Under the "internal conflict" of AI logic, Microsoft (MSFT.US) has fallen into a "golden pit"

Vivek Arya, a senior analyst at Bank of America, recently described the recent sell-off in the software sector as a “logical contradiction”: on one hand, hyperscale cloud providers and AI computing companies are being sold off due to massive capital expenditures (market believes investments won’t yield returns); on the other hand, the market claims AI will disrupt SaaS and replace human labor.

According to normal logic, these two views cannot both be correct at the same time. If AI is the future, then companies investing heavily in computing power are the right choice, and investors should buy hardware stocks and sell SaaS; if AI is not as important as market expectations suggest, then they should sell hardware stocks and buy traditional SaaS.

However, both types of stocks are currently plunging simultaneously, which seems to only indicate that the market doesn’t understand AI and is unwilling to delve into the logic, merely reacting blindly.

As a key player in the current AI revolution sweeping the global tech industry, Microsoft (MSFT.US), one of the “Big Seven” that investors have always focused on, has become a typical example of market mispricing. Microsoft’s stock price is being sold off due to massive capital expenditures, implying investors don’t believe AI computing infrastructure is justified; yet, they are also worried that Microsoft’s software business will be disrupted by AI, believing that after AI replaces human labor, companies will no longer need to purchase Microsoft’s “seat accounts.” Even non-tech experts can see how absurd this logic is.

Clearly, the current market valuation of Microsoft has become seriously disconnected from its fundamentals. However, this irrational volatility presents a rare entry opportunity for investors. Regardless of which side’s judgment proves correct in the end, Microsoft will remain resilient.

Positive and Negative Catalysts

The biggest concern about Microsoft currently stems from its quarterly capital expenditure of $37.5 billion and plans to continue increasing investments. Investors generally worry that the scale of cloud providers may lead to oversupply of computing capacity, and that Microsoft’s capital spending and Azure revenue growth are not clearly correlated, making returns on investments uncertain.

But these concerns fundamentally stem from a misunderstanding of the cycle of infrastructure buildout. According to the latest earnings call, about two-thirds of the current quarter’s capital expenditure was directed toward short-cycle assets like GPUs and CPUs, primarily to meet the ongoing demand from customers. The company explicitly stated that Azure’s customer demand consistently exceeds supply, requiring continuous investment to balance new capacity with the rapid growth of Azure, the computing needs of Copilot series products, R&D innovation, and replacement of aging servers and network equipment.

More importantly, Microsoft’s capital expenditure isn’t just about purchasing chips from Nvidia (NVDA.US) and AMD (AMD.US); it is accelerating vertical integration to build its own computing moat. During the Q2 earnings call, CEO Satya Nadella repeatedly emphasized the deployment of the self-developed Maia200 AI accelerator.

Nadella said: “At the chip level, we have Nvidia, AMD, and our own Maia chips, which deliver optimal overall cluster performance, cost, and supply across multiple hardware generations. This week, we launched Maia200 accelerators, with FP4 precision compute power exceeding 10 petaFLOPS, and a total cost of ownership (TCO) optimized by over 30% compared to the latest hardware. We will start large-scale deployment with inference and synthetic data generation for the superintelligent team, as well as inference services for Copilot and Foundry.”

For non-technical investors, technical metrics like FP4 precision of 10 petaFLOPS may seem meaningless, but a 30% reduction in TCO is a tangible and significant positive. Microsoft is also simultaneously advancing its self-developed Cobalt200 CPU, which improves performance by over 50% compared to the previous generation. This is a key positive: the company is gradually reducing dependence on Nvidia, avoiding paying premiums for all hardware, and in the long run, profit margins will recover—one of the core reasons why bearish sentiment on Microsoft exists.

This quarter, Microsoft’s gross margin fell to 68%, seemingly under pressure, but the main reason is that the company is in an investment phase: high-cost assets are being amortized as they go live. As the company gradually replaces Nvidia and AMD chips with its own Maia and Cobalt chips, the cost per inference (per user query to Copilot) will significantly decrease, driving profit margins to rebound. As the penetration of self-developed chips increases, efficiency gains will outpace depreciation, supporting management’s expectation of a slight increase in operating profit margin for 2026.

Beyond capital expenditures, OpenAI is also a major point of divergence for the market.

Recently, Microsoft recorded a $10 billion GAAP gain from its restructuring of OpenAI investments, but this is just an accounting treatment; the actual operational logic is more complex. About 45% of Microsoft’s remaining performance obligations (RPO) come from selling compute power to OpenAI.

