Earnings Call Highlights: Microsoft (MSFT) | Q1 FY2027 — Azure Up 48%, Cloud Accelerates to 30%, and Capex Comes In Light Because Third-Party Capacity Slipped, Not Demand
1. Executive Takeaway
First, capex came in below expectations, but this is a delivery-timing issue, not a demand issue. Capex including finance leases was $45.2B, of which finance leases were $7.4B and cash paid for PP&E was $39.37B. CFO Amy Hood attributed the shortfall directly to delayed third-party capacity deliveries and moved that capacity into the second half of the fiscal year. She also stressed twice that demand still exceeds supply. In other words, the market's biggest fear, a hyperscaler hitting the brakes, did not happen. What happened is that suppliers could not keep up.
Second, a large part of Azure's 48% growth came from software squeezing out efficiency, not from new hardware. Management's explanation was blunt: engineering teams hunted for efficiency at every layer of the platform. As GPU fleet efficiency improved, the freed-up compute went to Foundry first, while the CPU fleet found and sold capacity that had been sitting idle. Token pricing on some high-quality models also came in above expectations. That explains how capex could miss while Azure beat.
Third, the business model is expanding from selling seats to selling usage. The number of Copilot customers consuming credits through usage-based billing grew more than 75% in the quarter, and agent-opened PRs now account for over a third of all PRs on GitHub. AI inference demand is starting to tie into business-unit budgets rather than just an annual IT license purchase, which is the key to whether inference compute demand can keep compounding.
2. Revenue and Financials
This is Microsoft's first quarter under its new segment structure: the three former segments were folded into Agents and Infra (apps and agents, the multi-model system, and Azure infrastructure) and Devices and Consumer (consumer offerings), with prior periods recast.
Metric (GAAP) | Q1 FY2027 | YoY | QoQ |
Revenue | $92.28B | +18.8% | +2.5% |
Gross margin | 67.8% | -1.2 pts | — |
Operating income | $45.92B | +21.0% | +13.1% |
Operating margin | 49.8% | Up slightly | — |
Net income | $36.63B | +32.0% | +2.4% |
Diluted EPS | $4.92 | +32.3% | — |
What the numbers actually say:
GAAP net income rose 32%, but only 18% excluding the OpenAI investment impact. A year ago Microsoft booked a $3.1B loss on its OpenAI investment; this quarter it was a $207M gain, and that base effect inflates GAAP growth. Excluding it, net income was $36.42B and EPS $4.89, both up 18%, which is the true pace of the core business.
Gross margin fell 1.2 points year over year, mainly because the revenue mix shifted toward Azure. Microsoft Cloud gross margin slipped to 66%. Management says product-level margins are actually improving, mostly in Azure, but Azure's rising share more than offsets that.
An operating margin near 50% comes from cost discipline. Operating expenses grew only 6% against 19% revenue growth, showing strong operating leverage.
Segments and product lines:
Item | Q1 FY2027 | YoY |
Agents and Infra revenue | $76.80B | +25% |
Agents and Infra operating income | $40.56B | +24% |
Devices and Consumer revenue | $15.48B | -3% |
Microsoft Cloud revenue | $63.9B | +30% (+31% in constant currency) |
Azure | $33.17B | +48% |
Microsoft 365 cloud | $27.78B | +17% |
Productivity and server licensing | $8.41B | -3% |
Windows OEM and devices | $3.55B | -22% |
▲ Microsoft Q1 FY2027 earnings card: revenue $92.28B, gross margin 67.8%, operating margin 49.8%, EPS $4.92, Q2 revenue guidance $96.0B–$97.1B (Image source: Simple Tech Trend original; data from SEC 10-Q and company earnings release)
Cash flow and capex: operating cash flow was $58.0B, up 29%; free cash flow was $18.6B, held down by capex. Roughly three quarters of capex went to short-lived assets, mainly CPUs and GPUs including higher memory and storage costs, with the rest on long-lived assets such as data center buildings and land. Commercial remaining performance obligation (RPO) rose 73% to $677B. About 30% will be recognized in the next 12 months (up 29%), the portion beyond 12 months grew 98%, and the weighted average duration is 2.3 years.
