Google Vs. Nvidia: Inside The AI Hardware Showdown

Alphabet’s capital expenditures have blown past early-2025 expectations—rising from an initial ~$60 billion estimate to guidance that has climbed through $75B and $85B to a current projection of $91–93B—driven primarily by investment in AI infrastructure (servers, storage, power/cooling and chips) for Search, Ads, YouTube, Gemini and Cloud. Google remains a top direct customer for Nvidia as hyperscalers drive heavy GPU demand, but GPU spending appears to be growing faster than Google Cloud revenue (Cloud +33.5% YoY to $15.1B last quarter versus capex up ~75% YoY), creating potential pressure on cash flow and returns. To balance flexibility with cost control Google is pursuing a dual-track approach—continuing to buy Nvidia GPUs while aggressively scaling its own TPUs for inference (Ironwood v7 reportedly >4x faster than v6 and ~10x peak vs v5p, with each TPU generation delivering 2–3x better performance-per-dollar)—and has begun monetizing TPUs externally (e.g., Anthropic’s large commitment). The strategic shift to TPUs should materially reduce Google’s per-query compute costs, increase its negotiating leverage with Nvidia and could boost margins and valuation as workloads migrate, though CUDA lock‑in and enterprise expectations mean Nvidia displacement will be gradual.

Analysis

Alphabet has dramatically ramped capital expenditures in 2025, with guidance moving from an initial ~$60 billion to $75 billion in February, to $85 billion mid-year, and to a current projection of $91–93 billion, driven explicitly toward AI infrastructure—servers, storage, power/cooling and a large volume of chips for Search, Ads, YouTube, Gemini and Google Cloud. Management signals that 2026 is likely to see a further increase, indicating sustained heavy investment that is already ~75% higher YoY versus a Google Cloud revenue increase of +33.5% YoY to $15.1 billion last quarter.

Google remains a top direct purchaser of Nvidia hardware; Nvidia reported that anonymous customers accounted for 39% of Q2 FY26 revenue (likely hyperscalers), and Nvidia sales rose ~58% driven by pricing power. The mismatch—GPU expenditures rising faster than Cloud revenue—implies potential pressure on free cash flow and returns, creating an incentive to control costs and diversify compute sources.

Google’s explicit dual-track strategy pairs continued Nvidia flexibility with accelerated deployment and commercialization of its custom TPUs. The new Ironwood v7 TPU claims >4× speed vs v6 and ~10× peak versus v5p, with each TPU generation delivering 2–3× better performance-per-dollar; external commitments such as Anthropic’s up-to-1M TPU deal and anticipated gigawatt-scale buildouts suggest meaningful external demand, which should lower per-query costs, improve margin dynamics and increase Google’s negotiating leverage with Nvidia, although CUDA lock-in and enterprise expectations mean displacement will be gradual.

Key Decisions for Investors

• Investors should consider maintaining or modestly increasing exposure to Alphabet to capture potential margin expansion from TPU-driven cost reductions and external TPU monetization, while sizing positions to reflect elevated capex that could pressure near-term free cash flow
• Monitor upcoming quarters for capex guidance versus Google Cloud revenue growth, soft metrics of TPU adoption (external commitments and Vertex AI usage) and Nvidia revenue mix; if GPU spending continues to materially outpace Cloud revenue or TPU adoption lags, reduce exposure or implement hedges
• Reassess Nvidia positions in light of Google’s TPU scaling: avoid adding to NVDA exposure based solely on hyperscaler demand until there is clear evidence Nvidia can maintain pricing power and share despite increasing TPU workloads, and consider tactical hedges against potential GPU demand erosion