The Zhitong Finance App learned that Morgan Stanley published a research report reaffirming the “gain” rating of Microsoft (MSFT.US), with a target price of 650 US dollars. A team of analysts from Daimo, led by Adam Wood, said that after Microsoft held the Ignite conference in San Francisco and exchanged views with CEO Nadella and CFO Amy Hood, the bank believes Microsoft is one of the few software companies that have realized the return on AI investment.
Computing power strategy: Don't blindly maximize long-term GPU contracts and preserve computing power for diverse businesses
Damo pointed out in the report that when it comes to allocating computing power resources, Microsoft management regards GPUs as combined assets for allocation and refuses to simply pursue signing long-term GPU-as-a-Service (GPaaS) contracts. Management said that the company will actively reserve computing power resources and supply its own first-party products, Foundry platform, and external enterprise customers, rather than locking all computing power resources into long-term contracts. From Microsoft's perspective, Model Lab is only an important customer and does not represent the entirety of the cloud business. Preserving flexible computing power is an active investment in the value of its own platform and application, with the intention of seizing monetization opportunities for high value-added services on top of infrastructure.
In response to the Neocloud (Neocloud) architecture strategy that the market is concerned about, Microsoft stated that the underlying strategy has not changed. The priority for construction is still to build its own data center, followed by the long-term leasing model where Microsoft takes control of the computer room decoration, network, and hardware equipment. This model can better control the operating efficiency, deployment pace, and equipment capital expenses of data centers. The root cause of the cancellation of some third party cooperation orders is due to the supplier's insufficient ability to fulfill the service contract. It is not a shift in Microsoft's overall computing power procurement strategy, nor does it represent a major change in the market computing power supply and demand pattern.
At the same time, improving Microsoft's computing power efficiency is not limited to the GPU level. CPU utilization and storage systems are also the focus of efficiency optimization, and the newly launched infrastructure has also inherited this optimization plan, so the efficiency dividend is not limited to existing server clusters. However, the bank cautioned that efficiency improvement cannot be used alone as an additional upward variable in computing power planning; efficiency improvements can simultaneously drive an increase in revenue carrying capacity and reduce delivery costs.
Profit and gross profit margin: Product-level improvements are prioritized, and the cloud business structure will gradually suppress the overall gross profit margin
Damo specifically pointed out a key point that is easily confused by the market: the core goal of Microsoft management is to increase the gross profit margin of Azure and M365 products. The comprehensive gross margin of Microsoft Cloud disclosed in financial reports is only the result of business development, not a direct assessment target of management.
Even if the profit level of individual products continues to improve, once the share of the Azure business with high growth but relatively low gross margin continues to expand, it will still dilute the Microsoft Cloud consolidated gross profit margin. Investors can't see the overall gross margin under pressure; they simply judge that the profitability of various cloud services themselves is weakening. Subsequent evaluation of profit improvements should focus on tracking resource utilization, infrastructure efficiency, model operation costs, and the consumption scale of high-value platform services.
Enterprise AI Flywheel: Positioning multi-model orchestration layers, not limited to self-developing large models, Microsoft IQ can break through a closed loop of capabilities
Microsoft is not betting on a single model itself. The core value of enterprise customers comes from the entire system surrounding the model: enterprise business context, model scheduling framework, memory capacity, and secure and trustworthy control. Enterprises can mix and use cutting-edge models, Microsoft's self-developed specialized MAI model, and open source weighting models. The Microsoft platform is responsible for selecting the most cost-effective model to complete the work according to task requirements. At the same time, it leaves control of the enterprise's commercial data to the customer, and uniformly controls the security, permissions, and expenses of all AI agents.
Microsoft IQ unveiled at this conference integrates Work IQ, Fabric IQ, Foundry IQ, and Web IQ to further open up the full link of the enterprise AI flywheel and closely integrate enterprise business data, model orchestration, and application distribution. Damo believes that the more models available on the market, the higher the strategic value of Microsoft as a platform layer connecting models to enterprise business workflows.
Copilot commercialization: subscription + pay-as-you-go two-wheel drive, non-code scenario penetration is still early
Microsoft has adopted a hybrid business model where subscription is combined with usage billing: subscriptions bring stable basic revenue, and usage billing increases incrementally according to the customer's actual AI workload. Management revealed that large customers have already spent more on Copilot Cowork products than Microsoft's other products combined, but the CFO team emphasized that this case was an extreme sample and did not represent a broad customer base. The commercialization time for the complete product was still short, and it was not possible to establish a reliable penetration rate.
Currently, Copilot's acceptance of coding-related workflows is strong, but consumption for a wider range of knowledge office scenarios is still being nurtured, and the growth potential is impressive, but the pace of implementation is still an important uncertain variable.
Management is an analogy to the early development of Azure. When enterprises embrace AI agents, their demand for controllable costs and observable expenses is very prominent. Microsoft launched Agent 365 business analysis and cost management tools to bind AI expenses to business implementation results and budget alerting capabilities to help enterprises control fluctuating AI usage and remove barriers to large-scale deployment, but the direct revenue brought by these tools depends on how quickly the enterprise's production-level AI workload expands.
Self-developed chip strategy: Maia and Cobalt complement efficiency, not all workloads migrated to self-developed hardware
Microsoft's custom chips are part of a diversified computing power fleet, and it is not necessary to migrate the entire business to self-developed hardware. Taking the Maia AI chip and the newly released Cobalt 200 as an example, the Cobalt 200 targets intelligent AI workloads, and the performance is up to 50% higher than the previous generation Cobalt 100. Microsoft will continue to purchase chips from external suppliers. The goal of self-developed chips is to improve performance and cost in specific scenarios, as a supplementary gripper for overall computing power efficiency rather than an exclusive alternative, continuing the idea that computing chips evolved from a single Intel (INTC.US) to AMD (AMD.US) and Cobalt in multiple ways in the early years.