office 365 business. In a recent report, Goldman Sachs stated that short-term market worries are overshadowing the rationality of Microsoft's medium-term strategy.

According to the report, Microsoft is proactively adjusting its computing power allocation structure: the proportion dedicated to internal research and development and Copilot has increased from about 10% to 20%. If all the newly added GPU capacity were allocated to Azure, the growth rate for Azure in the second quarter could reach nearly 40%, indicating that some growth potential is being strategically deferred.

Regarding capital expenditures, the forecast for fiscal year 2026 is $148 billion, a 400% increase compared to fiscal year 2022. The current allocation of computing power is approximately 70% for Azure and 30% for Copilot and internal R&D. The allocation ratio for newly added GPUs is more balanced, closer to 60/40. Goldman Sachs believes that Microsoft is prioritizing long-term strategic positioning over short-term Azure revenue, which will translate into a stronger competitive moat and returns in the medium term.

**The "Iceberg Model": Visible and Invisible Compute Investment**

Goldman Sachs introduced an "iceberg analogy" framework in its latest report to interpret Microsoft's current compute capital expenditure strategy. The part "above the waterline" refers to compute investments that are directly monetized and reflected in the quarterly performance of Azure and Office 365. The part "below the waterline" comprises compute deployments that do not currently generate direct revenue but are crucial for Microsoft's long-term strategy; these investments, while not reflected in short-term financials, create room for future monetization.

Microsoft management indicated that if all newly added compute capacity were allocated to Azure, the second-quarter growth rate on a constant currency basis would exceed 40%. The actual reported growth was 38%. Goldman Sachs estimates that with full allocation, Azure growth could reach the low 40% range, equivalent to an increase of about 5 percentage points. This confirms that Microsoft is strategically deferring some compute capacity for internal R&D and long-term initiatives like Copilot.

**Updated Compute Allocation: Internal Usage Share Rises Significantly**

Goldman Sachs has raised its estimate of the proportion of compute power used internally by Microsoft from about 10% to 20% in its latest report. This adjustment is based on guidance from the Q2 earnings call and recent tracking of token usage for first-party applications like Copilot and by top engineers. Microsoft has previously stated its commitment to developing AI models at the absolute forefront.

For the use of compute capital expenditure in fiscal year 2026, Goldman Sachs established a detailed framework:

* **Azure AI Compute** claims the largest share, applying a 4-5x "compute/quarterly new revenue" multiplier, meaning each dollar of capital expenditure can translate into $0.20 to $0.25 of revenue in the following quarter. This portion implies a gross margin over a 6-year lifecycle that is lower than the current core Azure level. * **Internal R&D Compute** encompasses employee AI adoption, increased developer compute demands, and the development of internal models by Microsoft AI (MAI). * **Application Compute** includes usage-based AI compute and Copilot compute that is not yet monetized (such as for Bing and free Copilot applications), applying a 3-4x multiplier.

Overall, Goldman Sachs estimates that approximately 70% of total compute power in FY2026 will be allocated to Azure, with 30% used for Copilot and internal purposes. The allocation for newly added GPUs is more balanced, closer to a 60/40 split.

**Four Drivers Behind Increased Internal Allocation**

Microsoft's compute demand is being driven upward by four structural factors, leading to a profound evolution in its capital expenditure structure.

First, **Copilot adoption is continuously climbing**. Increased reliance on Copilot by both Microsoft employees and external customers is directly driving up compute consumption.

Second, **internal R&D efforts have intensified significantly**. Microsoft clearly stated in February that it is developing AI models at the absolute forefront. Leveraging its IP access agreement with OpenAI until 2032, the company will build specialized models based on forked versions of OpenAI models, tailored to its own domain data, primarily serving enterprise-grade Copilot applications.

Third, **the operating expense structure is undergoing a fundamental shift**. Microsoft expects future OpEx growth to be increasingly driven by capital expenditure rather than headcount, as engineering workloads become more compute-intensive. Employee use of internal Copilot also requires incremental compute support. This trend is already reflected in the 5% workforce reduction in 2023 and multiple rounds of personnel adjustments in 2025.

