Will Microsoft Cloud gross margin be at or above 65% in Q3 FY26?

Hood's Cloud GM forward guidance has landed within 50bps of guide for 4 consecutive quarters (Q3 FY25-Q2 FY26). She guided to a precise 65% for Q3 FY26, not a range — implying tight visibility. Modal outcome is 64.5-65.5%. The mechanism driving the 200bps step-down (short-lived asset mix shift to 67%, depreciation flow into COGS) is GAAP-mechanical and largely already in run-rate. Maya 200 silicon (+30% TCO) and 50% inferencing throughput gains provide partial offset. However, committee math suggests software efficiency 'currently matches not offsets' depreciation absorption, and memory pricing inflation is a live catalyst. Slight upward tilt above 50% from precision history; downward tilt from compression mechanics. Net 58%.

The committee held UNIT_ECONOMICS at PLAUSIBLE rather than FRAGILE precisely because Cloud GM is expected to land at or near the guided 65%. But the 200bps step-down exceeds the trailing ~100bps/qtr trend, and the underlying compression mechanism (short-lived asset depreciation absorption) is accelerating, not stabilizing. OpenAI 45% RPO concentration is a unit-economics black box per Atomic Auditor — if OpenAI workloads run below blended margin, print lands 50-100bps below 65%. Memory pricing inflation flagged by Hood is the operational catalyst that could push the print 50-100bps below guide if it accelerates. With trough-guide reliability at ~75-85% (15-25% miss rate by >50bps), my estimate centers around 55%.

The question is binary at the exact guided number. Given that MSFT's Q3 FY26 quarter just ended (March 31, 2026) and earnings will print in late April 2026 (roughly when this prediction is made), most of the revenue-determining factors are already realized. Hood's guide-vs-print precision argues this will land within ~50bps of 65%. Distribution roughly symmetric around 65% means ~50% probability the print is at-or-above versus below; the tilt toward YES comes from (a) revenue beat pattern lifting denominator (gross profit growth keeps pace), (b) Maya 200 first-quarter populating COGS at meaningful scale, (c) capex sequentially decreasing on finance lease timing. Tilt toward NO comes from compression mechanics being structural. Net 60%.

Three forces in tension: (1) Hood's guide precision says modal outcome is exactly 65%, (2) compression mechanics (short-lived asset mix at 67%, capex revision higher) say underlying trend is still compressing, (3) software efficiency offsets are real but committee math says they currently match not exceed depreciation absorption. The threshold is at the guide so binary is roughly 50/50 with adjustments. Adjustments: precision history (+5%), revenue beat pattern (+3%), memory pricing risk (-3%), OpenAI margin opacity (-3%). Net 56%, modest tilt YES.

Contrarian read: a 200bps step-down guide is a big move, and Hood's incentive when guiding to a trough is to underpromise — but she also has incentive to NOT widely overshoot her own commentary on stabilization. The committee math 'efficiency currently matches not offsets depreciation absorption' implies the trend through Q2 FY26 was still compressing. If you extrapolate -100bps/qtr from 67% to Q3, you land at 66% (above guide); but the 200bps step-down implies management sees an acceleration in compression that overshoots normal trend. Memory pricing flag is a real downside catalyst not yet in run-rate. Slight tilt toward 50/50, with mild lean YES.

Optimistic read leaning on Hood's track record. 4-of-4 within 50bps means base rate on hitting precise guide is high. AWS 2018-2020 and MSFT FY18-20 both rebounded from similar compression — favorable historical analog (n=2 small but directionally relevant). Maya 200 +30% TCO is the silicon-cost lever that did not exist in prior cycles. Beat pattern on revenue (4-of-4) lifts denominator and softens compression mechanically. Hood's specific 65% number (not range) signals high confidence in landing there. 62% YES.

Threshold at exact guide; Hood's precision history says modal outcome is at-or-near 65%. Committee held UNIT_ECONOMICS PLAUSIBLE meaning base case is in line with guide. Memory pricing and OpenAI margin opacity are the downside risks. Slight lean YES at 55%.

Devil's-advocate take: 200bps step-down is large, mechanism (asset mix to 67%) is accelerating, and management may be calibrating guide to a number they expect to slightly miss given the compression trajectory has accelerated quarter-over-quarter. True coin-flip at 50%.

Balanced view: Hood's guide precision (4-of-4) + Maya 200 + revenue beat pattern tilts toward YES. Compression mechanics + memory risk + OpenAI margin opacity tilt toward NO. Net modest lean YES given precision history. 58%.