Revisiting Our Microsoft Thesis After The Capex Crossed $100 Billion

In January, we modelled FY2026 capex at $80-85 billion. The current run rate implies $100-115 billion. That is a $15-30 billion upside revision on capex alone, absorbed inside the valuation framework. Consensus sell-side estimates were slower to adjust than the Capital Desk; most models still carry FY2026 capex at $90-95 billion. The reality is higher.

At the same time, Azure growth has held above 30% on a constant-currency basis for six consecutive quarters. Commercial cloud revenue (Azure plus commercial Office plus Dynamics) hit approximately $172 billion annualised at end of FY2025, up 21% year over year. That is a larger absolute revenue scale than the entire AWS business at comparable commercial-cloud definition. Microsoft is no longer the challenger to AWS; it is the larger commercial cloud franchise by revenue.

The OpenAI commercial relationship has also matured. Revenue pass-through from OpenAI-related workloads is now estimated at $15-20 billion annualised, carrying lower margins than core Azure but contributing meaningfully to revenue growth. The accounting treatment has drawn some criticism as potentially lower-quality revenue. The Capital Desk treats it as real commercial cloud revenue; the underlying compute consumption is genuine and priced at market rates. The OpenAI relationship now sits inside a broader set of AI commercial agreements with Anthropic via Azure, with Meta via cross-licensing, and with internal Microsoft products.

In late 2024 and early 2025, Microsoft's capex trajectory was the single most debated item in the mega-cap capex cycle. The stock spent most of that period range-bound in the $380-420 zone as investors worked through the implications. Since then, the stock has moved to $530 on a combination of Azure beat-and-raise results and the collective market acceptance that AI infrastructure capex is a structural feature rather than a discrete event.

That market re-framing is the single largest shift since our prior piece. The debate is no longer whether Microsoft should spend $100 billion a year on AI infrastructure. The debate is whether the return on that investment is in line with the invested capital. The Capital Desk's updated view is that the returns are currently tracking above cost of capital, based on the operating margin expansion in commercial cloud.

Microsoft operating margin reached 45.6% in FY2025, up from 41.8% in FY2023. Every percentage point of margin on the $281 billion revenue base is worth roughly $2.8 billion of operating profit. The 380 basis point expansion over two years has added $10.6 billion of operating profit. That is meaningful. The question is whether the rate of expansion can continue given the depreciation from the capex cycle is set to accelerate into FY2027-28. Operating leverage in commercial cloud is the most measurable signal that the AI investment cycle is producing commercial returns in real time.

Microsoft returned approximately $45 billion to shareholders in FY2025 through buybacks and dividends combined. That represents 63% of free cash flow, compared to 55% in FY2024 and 74% in FY2023. The capital return pace is relatively steady even as capex has stepped up. Total debt sits at $63 billion against $79 billion in cash; net cash of $16 billion is a healthy balance sheet position.

The updated Capital Desk view on Microsoft's capital allocation has three components. First, the buyback cadence is being maintained despite the capex absorption, which demonstrates internal confidence in the forward cash flow trajectory. Second, the dividend has grown at 10%-plus annually, which supports income-oriented investor demand. Third, the balance sheet flexibility for M&A remains intact, which matters given the likely continued consolidation of AI-native software companies over the next two years.

The net capital allocation story is strong. Management is funding the AI infrastructure build without impairing the return of capital programme or the M&A optionality. Few large-cap technology companies have this combination of flexibility currently. The valuation deserves a modest premium for capital allocation discipline specifically.

Commercial cloud revenue accelerated from $115 billion in FY2023 to $172 billion in FY2025. That is approximately $57 billion of incremental revenue over two years, a larger absolute revenue addition than the entire FY2025 revenue of SAP or Oracle individually. The scale of the growth is what makes the capex decision rational. Every dollar of Azure revenue converts to operating profit at approximately 35-40% operating margin at scale. Growing a business at that margin profile by $30 billion a year supports capex absorption of $20-25 billion without damaging the overall company profit trajectory.

The gross margin on commercial cloud has been stable in the 70-72% range despite the AI workload mix shift. AI training workloads have lower gross margins than traditional cloud compute because of the specialised hardware costs; but inference workloads have gross margins approaching or exceeding traditional cloud levels. The mix matters. FY2025 showed inference workloads growing faster than training workloads, which supports the structural margin profile.