3. Technology and Business Highlights
Azure: efficiency is capacity
The most interesting thing this quarter is not the 48% itself but how it was produced. Management describes a single fungible fleet: data center design, silicon, systems, software and models are optimized together, then capacity is allocated dynamically across compute, data, Foundry and Copilot based on demand signals.
Concrete results:
Microsoft added 24 new data centers across five continents this quarter.
Dock-to-live time for new GB300 racks in its biggest regions fell about 20% quarter over quarter; GB300 makes up the majority of current GPU deployments.
Throughput for Copilot workloads rose more than 50% quarter over quarter.
The first NVIDIA Vera Rubin GPUs have been docked.
Squeeze twice the throughput out of the same GPU and you need half as many new racks. That is why capex was low and Azure was high this quarter.
Custom silicon: Maia and Cobalt
Maia 200 delivers up to 1.7x more tokens per watt across AI experiences in Teams and Copilot.
Maia 300 is claimed to deliver 40% more tokens per watt than leading GPU-based systems. We broke down its system specs and open-spec strategy in Microsoft at OCP 2026: Maia 300 and Frontier AI Infrastructure.
Cobalt CPU keeps gaining adoption for general-purpose cloud workloads, and the CPU fleet also monetized idle capacity this quarter.
Foundry and multi-model: an amplifier for inference demand
Foundry now has more than 120,000 customers and revenue more than tripled year over year. Customers building agents with multiple models grew 5x, and revenue from open-source models on Foundry grew 10x. Satya Nadella's framing: customers will move back and forth between frontier models, open weights and Microsoft's own MAI models, and Microsoft's job is to keep them from being locked into any single model.
For inference compute, the implication is that the more diverse the model choices, the more fragmented the workloads, and the more you need a large compute pool that can be reallocated quickly. That is exactly where a hyperscaler beats a single-model provider.
Copilot: from seat licenses to usage-based billing
Copilot daily users rose nearly 50% quarter over quarter and more than 5x year over year.
Customers with more than 50,000 Copilot seats grew more than 8x; Volkswagen announced a rollout to more than 330,000 employees.
Agent-generated PRs on GitHub grew 6x over the past six months and now make up more than a third of all PRs; GitHub Copilot Fusion offers multi-model orchestration with near-frontier quality, up to 40% faster and up to 77% cheaper.
There was noise too: the CFO acknowledged weaker sales execution in M365 E3 and standalone SKUs, and paid M365 commercial seats grew just 5%. Slower seat growth carried by ARPU growth is a structural shift worth tracking.
4. Management Outlook
Q2 FY2027 guidance (in US dollars unless noted; management expects no meaningful FX impact):
Item | Q2 FY2027 guidance |
Total revenue | $96.0B–$97.1B (+18–19% YoY) |
Agents and Infra revenue | $80.1B–$80.7B (+24–25% YoY) |
Azure revenue growth | 49–50% (constant currency) |
M365 commercial cloud growth | About 19% (constant currency) |
Microsoft Cloud gross margin | About 65% (down YoY) |
Devices and Consumer revenue | $15.9B–$16.4B |
Windows OEM and devices | Down in the mid-30s % |
COGS | $32.7B–$32.9B (+26–27% YoY) |
Operating expenses | $18.5B–$18.6B (+9% YoY) |
Effective tax rate | About 20% |
Capex | Over $50B in Q2; about $170B for calendar 2026 |
Reading the language:
The Azure guide is conditional. Amy Hood said hitting 49–50% requires Q1's efficiency gains to carry into Q2 and more efficiency gains still to be found, because some supply expected in Q2 moved to the second half; she also acknowledged token pricing can change within the quarter. This is not a 'confirmed' tone but a 'confident, pending execution' tone.
