Microsoft And Alphabet's $375B Q1: The AI Capex Industrial Phase
How Microsoft's $190B plan and Alphabet's zero buybacks made capital structure the deciding variable in AI infrastructure leadership
Welcome to Global Data Center Hub. Join investors, operators, and innovators reading to stay ahead of the latest trends in the data center sector in developed and emerging markets globally.
The AI capex debate in 2024–2025 centered on headline spend levels whether Amazon could clear $200B, Alphabet sustain ~$90B+, and Meta execute gigawatt-scale buildouts. Scale was the primary narrative.
That framing has now broken.
Microsoft’s ~$190B 2026 capex came in far above expectations, but driven by component inflation, not incremental capacity.
Alphabet also paused buybacks and acquired Intersect for $5.9B, signaling a shift toward energy and infrastructure control.
Together, these moves mark a transition into an industrial capex regime.
Procurement costs, energy access, and supply chain ownership are now as critical as raw compute demand, forcing capital structures to adapt in real time.
Markets are repricing accordingly. Microsoft fell despite strong beats across revenue, Azure, and AI metrics, as component inflation is now treated as structural.
Alphabet showed a similar pattern, where headline earnings strength masked softer underlying recurring profitability.
Microsoft: The Cost of Executing Demand
Microsoft delivered the strongest demand signal of the cycle. RPO hit $627B, up 99% YoY, with ~30% expected to convert within a year.
Copilot surpassed 20M paid seats, and large enterprise adoption is accelerating sharply.
Infrastructure is scaling in response.
Microsoft added ~1 GW of capacity this quarter and is on track to double its global data center footprint in two years.
The new Wisconsin Fairwater facility is built as a dense AI training cluster using NVIDIA Blackwell GPUs and high-speed networking.
First-party silicon is becoming core to the model.
Cobalt CPUs are now deployed in nearly half of Azure regions, while Maia 200 targets lower-cost inference to reduce reliance on merchant GPUs. The strategy aligns with broader hyperscaler vertical integration in AI compute.
Capital intensity is rising fast.
Capex reached $30.9B (+85% YoY) and is expected to exceed $40B next quarter, with margins at multi-year lows. The OpenAI restructuring reduces revenue-sharing pressure while preserving long-term model access through 2032, easing margin drag without sacrificing strategic positioning.
Read the full Microsoft Q3 FY2026 analysis: The $190B Capex Plan That Repriced AI
Alphabet: When Buybacks Become a Capex Question
Alphabet’s Q1 sent a clear capital allocation signal. Buybacks dropped to zero from $15.1B a year earlier, despite $126.8B in cash and securities. The shift reflects prioritization of infrastructure over financial returns.
Capex surged 107% YoY to $35.7B, with ~60% focused on servers and AI hardware. Full-year 2026 guidance now sits at $180–190B, with 2027 expected to be “significantly higher,” extending the investment cycle further into the decade.
Strategic acquisitions reinforced vertical integration. Wiz strengthens cloud security, while Intersect signals a move toward internalizing energy infrastructure, reducing reliance on external utilities and PPAs.
Cloud revenue hit $20B (+63% YoY), with margins rising to 32.9%. Backlog reached $467.6B, but demand now exceeds available capacity, making growth increasingly dependent on power access and AI silicon supply.
TPU 8 commercialization marks a competitive shift. Alphabet is now selling its in-house chips externally, challenging Nvidia on cost and performance while lowering its own inference economics.
Read the detailed Alphabet’s $190B Reset breakdown: Buybacks Pause as Power Becomes the Constraint
Power, Components, and the Sequencing Problem
The cohort has converged on a single operating reality.
Combined 2026 hyperscaler capex is approaching $700B: Amazon ~$200B, Alphabet $180–190B, Microsoft ~$190B, Meta ~$125B.
Spend is concentrated in a tight constraint stack advanced packaging, GPUs/TPUs, liquid cooling, transformers, and grid power.
Power remains the primary bottleneck.
Microsoft has secured 10+ GW of clean energy and is restarting nuclear at Three Mile Island (835 MW expected by 2027, ahead of schedule).
Amazon added 4 GW in 2025 with plans to double by 2027. Meta is building gigawatt-scale campuses in Ohio and Indiana.
Alphabet has acquired a renewable infrastructure platform outright.
Component supply is now the second constraint, formalized in Microsoft’s $25B disclosure.
Execution is now sequential.
Power first, then land and grid interconnects, followed by silicon allocation, networking, and cooling.
Hyperscale efficiency now depends on coordinating all four procurement layers in parallel, not stepwise.
The implication for emerging markets is tightening.
Buildouts are clustering in regions with reliable grids, water access, regulatory stability, and long-duration policy continuity.
That concentrates investment in the U.S., select European hubs, and parts of Asia-Pacific.
The infrastructure cycle is not broadening globally it is concentrating geographically.
What Resolves Next
Three signals will define the next four quarters.
First, Microsoft’s gross margin. A move from 68.7% to 67.6% is fine, but below 65% would challenge current valuation assumptions around operating leverage.
Second, Alphabet’s 2027 capex. A $210–230B range implies a longer build cycle into 2028 and a structural pause in buybacks. A $190–200B range suggests the cycle is closer to peaking, with capital returns resuming sooner.
Third, merchant silicon adoption. Microsoft’s Maia 200 and Alphabet’s TPU commercialization both aim to reduce GPU dependence. If uptake accelerates, margins improve and Nvidia exposure declines; if not, the strategy remains primarily an internal capacity expansion play.
The Bottom Line
Microsoft formalized component inflation as a separate capex line. Alphabet paused buybacks entirely and bought generation outright. Both disclosures point at the same conclusion: capital structure is now the deciding variable in AI infrastructure leadership.
For investors, the question is no longer whether AI demand justifies capex. RPO at $627B and backlog at $467.6B answered it. The question is whether component inflation and depreciation compress margins faster than monetization expands.
For operators, the binding constraint has bifurcated. Procurement strategies must execute power, land, silicon, and networking cycles in parallel.
The Q1 prints did not change the demand story. They formalized the cost of executing it.
Useful framing. The "power first, then land, then silicon" sequencing is the right read — power decisions are 5-7 year commitments while silicon turns over in 18 months. Does that mean buying your own renewable, like Alphabet/Intersect, is the new vertical integration play that replaces PPAs entirely?