I bought MSFT shares at an average price of $370 during the hyperscaler dip last week, and here are my thoughts.
The AI Trade Stack
We are really facing a strong sectoral narrative here. Brief intro to the current state of the AI trade: the trade includes the bottom layer of semis and the semi supply chain, which is where the largest economic value is accruing at this point (SK Hynix, Samsung, Micron, Broadcom, NVDA, AMD, etc.). Memory is now getting hotter than companies on the GPU chain. On the second layer, we have hyperscalers converting semis into tokens through their data center buildout (MSFT, GOOGL, AMZN, META, etc.). Then, you have the corporate applications and software — NOW, CRM, Workday — and also OpenAI and Anthropic.
Last week's dip reflects increasing concern about the CapEx buildout by hyperscalers and reconfirmed confidence in semi names, particularly the momentum in memory. Major hyperscalers have been doubling down on the buildout budget and thus the financing activities: Google issued an $80B equity offering, SpaceX had a massive IPO, and right after, made its first $40B investment-grade debt issuance. The financing is necessary to support a multi-year buildout cycle and reflects management's view that the CapEx will continue and may not fully translate to the top line in the next 2–3 years. On the other hand, Micron's recent earnings showed strong booking momentum, continuing to support its high multiple.
All hyperscaler stocks have been dipping: MSFT is getting hit the hardest, while GOOGL is in a relatively better position. On one side, I think the bias reflects persistent bearish sentiment on the SaaS sector. On the other hand, on the cloud computing front, Google is viewed as the leader, and MSFT's recent Azure growth acceleration lags its competitors. I think both the bias against SaaS and the concerns about CapEx spending present a great opportunity to buy a company that leads both themes at a historically low multiple.
The SaaS Business
I've been bullish on SaaS companies with a focus on enterprise consumers, proprietary data, and a clear pathway to monetize the agentic economy. In my previous write-up about ServiceNow, I defended SaaS companies against the bearish thesis of seat reduction and disruption risk in more detail, so I'll go more briefly here.
For Microsoft, seat reduction wouldn't be a primary concern, as no company will voluntarily cancel an MSFT subscription except due to layoff effects. The reason: MSFT is cheap — most enterprise subscription values range from $10 (Office 365 E1) to $57 (Microsoft 365 E5) per user per month. At the same time, MSFT suites are deeply embedded in workflow through both the dominance of Office 365 in Windows systems and increasing cross-sell activity: Microsoft Teams is gaining ground against Slack, and enterprise products like Dynamics 365 also have strong momentum.
That low SKU means any meaningful upsell with Copilot will more than compensate for seat loss from layoffs. MSFT has an upcoming subscription tier, Microsoft 365 E7, which bundles E5 capabilities with Copilot credits. The SKU is $99 — a 74% uplift from E5 — and per partner interviews conducted by JP Morgan and Wells Fargo, clients show "very strong" interest in E7, and "several mega PE firms will shift to E7 immediately after its release." That shows a strong ability to upsell, linked to the stickiness of current MSFT subscriptions. Executives do have concerns over AI disruption, so some interviewees express that they don't want to enter "multi-year binding contracts" because of the fast pace of AI development. MSFT's bundle is not presented as a standalone AI solution or a claim to occupy the center of clients' AI workflow, but as a reasonable add-on to the current system — which I think is the right framing.
The strong recent Copilot growth gives me more confidence in MSFT's ability to play out this thesis even better than some software peers. Per Wells Fargo estimates, Copilot paid seats stood at 9.2M in FY25; in FY26 Q3 we saw a 5M increase from roughly 15M to over 20M, and with that trend continuing into Q4 FY26, paid seats will reach 26M. Assuming a $30 SKU, that brings $3.8B in Copilot revenue for FY26 — a 169% increase from FY25. For context, FY25 itself saw a 241% increase from FY24. The momentum is strong on paper, but it has to be admitted that 26M paid seats is still dwarfed by MSFT's 450M+ global commercial installed base, so we will see the growth rate slow in FY27 and onward.
Comparing horizontally, MSFT's AI adoption leads across SaaS. It's hard to compare exact seat growth since different SaaS companies report differently. Take ServiceNow: CEO McDermott cited that the number of customers spending $1M or more on Now Assist grew over 130% YoY in Q1 — it's reasonable to infer that the overall growth rate is lower than 130%, otherwise McDermott would cite a prettier headline number. MSFT Copilot outpaces that. On qualitative evidence, JP Morgan interviews reveal that E7 has strong indicated demand and "all the megafund PE firms will move to E7 once the GA release drops." Setting valuation aside, MSFT's Copilot adoption is among the strongest in enterprise SaaS.