On one hand, this is positive: OpenAI is a super-large customer, and its recent financing reduces default risk; but on the other hand, it introduces significant customer concentration risk—Microsoft’s AI infrastructure buildout is highly tied to OpenAI’s success, and how OpenAI will realize profits and cover its huge investments remains uncertain.

Recent news about Anthropic complicates the situation: Anthropic lost a government contract, which was taken over by OpenAI. Logically, as OpenAI’s cloud provider, Microsoft should benefit, but the market reacted negatively. The reason is that if OpenAI continues to grow, Microsoft only benefits from infrastructure, but could this weaken its long-term competitive advantage in software?

There are even concerns that Microsoft’s own Copilot tools are increasingly integrating third-party models. If these are seen merely as external technology gateways rather than truly innovative products, Microsoft might lose its competitive edge in software.

But it must be recognized that Microsoft’s core business is never driven by individual consumers but by enterprise clients. As long as enterprises continue to choose Microsoft’s products and services, the company’s fundamentals will not fundamentally weaken.

Unreplaceable Core Business: Microsoft’s Perfect Hedge

In fact, investors shouldn’t be bearish on AI, especially for giants with their own cloud services. “Overcapacity” in computing power shouldn’t be a reason for concern. Even without selling to AI customers, computing resources can be used in countless scenarios or even for internal use. Microsoft is adding 1 gigawatt of capacity in three months, building “AI super-factories” in Atlanta, Wisconsin, and other locations, connecting multiple data centers via high-speed networks to form a massive, organic computing ecosystem.

Currently, Microsoft Copilot has 15 million paid seats, with a growth rate of 160%—though still a small fraction of total users, the growth momentum is strong. Moreover, like the other Big Seven tech giants, Microsoft’s computing capacity remains in high demand, requiring continuous expansion.

Using Bank of America’s “logical contradiction” analysis, Microsoft is actually a highly valuable strategic asset and even a hedge: if AI succeeds and replaces human labor, Microsoft will be a winner because it controls Azure’s infrastructure and has enterprise-grade intelligent tools to replace human work; if AI only modestly improves efficiency, Microsoft will still be a winner because it has 450 million Microsoft 365 commercial seats, with high switching costs, ensuring its core business remains stable.

The current market’s sell-off due to high capital expenditure essentially offers investors a great opportunity: to buy the world’s leading AI infrastructure assets at a pessimistic valuation of “software business about to disappear,” while in reality, Microsoft’s software business shows no signs of decline.

Valuation Analysis

Market excessive pessimism has pushed Microsoft’s valuation into an attractive range.

First, look at the company’s fundamentals: Microsoft’s remaining performance obligations (RPO) total $625 billion, up 110% year-over-year. This alone demonstrates Microsoft’s core value and the necessity of its compute infrastructure. About $281 billion of this comes from OpenAI, meaning over $340 billion comes from diversified long-term contracts, indicating a healthy structure.

On the valuation side, Microsoft’s current forward P/E ratio is only 25.53, well below its long-term average of 33.21, representing a low point since 2022. The last time such a valuation level appeared, it didn’t last long.

In short, the market is currently pricing in extreme skepticism about Microsoft, leaving no room for its operational stability, long-term AI industry potential, or the efficiency gains from self-developed chips and dense data centers. Once the scale of self-developed chips drives costs down and profit margins recover, the current price will prove to be a significant undervaluation.

Conclusion

The market currently views Microsoft’s massive capital expenditures as a burden, but data shows these investments will ultimately build a defensive moat. Microsoft’s aggressive deployment of self-developed chips and infrastructure is essentially locking in future compute costs for the next 5-10 years. With a P/E of 25.53, it completely ignores a key fact: Microsoft’s compute capacity is in a supply-constrained state.

There will always be a side of the market that misjudges the future of the software industry, but based on Azure backlog and historical return on invested capital, the misjudged party will never be Microsoft.

Of course, risks remain—for example, even after raising funds, OpenAI’s development still faces significant uncertainty; slower-than-expected commercialization of Copilot; potential slowdown or reversal in user adoption; revenue growth slowing from 17% this quarter to 15-16%, which could trigger negative market reactions; and ongoing industry competition from Google (GOOGL.US), Amazon (AMZN.US), and others.

But even considering these risks, at the current valuation level, Microsoft’s risk-reward ratio remains highly attractive.