Warren Buffett Watched Alphabet's Stock Price Climb 9,000% Before He Decided to Invest. Here's Why He and Greg Abel Are Making It One of Berkshire Hathaway's Biggest Investments Now.

Berkshire Hathaway made Alphabet its third-largest marketable equity holding after initiating the position in Q3 2025 near $200 per share, following Buffett's long-standing regret over missing the stock's earlier rise. Alphabet plans roughly $200 billion of capex this year for AI data centers and servers; Google Cloud CEO Thomas Kurian said average new-server payback is under two years, implying about a 20% annualized lifetime return, while custom TPUs pay back in roughly half that time. Long-term customer contracts averaging five years support demand visibility, and Alphabet trades below 17x forward earnings.

Analysis

The investable question is not whether Alphabet can finance AI infrastructure, but whether incremental AI revenue scales faster than depreciation and power costs. A sub-two-year hardware payback claim is attractive only at sustained utilization; because accelerators depreciate far faster than buildings, a demand or pricing reset would hit operating margin before cash capex can be reduced. The key confirmation over the next 1-3 quarters is Cloud revenue growth accelerating while consolidated capex-to-revenue stabilizes and depreciation growth does not outrun gross-profit expansion.

Custom TPUs create a less-appreciated strategic option: they lower Alphabet's unit cost versus externally sourced GPUs and improve negotiating leverage with NVDA. That is incrementally positive for GOOG margins, but it also raises the risk that a larger share of hyperscaler AI spend migrates from merchant accelerators to captive silicon; the effect is more material to NVDA's long-duration multiple than to its near-term revenue. The structural winner is likely Google Cloud if TPU availability becomes a sales differentiator for enterprise workloads, while smaller cloud challengers lacking proprietary silicon face a higher cost-of-capital hurdle.

Berkshire accumulation is not a clean near-term catalyst because position changes are disclosed with a lag and may already be reflected in institutional positioning. Consensus may also be underestimating regulatory optionality: an adverse search remedy could make distribution economics less durable, but it could simultaneously unlock value if investors assign higher standalone multiples to YouTube, Cloud, and Waymo-like assets. A break in Cloud backlog, falling utilization, or a material adverse remedy in the antitrust cases would falsify the margin-expansion thesis.

Key Decisions for Investors

• Initiate a 6-12 month long GOOG position only on post-earnings evidence that Cloud growth accelerates while depreciation and capex guidance remain contained; target 15-20% upside from multiple stability plus earnings revisions, with a 8-10% stop if Cloud growth decelerates or capex rises without corresponding backlog disclosure.
• Express the custom-silicon theme as a small 6-month pair: long GOOG / short NVDA, sized beta-neutral. The thesis is TPU-driven share-of-wallet and lower AI inference cost; exit if NVDA data-center guidance or hyperscaler GPU purchase commitments reaccelerate materially.
• Do not chase BRK.A on expected Alphabet buying. Treat the next 13F as an event watch item rather than a trade catalyst; a disclosed increase would be more useful as confirmation of valuation discipline than as new information on Alphabet fundamentals.
• Monitor GOOG quarterly disclosures for Cloud operating margin, remaining performance obligations/backlog, depreciation growth, and server utilization. If revenue growth fails to cover the incremental depreciation burden for two consecutive quarters, reduce or avoid the long regardless of headline AI demand.