On the terminal value side, the unit economics work the same way as I argued with ServiceNow. As companies lay off labor in the agentic economy, they need to compensate with more token usage. Current software takes less than 10% of a seat's total budget, while most flows to labor — and MSFT takes a portion of that 10%. Say 50% of the displaced labor budget is redirected into token spending for agents: MSFT's gross profit per seat rises materially. If anything, MSFT has the IP of OpenAI's models through their prior deal and greater scale, so the gross margin erosion from inference cost should be lower compared to other enterprise SaaS. Besides, MSFT is likely taking a larger slice of the cake over time than they do today.
In discussing monetization of the agentic replacement, the value accrues to the so-called orchestration layer — who controls the agents, who runs them. MSFT is well-positioned for that. Per UBS and JP Morgan interviews, the orchestration layer will likely not be a winner-take-all story, and corporates prefer a multi-vendor system where "general vendors" like Google or MSFT sit at the center. MSFT is actively building for that center: they have developed an ecosystem across the full agentic pipeline — Chatbot (Copilot), Build (Copilot Studio), Govern (Agent 365), Storage and Connection (Azure). Microsoft's model was previously locked to OpenAI, but they are now starting to diversify to Anthropic and even cheaper, more efficient DeepSeek models — which I think is a positive signal, as it makes MSFT more like a platform business and makes the terminal value upside even more pronounced.
The Data Center Business
From recent quarterly reports, Azure continues its quality growth, and a slight lag in acceleration does not change its competitive position meaningfully. Azure is growing at 39% YoY, accelerating 1% against a tough comp. The lag versus GCP and AWS is partially attributable to that tough prior-year comparison.
The growth is largely capped by supply, not demand. Semianalysis's semiconductor rental dashboard shows that on-demand GPU rental capacity is sold out across all types in the first half of 2026. H100 rental pricing rose from a low of $1.70/hr last October to $2.65 in May — a 55% increase. Both unit pricing and capacity clearly reflect supply-constrained demand, and MSFT expects those constraints to persist through at least the latter half of 2026. So we should see similar growth or modest upside in the next several quarters.
Further out, there is genuine uncertainty. It's not yet clear whether compute demand will match the growth of supply. MSFT capacity is expected to expand at roughly 5GW per year from 2027 onward, compared to 2GW and 4GW in 2025 and 2026 — roughly linear growth. The bull narrative is that token usage follows a parabolic path as enterprise moves from chatbots to agents, keeping utilization high. Under a softer scenario, if enterprise demand doesn't materialize in the next 1–2 years, hyperscalers can pull back CapEx. The worst case is cash keeps getting burned with the AI promise unrealized. That uncertainty is clearly priced into the current multiple — but I think, with willingness to sit with that uncertainty, MSFT is still attractive, because its computing business and its SaaS business reinforce each other. In the upside case, the justification for CapEx rests on enterprise application adoption, and MSFT sits at the center of the enterprise application layer. The software business is not just a legacy cash cow — it is an integral part of the bull thesis.
Valuation
I ran a DCF to assess returns under different CapEx scenarios rather than to produce a precise price target. In the bull case, I assume Azure maintains its current expansion pace and peaks in 2027–2028 with about 2 more points of acceleration, translating to ~30% Intelligent Cloud segment growth. The recently announced model diversification — bringing DeepSeek and Anthropic models to Azure AI — also acts as a tailwind. That generates a target of roughly $600 in the bull case and $450 in the reduced-CapEx case. Bear case is $312. The risk-return is attractive in my view.
On multiples: MSFT is at its cheapest in P/E terms, and also cheap on unlevered FCF. Investors are pricing the stock as if FCF isn't compressed by near-term CapEx, which understates the normalization upside. Currently, MSFT trades at 21.57× Price/NTM EPS, 23.75× Price/LTM EPS, and 14.48× Price/NTM Cash Flow. FY27 consensus EPS is $19.37; at a target 25× P/E, that yields $484. DCF-derived base case is $512.
Risks
CapEx downside: Customer concentration and reliance on NVDA make MSFT more exposed to the downside of the CapEx buildout than it appears.
AI adoption: Corporate adoption slows, or competition in the application layer from other AI companies intensifies faster than MSFT can capture share.