Most importantly, commercial cloud customer concentration is diversifying. The top ten customers represent under 20% of commercial cloud revenue, and the largest single customer (OpenAI) is estimated at 7-8% of commercial cloud revenue. That concentration is manageable. Compare to a typical SaaS company where a single customer can represent 20%-plus of revenue. Microsoft's diversification is structurally superior.

Microsoft insider activity over the past six months has been modestly constructive. No open-market purchases of meaningful scale but steady exercise-and-hold patterns among several senior executives rather than exercise-and-sell. That pattern typically signals moderate internal confidence. Executive stock ownership levels have increased approximately 4% collectively over the past twelve months.

Notably, CEO Satya Nadella and CFO Amy Hood both continued their programmatic sale plans without acceleration or deceleration during the capex debate period of Q2 and Q3 2025. That stability is its own signal. Executives who believe the stock is overvalued typically accelerate programmatic sales; executives who believe the stock is undervalued typically pause or reduce them. The steady cadence suggests the management view is approximately aligned with the current market price.

For the Update thesis, the insider activity pattern reinforces the 'fairly valued to modestly constructive' framing. If insider behaviour shifts materially in either direction over the next two quarters, the Capital Desk will revisit the position. Our internal model has been repaired with the higher capex assumption and the resulting fair value range is what we are guiding to now.

The biggest accounting consequence of the capex cycle is the depreciation wave that will absorb operating margin into FY2027-28. Capital assets placed in service in FY2024-25 will depreciate through FY2028-31 depending on useful life assumptions (servers depreciate faster than building infrastructure). The Capital Desk estimates depreciation expense will grow from approximately $20 billion in FY2025 to $45-55 billion by FY2028.

That is a $25-35 billion expense headwind across three years. Operating margin compression from depreciation alone could reach 8-12 percentage points absent offsetting revenue growth. The offsetting factor is revenue growth; if commercial cloud revenue continues compounding at 18-22%, the incremental gross profit covers the depreciation headwind and still produces modest operating margin expansion.

This is the specific margin math the market has not fully priced. Bulls focus on revenue growth. Bears focus on depreciation wave. The Capital Desk's view is that both perspectives are partially correct; the net is modestly positive for operating margin but creates quarterly volatility that sell-side models may struggle to predict accurately. Expect larger-than-usual EPS beat-and-miss variance through FY2027.

The January piece argued that the capex cycle would continue growing through FY2026 before normalising. That has proven correct on the first half; the normalisation timeline is now less clear. Our prior $450-490 fair value range anchored to consensus FY2026 free cash flow of $78 billion. The updated consensus is closer to $70 billion on higher capex. That reset compresses our implied fair value on pure cash flow yield mechanics.

Offsetting that compression is the operating margin expansion and revenue growth outperformance. Commercial cloud revenue has grown at 21% versus our prior 18% expectation. Operating margin has expanded 80 basis points faster than our prior model. Those factors push fair value higher. Net, the revised range of $520-580 captures both revisions.

What the prior piece got wrong was the speed of the capex escalation. We modelled a gradual ramp to $85 billion by FY2027. The actual trajectory has been steeper. That miss was real and the Capital Desk owes the correction. The model has been updated.

In the prior piece we compared Microsoft's AI capex cycle to the 2015-2019 Azure infrastructure build. That comparison still holds. In that cycle, Microsoft's capex grew from $5.9 billion to $13.9 billion while free cash flow expanded from $25.2 billion to $38.3 billion. The investment cycle produced a return profile that justified multiple expansion and share price compounding above 25% annually across the window.

The current cycle is larger in absolute magnitude but similar in structure. AI infrastructure capex is the 2015-era Azure capex equivalent. The harvest phase is still coming. The Capital Desk maintains that the harvest phase will arrive sometime in FY2027-28, with free cash flow expanding to the $95-110 billion range as capex normalises to 25-28% of revenue from the current 23% trajectory peak.

Historical parallel precision matters here. The 2015-2019 Azure cycle produced operating margin expansion of roughly 900 basis points across the window. The current AI cycle has produced roughly 400 basis points in two years. The remaining expansion, if the parallel holds, would bring Microsoft's operating margin toward 48-49% by FY2028. That is a meaningful valuation lever.