Not Nvidia, Not Micron. This Magnificent Warren Buffett Stock Could Be the Quiet Winner of the AI Arms Race -- Here's the Case.

Berkshire Hathaway has built an approximately $38 billion Alphabet position, or about 106 million shares and 13% of its equity portfolio, following accelerated purchases and a reported $10 billion private-placement investment. Alphabet's Google Cloud revenue reportedly grew 63% year over year to $20.0 billion in Q1 and 82% to $24.8 billion in Q2, with operating margins rising to 36% and backlog reaching $514 billion. Despite planned annual capex of $195 billion-$205 billion and temporarily negative free cash flow, the article argues Alphabet's 16x forward P/E undervalues its AI, cloud, custom-chip, and Waymo businesses.

Analysis

The investable implication is not simply that GOOG is an AI beneficiary; it is that vertical integration can shift the AI profit pool away from merchant silicon if TPUs achieve meaningful external adoption. Every workload retained internally on TPU reduces Alphabet's effective compute cost and potentially displaces incremental NVDA accelerator demand at the margin, while GOOG monetizes the same asset through Cloud and model distribution. This makes long GOOG / short NVDA a credible 6-18 month relative-value expression, but only if TPU availability and customer uptake become visible rather than remaining a captive-cost advantage.

Near term, the market is likely to focus on capex-to-revenue conversion, not the headline backlog. A large contracted-demand figure does not establish revenue timing, cancellation protection, or hardware pass-through economics; investors should monitor Cloud growth net of AI infrastructure depreciation, operating-margin progression, and RPO conversion over the next two earnings cycles. The key risk is that incremental Cloud revenue is low-margin GPU resale while Search faces higher inference costs and AI-answer cannibalization, producing FCF pressure without a commensurate earnings re-rating.

Consensus may be underweight Alphabet's ability to fund AI through high-margin distribution assets, but it may also be too willing to capitalize Waymo and TPU optionality before standalone economics are disclosed. The more actionable catalyst is evidence that AI products protect Search query monetization and raise Workspace/Cloud ARPU; absent that, a lower headline multiple can be warranted by structurally higher capex and depreciation. A sustained Cloud-margin decline below the current trajectory or a material reduction in companywide buybacks would falsify the "self-funding platform" thesis.

Key Decisions for Investors

• Initiate a 6-12 month long GOOG / short NVDA pair, sized beta-neutral, after the next earnings print if GOOG Cloud growth remains above 40% and Cloud operating margin holds above 30%. Target 15-20% relative outperformance; stop if Cloud margin falls below 28% or management signals TPU economics are not improving compute cost.
• For directional exposure, accumulate GOOG on capex-driven post-earnings weakness rather than chase AI enthusiasm. Use a 12-month downside threshold of roughly 15% from entry and underwrite upside only if Search monetization and Cloud revenue both accelerate; do not attribute valuation to Waymo without disclosed contribution margin or cash-burn data.
• Maintain MU and NVDA exposure only as a tactical 1-3 month AI-capex trade, not a clean structural proxy for all AI spending. Reduce if hyperscaler commentary shifts from capacity scarcity toward utilization optimization, because HBM and accelerator order revisions would transmit quickly through inventory and multiple compression.
• Set an alert around external TPU customer disclosures, Cloud backlog conversion, and Alphabet depreciation growth. Verified third-party TPU wins would strengthen the GOOG/NVDA relative trade; backlog growth without revenue conversion or rising depreciation would instead favor staying neutral.