Most important: Q3 capex vs $34.3B consensus and $145B annual anchorIn the print: capex $, Azure growth %, operating margin, RPOOn the call: H2 capex cadence, Maia 200 fleet share, Fairwater timelineDownstream: NVDA, AVGO, TSMC, SK Hynix, ETN, PWR
House view
Street prices MSFT on Azure growth durability. The $37.5B Q2 capex and $145B annual reference signal a spend trajectory that tightens GPU and power delivery through CY26. We read this print as a capacity cadence signal that firms NVDA backlog and ETN power orders through mid-2027.
What is priced in
Street expects Q3 FY26 revenue ~$81.2B, EPS ~$4.05, Azure growth 37-38% constant currency, capex ~$34.3B, and operating margin ~45.5% vs guide of 45.1%. Not priced in: whether the $145B annual capex anchor holds or ratchets higher, Maia 200 custom silicon mix shift that could pressure NVDA/AVGO allocation assumptions, or whether the 80%+ AI capacity growth target pulls forward CY27 power commitments to ETN and PWR.
What decides the capex readRanked by capex-trajectory impact
Priority 1 · primary capex signal
Q3 capex vs $34.3B consensus and $145B annual anchor
Q2 printed $37.5B. If Q3 stays above $34B, the $145B annual run rate is intact or accelerating. This is the largest single demand signal for NVDA GB200 shipments and TSMC CoWoS allocation through CY26.
✓Q3 capex at or above $35B with reaffirmation of $145B annual reference and mention of new capacity coming online in H2.
✗Q3 capex below $32B with language about optimizing spend cadence or deferring site buildouts to CY27.
Priority 2
Maia 200 custom silicon fleet share vs NVDA/AMD mix
Maia 200 is online with 30%+ TCO improvement. Rising custom silicon share shifts capex dollars from NVDA toward AVGO/internal design. The mix split determines who captures the next $10B+ of incremental spend.
✓Maia 200 deployed across multiple Azure regions with specific workload share cited, plus continued NVDA GB200 procurement language.
✗Maia 200 limited to internal workloads only, with no fleet share disclosure and heavy emphasis on third-party GPU procurement.
Priority 3
Forward capacity anchoring: 2GW Fairwater and doubling footprint timeline
Fairwater (2GW, online next year) and the 2-year doubling target set the outer boundary of power and cooling demand. Any acceleration pulls forward ETN switchgear and PWR electrical orders.
Working the eventRelease drop vs. Q&A
In the release · first 60 seconds
Quarterly capex (incl. finance leases)
$37.5B (Q2 FY26), Street expects ~$34.3B for Q3
Whether Q3 capex stays above $35B, confirming the $145B annual run rate, or drops below $32B signaling a cadence pullback.
FY26 capex guide or annual anchor
$145B annual reference from Q2 FY26 commentary
Any reaffirmation, raise, or softening of the $145B figure. A raise above $150B would be a new demand signal for NVDA and ETN.
Revenue
Street ~$81.2B; guide midpoint $81.2B
Whether revenue clears $81.8B, indicating Azure demand is absorbing the capex ramp, or misses below $80.6B.
EPS
Street ~$4.05; Q2 FY26 printed $4.14
Whether EPS holds above $4.05 despite margin pressure from capex depreciation. A miss below $3.95 would raise FCF concern.
Azure revenue growth (constant currency)
Guided 37-38% for Q3 FY26; Street ~37.1%
Whether Azure growth hits 38%+ confirming demand justifies the capex ramp, or decelerates below 36% creating a spend-vs-demand gap.
NVDA and TSMC still see firm orders near-term since capex is committed, but demand softness raises risk of order deferrals in H1 CY27. ETN backlog intact but new orders slow.
2015 signals · 54 high-qualityResearch read-through · not a trade recommendation
✓Fairwater on track for CY27 with additional mega-site announcements, reaffirming the 2-year doubling commitment.
✗Fairwater delayed past CY27 or doubling language softened to demand-dependent phrasing without new site names.
Guided ~45.1% for Q3; Street ~45.5%; Q2 printed 46.7%
Whether margin holds above 45% or compresses below 44%, which would signal depreciation from the capex ramp is hitting faster than expected.
On the call · where capex read moves
H1/H2 capex split and Q4 FY26 capex trajectory
A back-half acceleration above $38B/quarter would confirm the $145B anchor is a floor, not a ceiling, firming NVDA H2 delivery schedules.
Maia 200 deployment regions and workload share
Specific fleet share numbers for Maia 200 shift the custom-vs-merchant GPU mix read. Rising Maia share is bullish AVGO, mixed for NVDA.
Fairwater 2GW site timeline and new mega-site announcements
On-track Fairwater plus new sites extends ETN switchgear and PWR electrical backlog into CY28. Delays ease the power constraint.
Inference throughput gains and tokens-per-GPU improvement
Q2 cited 90% more tokens per GPU YoY. Further gains could slow GPU procurement growth even as demand rises, easing NVDA order urgency.
RPO growth and OpenAI $250B contract consumption pace
RPO was ~$400B with 2-year duration. Faster OpenAI consumption locks in capex commitments and tightens SK Hynix high-bandwidth memory supply.
NVDA faces a near-term order gap as MSFT leans on inference efficiency and Maia 200. AVGO benefits if custom silicon fills the GPU gap. SK Hynix and Micron see softer pull from MSFT specifically.
NVDA, AVGO, and TSMC all face demand revision risk. SK Hynix high-bandwidth memory pricing power weakens. ETN and PWR new data center orders at risk of deferral. Amazon noting memory supply volatility compounds the concern.