Capex is delayed acceleration, not a cut. Management said capex will accelerate in the second half of the fiscal year, driven by CPUs (including higher memory and storage costs), GPUs and the third-party capacity that shifted to H2. The roughly $170B calendar-year figure is lower than some expected because of delivery timing, not demand.
Operating margin 'down slightly' year over year. Cost guidance includes an estimate for the recently announced voluntary employee exit offer, and COGS growing 26–27%, well ahead of revenue, shows depreciation and compute costs catching up.
A looser stance on financing. Management said it will tap additional funding sources when attractive, noting it already uses leasing and has not yet tapped the debt markets. For an AAA-rated, free-cash-flow-positive company, that line signals capex intensity is now large enough to line up funding in advance.
5. Supply Chain and Customer Signals
Signal one: third-party GPU capacity was this quarter's bottleneck. Delayed third-party capacity came up again and again in the CFO's remarks: the capex miss, Azure that could have been higher, Q2 supply slipping, H2 acceleration all point to the same cause. Management's wording was third-party GPU-as-a-service contracts, capacity that typically comes from neoclouds and data center developers. The implication: the AI infrastructure bottleneck is shifting from chips to how fast data halls, power and full racks can actually be delivered. That is the other side of the same story as Celestica's Q3 FY2026 earnings call calling for $40B of revenue in 2027 and Corning's customers prepaying to fund capacity: demand is queuing while supply-side expansion lags.
Signal two: memory and storage costs were called out explicitly. The capex commentary mentioned higher memory and storage costs twice. Memory is taking a growing share of CPU and GPU server BOMs, which keeps squeezing how much compute each capex dollar buys.
Signal three: custom ASICs and NVIDIA side by side, and an increasingly heterogeneous fleet. Vera Rubin has arrived, GB300 is the deployment workhorse, Maia 200 is live in Teams and Copilot, Maia 300 is on the way, and Cobalt CPUs keep expanding. A heterogeneous fleet means interconnect and network architecture must become more open and standardized. For the optical scale-up standards Microsoft is pushing at OCP, see Microsoft: Why Standards Matter for AI Scale-Up.
Signal four: OpenAI is still the largest single customer. The 10-Q discloses about $10.0B of revenue from commercial arrangements with OpenAI this quarter, including revenue sharing, roughly 11% of total revenue. At the same time Satya spent a lot of time on enterprises owning their own weights and cutting costs with MAI or open-source models. Microsoft is deliberately reducing its dependence on any single frontier model and betting growth on enterprise breadth. He even said a few quarters of growth from four customers buying lots of capacity is not a business.
Signal five: hyperscaler capex rhythms are diverging. Compare Meta's Q3 2026 with record $34.9B quarterly capex and negative free cash flow: Microsoft spent $45.2B this quarter and still generated positive free cash flow of $18.6B, but has started talking about outside financing. The shared signal is that AI infrastructure investment shows no sign of slowing into 2027; the difference is whose balance sheet can sustain it longer.
6. Summary
Microsoft told the market two things this quarter: demand is still queuing, and efficiency is now as important a growth driver as new capacity. Azure at 48% shows software optimization can sell more compute without adding hardware, but that growth is less predictable than building data centers. Whether the Q2 guide is met depends on when third-party capacity arrives and whether more efficiency can be squeezed out.
Three metrics to track next quarter:
Whether actual Q2 capex exceeds $50B, and whether third-party capacity lands on schedule in the second half of the fiscal year. This directly sets the shipment cadence for optical modules, switches and rack-level assembly.
Whether Azure reaches 49–50%, and whether management again credits any beat to efficiency rather than new capacity.
Whether Microsoft Cloud gross margin holds around 65%. The faster it falls, the faster depreciation and memory costs are catching up with revenue.
Capex was not cut, just pushed back. For the AI infrastructure supply chain, the second half is when this Microsoft cycle really pulls in volume.
This article is for technology and industry trend analysis only and does not constitute investment advice.
Comments