How will Microsoft's (MSFT.US) AI investment returns be realized? Morgan Stanley reviews the Ignite conference: Flexibly allocating GPU computing power and betting on end-to-end enterprise AI commercialization
Morgan Stanley released a research report, reiterating an "Overweight" rating on Microsoft with a target price of $650.
According to Wall Street Insights, Morgan Stanley has issued a research report reiterating its “Overweight” rating on Microsoft (MSFT.US), with a target price of $650. Led by Adam Wood, the Morgan Stanley analyst team stated that after attending Microsoft's Ignite conference in San Francisco and speaking with CEO Satya Nadella and CFO Amy Hood, the bank believes Microsoft is among the few software companies already realizing returns on AI investments.
Computing Power Strategy: Avoid Blindly Maximizing Long-term GPU Contracts and Retain Capacity for Diverse Services
The report notes that when it comes to allocating computing power, Microsoft management views GPUs as part of a diversified asset portfolio and avoids simply signing long-term GPU-as-a-Service (GPUaaS) contracts. Management stated that the company actively reserves computing resources to supply its own first-party products, its Foundry platform, and external enterprise customers, rather than locking all resources into long-term agreements. From Microsoft's perspective, Model Labs are just important customers—they do not represent the entirety of the cloud business. Retaining flexible computing power is an active investment in the value of its platforms and applications, aiming to capture high-value-added service monetization opportunities on top of its infrastructure.
Addressing market interest in the new "Neocloud" architecture strategy, Microsoft says the underlying strategy remains unchanged: the top priority is to build its own data centers, followed by long-term leasing arrangements where Microsoft retains control over data center layout, networking, and hardware. This approach allows better management of data center operations, deployment pace, and capital expenditure on equipment. The cancellation of some third-party cooperation orders stemmed from insufficient vendor service delivery capabilities, not a shift in Microsoft’s overall computing procurement strategy; nor does it suggest a significant change in the market's supply-demand dynamics for computing power.
In addition, Microsoft’s improvements in computing efficiency are not limited to the GPU layer. CPU utilization and storage systems are also key focuses for optimization, and new infrastructure deployments incorporate these efficiency strategies—thus, efficiency gains are not limited to existing server clusters. However, the bank cautions that efficiency gains cannot be seen as an additional upward variable for computing planning alone; instead, efficiency improvements can simultaneously boost revenue capacity while lowering delivery costs.
Profitability and Gross Margin: Product-Level Improvement is the Priority; Cloud Business Structure May Temporarily Pressure Aggregate Gross Margin
Morgan Stanley points out a critical distinction that may be confusing for the market: Microsoft management’s core goal is to improve the standalone gross margins of Azure and M365 products. The reported aggregate Microsoft Cloud gross margin is simply an outcome of business dynamics, not a direct management metric.
Even if the profitability of individual products continues to improve, the overall cloud gross margin may be diluted if the proportion of fast-growing but lower gross margin Azure business continues to expand. Investors should not simply assume that pressure on aggregate gross margin reflects weaker profitability for cloud services. In future earnings assessments, attention should be focused on resource utilization, infrastructure efficiency, model operating costs, and the scale of high-value platform services consumption.
Enterprise AI Flywheel: Positioned as a Multi-Model Coordination Layer, Not Limited to Proprietary Large Models, and Microsoft IQ Closes the Capability Loop
Microsoft does not base its competitive edge on any single model. The core value for enterprise clients lies in the comprehensive ecosystem around the models: the context of enterprise operations, orchestration frameworks, memory capabilities, and secure, trusted management. Enterprises can use cutting-edge large models, Microsoft’s proprietary specialized MAI models, or open-source models. The Microsoft platform is responsible for selecting the most cost-effective model according to task requirements, while leaving control over commercial data to clients and managing the security, permissions, and costs of all AI agents in a unified way.
At this conference, Microsoft IQ was unveiled, integrating Work IQ, Fabric IQ, Foundry IQ, and Web IQ to further enable end-to-end enterprise AI flywheel, tightly connecting enterprise business data, model orchestration, and application deployment. Morgan Stanley believes that the more models available on the market, the higher Microsoft’s strategic value as the platform connecting models with enterprise workflows.
Copilot Commercialization: Dual-Engine of Subscriptions and Per-Use Billing, Early-Stage Penetration in Non-Code Scenarios
Microsoft has adopted a hybrid business model combining subscriptions with usage-based billing: subscriptions provide stable base revenue, while usage billing follows customers' real AI workloads for incremental gains. Management disclosed that some large customers have Copilot Cowork expenditures exceeding the sum of all other Microsoft products, but the CFO team emphasized that these cases are outliers and do not represent the broader customer base. The overall commercialization of the product suite is still recent, and reliable expectations for penetration rhythms cannot yet be established.
Currently, Copilot adoption in code-related workflows is strong, but consumption in broader knowledge work scenarios is still in the cultivation phase. The growth potential is significant, but rollout pace remains a key uncertainty.
Management compared this to the early development of Azure: as enterprises embrace AI agents, they are particularly focused on cost control and expenditure visibility. Microsoft has launched the Agent 365 business analysis and cost management tools, binding AI spending with business outcomes and budget alert capabilities to help enterprises manage variable AI usage and clear obstacles to large-scale deployment. However, the direct revenue from these tools depends on the pace at which enterprises scale up production-level AI workloads.
Proprietary Chip Strategy: Maia and Cobalt as Efficiency Supplements, Not a Complete Migration to Proprietary Hardware
Microsoft’s custom chips are only one element of a diversified computing fleet and are not intended to migrate all workloads to proprietary hardware. For example, the Maia AI chip and the newly released Cobalt 200 are aimed at AI agent workloads, with Cobalt 200 delivering up to 50% performance improvement over the previous Cobalt 100. Microsoft will continue to procure chips from external vendors; the goal of proprietary chips is to optimize performance and costs in specific scenarios, supplementing overall computing efficiency rather than serving as an exclusive replacement. This continues the earlier approach of computing chips evolving from sole Intel (INTC.US) reliance to a diversified portfolio including AMD (AMD.US) and Cobalt chips.
Disclaimer: The content of this article solely reflects the author's opinion and does not represent the platform in any capacity. This article is not intended to serve as a reference for making investment decisions.
You may also like
Anthropic triggers its own safety alert before IPO: AI progress exceeds expectations
Anthropic disclosed in its IPO filing that the rapid pace of AI development has triggered additional safety mechanisms within the company. Evaluation agency METR found that Anthropic's technological progress over the past year is equivalent to that of the previous two years, and AI's involvement in its own research and development is accelerating this process. The risk of losing control due to recursive self-improvement has also raised concerns in the field of AI safety.

NetEase Q3 revenue down 1.8% year-on-year, gross margin rises to 68.3%, gaming business profitability improves | Earnings Report Insights
NetEase's revenue in the third quarter saw a slight decline, but profitability continued to improve. The gaming business remained the core pillar, with gross profit up 5.4% year-on-year. The innovative and other businesses were the only segment to achieve year-on-year growth. Meanwhile, the company's net cash balance increased to 175.2 billion yuan, and it raised its quarterly dividend and extended its stock repurchase program, further enhancing shareholder returns.
DAO.US Q3 2026 net income reached 1.43 billion yuan, with positive operating profits for nine consecutive quarters, AI commercialization advancing in depth
On November 19, NetEase Youdao (DAO.US) announced its unaudited financial results for the third quarter of 2026.
Forex Today: Hawkish repricing of Fed policy outlook continues to support US Dollar