Fourth, **the baseline compute level across the software stack is rising overall**. Customers now have a basic expectation for AI features embedded within existing applications and are less accepting of separate charging models. For vertically integrated hyperscalers like Microsoft, this change directly translates into higher compute capital expenditure; for other software companies, it manifests as token fees paid to third-party LLMs or hyperscale cloud providers.

**Pathways to Improving Azure Market Sentiment**

In future quarters, as new capacity comes online in phases, the trade-off in compute allocation between Azure growth and first-party applications/internal R&D will become less of a zero-sum game. This means Microsoft can advance Copilot and frontier model development simultaneously without sacrificing Azure growth rates.

A core obstacle for investors currently is **insufficient information disclosure**. Against the backdrop of consistently rising capital expenditure intensity and repeated revisions to Azure expectations, the market urgently needs an analytical framework that can distinguish capital expenditure for Azure from that for other businesses, enabling clearer judgment on medium-to-long-term growth and investment returns.

Goldman Sachs estimates that in the second quarter, Microsoft had approximately $4 billion in GPU expenditure not directly linked to Azure or already monetized Copilot business. If this compute power were allocated to Azure using a 5x "compute/quarterly new revenue" multiplier (implying a three-year payback period), Azure could see an upside of about 3%, corresponding to a 5 percentage point increase in growth rate, raising the reported Q2 growth to 44% (43% on a constant currency basis).

Compared to other hyperscale cloud providers, Microsoft relies more on self-developed internal software solutions, which to some extent consumes capacity that could otherwise be deployed to Azure. For instance, compute demand from increased adoption of Microsoft's AI coding tools is counted as capital expenditure, whereas if other vendors purchased third-party software, it would appear as operating expense. This structural difference also affects the market's assessment of Azure's growth potential.

Some funds have recently flowed towards Google, primarily based on three factors: Gemini's competitive positioning, the potential opportunity with TPUs, and a relatively smaller exposure to application businesses. It is noteworthy that Google's stock price fell after its Q4 2025 results on February 4, as its 2026 capital expenditure was about 50% higher than market expectations. However, the earnings call also provided a clear disclosure framework – 60% of capital expenditure is for servers, with slightly more than half of that compute power allocated to the cloud business. Goldman Sachs believes that if Microsoft could provide disclosures of a similar dimension, it would help rebuild investor confidence.

**The Strategic Rationale Behind Copilot Investment**

Microsoft is shaping Copilot into an enterprise-grade LLM solution that balances performance and cost through differentiated delivery mechanisms. Copilot 365 is priced at approximately $30 per user per month and integrates multiple models. In comparison, ChatGPT Enterprise is priced around $60, offering unlimited GPT-4 access. Individual users might still prefer the latter's simplicity, but enterprises must weigh performance against cost and evaluate the long-term strategic value offered by the abstraction layer.

In terms of contextual intelligence, **WorkIQ** is becoming an important vehicle for Microsoft to demonstrate differentiated advantages. The tool, launched at last November's Ignite conference, analyzes anonymized, aggregated Microsoft 365 data to reveal collaboration patterns and work trends within organizations, providing data support for embedding AI into business processes.

Microsoft also demonstrates a unique strategic advantage as a **"fast follower."** Leveraging its strong distribution channels, enterprise-grade security capabilities, and service support, Microsoft can rapidly introduce technologies proven effective in consumer settings into enterprise scenarios. Practices over the past three-plus years show that enterprise software stacks have long iteration cycles. Meanwhile, management's urgent demand for AI implementation is pushing enterprises to seek incremental value beyond their existing software stacks, creating an opportunity for Microsoft.

More importantly, **Copilot has begun to materially contribute to the growth of M365 Commercial Cloud revenue**, a business estimated to reach $89 billion in FY2026. Excluding the impact of one-time items, M365 Commercial Cloud has maintained stable revenue growth over the past three quarters, reversing a previous trend of sustained slowdown. Goldman Sachs believes that Copilot's monetization is effectively offsetting pressures from the nearing end of the E5 upgrade cycle and new seat additions from lower ARPU segments (such as regional markets, small and medium-sized businesses, and frontline workers). The ARPU growth rate over the past three quarters has increased to 10-11%, significantly higher than the 5-7% in the prior year, also indicating an acceleration in Copilot adoption.