Why Google’s $185B AI Capex Signals a Fundamental Shift in the AI Race

Google’s 2026 AI‑focused capital expenditure

Alphabet announced a 2026 capex plan of $1.75‑$1.85 trillion, roughly double analyst expectations and exceeding the combined spend of the previous three years. About 75 % of the budget is earmarked for AI infrastructure.

Capex comparison among hyperscalers (2026)

Alphabet: $185 B (projected)

Meta: $115‑$135 B

Microsoft: ≈ $140 B (FY2026)

Amazon: ≈ $125 B

Total: > $600 B, a 36 % increase over 2025 record spend.

Allocation of Google’s AI spend

GPU clusters and accelerators

Data‑center construction and expansion

Network equipment for distributed training

Power infrastructure for energy‑intensive workloads

The remaining 25 % supports DeepMind research, strategic “other bets,” and improvements to user experience and advertising ROI.

Economic context

Google Cloud’s backlog grew 55 % quarter‑over‑quarter, reaching $240 B year‑over‑year, indicating strong enterprise demand. AI hardware typically depreciates at ~20 % per year; if current spending continues, the five major cloud providers could hold about $2 trillion in AI‑related assets by 2030, generating roughly $400 B in annual depreciation—comparable to their combined 2025 profits.

Revenue trends: Gemini AI reported 750 M monthly active users (up from 650 M) and a 78 % reduction in inference cost in 2025. Microsoft Azure AI revenue rose 175 % YoY, and AWS AI workloads are expanding rapidly.

Implications for data and engineering teams

Rising cloud costs: Capital intensity will be passed to customers; budgeting must account for higher GPU pricing and limited supply.

Resource concentration: Only a few organizations can afford the compute required for state‑of‑the‑art models, widening the capability gap.

Multi‑cloud & hybrid strategies: Teams are diversifying across providers—Google Cloud for Gemini‑integrated workloads, AWS for data lakes, Azure for Microsoft 365/Copilot, and on‑prem GPU clusters for experimentation.

Energy considerations: Power infrastructure is becoming a primary cost driver; data‑center location decisions now factor electricity availability and grid constraints.

Key metrics to monitor

Revenue per dollar of infrastructure: Google Search revenue grew 17 %; Google Cloud revenue grew 48 % YoY. The critical question is whether revenue can sustain >100 % infrastructure growth.

Cost‑efficiency trajectory: Gemini inference cost fell 78 % in 2025; maintaining a 70‑80 % annual decline is essential for the economic model.

Adoption and monetization: 750 M Gemini users exist, but conversion to paid usage remains uncertain; Microsoft’s AI revenue is tied to enterprise contracts.

Risk assessment

Not all hyperscalers may maintain the current spending pace through 2027. Meta appears most vulnerable due to the lack of direct AI‑revenue cloud services, while Microsoft and Google have clearer enterprise AI revenue streams. Amazon occupies an intermediate position with a strong AWS business but increasing AI competition.

Conclusion

Alphabet’s $185 B AI capex signals a shift from experimental AI to a large‑scale infrastructure‑building phase. The sustainability of this model depends on whether revenue growth can outpace or at least match the unprecedented capital outlays.