Berkshire Hathaway's investment trajectory in Google parent Alphabet (GOOGL.US) has formed a clear pattern: a substantial position initiated in the twilight of Warren Buffett's tenure, followed by continued expansion under successor Greg Abel. After Abel assumed the CEO role in January 2026, the combined holdings of Alphabet's Class A and Class C shares grew to approximately 57.835 million shares in the first quarter, then increased by another 48.1446 million shares in the second quarter to reach roughly 105.98 million shares. At quarter-end, the combined market value stood at about $37.764 billion, representing approximately 12.6% of Berkshire's 13F-disclosed equity portfolio and ranking as the firm's third-largest holding.
Wall Street asset managers and veteran analysts with a long-term bullish outlook on Alphabet's share price believe Google can monetize across multiple segments of the global AI application and computing value chain simultaneously—AI features in search capture user demand, Gemini and enterprise products expand subscription and application revenue, proprietary TPU AI computing infrastructure aggressively expands computing resource sales, and Google Cloud sells B2B AI cloud capacity and model services. Its significant equity stake in AI application leader Anthropic provides another frontier exposure to the cutting-edge AI model ecosystem.
Google's fundamental technological and ecosystem advantage in the unprecedented global AI wave lies in the synergistic optimization of its proprietary AI chips, model training, cloud-based AI inference computing services, and Google's exclusive massive product distribution system: enhanced model capabilities flow directly into existing Google online ecosystems and internet product lines, while surging user and enterprise demand converts into robust demand for TPU computing resources, AI cloud services, and subscriptions.
Financial data already reflect this transmission mechanism—Alphabet disclosed in Q2 2026 that Google Cloud revenue reached $24.8 billion, up 82% year-over-year, with an operating margin of 35.6% and a contract backlog of $514 billion. The company also confirmed it is training the larger Gemini 4 foundational model. From this perspective, Berkshire's heavy position corresponds to an AI business system that already generates revenue and profits with the potential to continue expanding its market boundaries.
Berkshire's Successive Accumulation, Gemini 4, and RSI Spark AI Sector Debate—Alphabet's AI Compounding Logic Unfolds
In Q3 2025, Berkshire first disclosed holding approximately 17.8461 million shares of Alphabet's Class A stock, valued at about $4.338 billion at quarter-end. Following Abel's January 2026 appointment as CEO, combined holdings of both share classes grew to approximately 57.835 million shares in Q1, then increased by another 48.1446 million shares in Q2 to reach approximately 105.98 million shares, valued at roughly $37.764 billion at quarter-end—about 12.6% of its 13F-disclosed equity portfolio and the third-largest holding when both classes are combined. This continuous accumulation spanning the management transition indicates Alphabet has evolved from a new investment into a core allocation.
From an investment logic perspective, Google combines mature business cash generation, world-leading frontier AI R&D capabilities, top-tier AI computing infrastructure, and an AI application commercial distribution system, allowing long-term capital to participate in both the compounding of existing businesses and the incremental market created by new technologies.
The most noteworthy aspects of Google's next-generation AI model—Gemini 4—are the potential simultaneous enhancement of multiple capabilities required for complex tasks and the advancement of Google's AI training process into what is known as RSI (Recursive Self-Improvement). RSI does not mean the model suddenly gains self-awareness; its currently verifiable form involves AI progressively participating in training strategy planning, algorithm generation, code writing, experiment design, result evaluation, error fixing, and subsequent training rounds.
RSI could create a "second demand curve" for computing power independent of end-user demand: training becomes a continuous closed loop where models propose algorithms, run hundreds to thousands of experiments in parallel, evaluate results, and initiate the next training round, rather than the traditional "train-release-done" cycle.
Market reports suggest Google has quietly tested its flagship model Gemini 4 Pro under the alias "gemini-3.8-flash" on the AI Arena benchmark platform. Disclosed benchmark data indicates it comprehensively surpasses OpenAI's Astra and Anthropic's Fable 5.1 in coding, agentic capabilities, and reasoning—earning the label "the world's most powerful AI model."
The score data for Gemini 4 Pro under the "Gemini 3.8 Flash" alias shows: DeepSWE v1.1 at 88.7%, 1.8 percentage points higher than Astra's 86.9%; GDPval-AA v2 at 2064 Elo; Terminal-bench 2.1 at 95.3%; and OSWorld-2.0 at 86.8%. These metrics cover software engineering, knowledge work, terminal operations, and computer use respectively. If formally verified, they suggest the model can more completely execute the workflow of "understanding requirements—formulating plans—calling tools—verifying results," with demonstrations of web pages, SVG, 3D models, and playable games showcasing these capabilities applied intuitively.
In terms of circulating market pricing, compared to Astra and Fable, it is also the most cost-effective among the "big three AI model providers": $2.25 per million input tokens and $11.25 per million output tokens. The combination of stronger task capability and lower invocation pricing could lower the economic threshold for enterprises adopting AI agents.
The computing logic shared by Astra and next-generation Gemini is that AI is transitioning from answering questions to continuously executing tasks. Test-time scaling, multi-round tool invocation, parallel exploration, and result verification increase computational steps for complex tasks; longer contexts raise prefill computation and KV cache capacity and bandwidth requirements. OpenAI recently disclosed that Astra has strengthened programming, browsing, computer operation, and complex task execution capabilities, expanding the scope of practical work the model can participate in. Product leaders stated that surging Astra demand has strained infrastructure; the company has suspended new subscriptions and upgrades for the $200-per-month Pro 20X plan since September 10, while existing subscribers continue with normal renewals.
The new-generation Gemini possesses the technological and AI commercialization foundation to become a powerful new bullish catalyst for Alphabet's share price.
For Alphabet, the key advantage is the ability to jointly optimize costs from chip to service level: TPU, HBM, high-speed chip interconnect, and scheduling software work together to improve the effective computing delivered per unit of power and hardware investment. As per-unit successful task costs decline and more programming and knowledge work becomes worthwhile to delegate to AI, the scale of new tasks can drive continued expansion of total computing demand, simultaneously growing Gemini's paid demand and Google Cloud's computing revenue.
The new-generation Gemini, along with OpenAI's GPT-6 Astra model, has ignited "AGI" discussions, while the RSI technical path focused on by AI leaders promises to be one of two core drivers of exponential AI computing demand growth: more powerful AI models with broader AI application tool usage, and next-generation AI training paths requiring stronger computing power. This strengthens the case for sustained growth in AI infrastructure demand.
Quantifiable engineering evidence for RSI has already emerged: the Dream-RSI study published on September 14 maintained the underlying coding agent unchanged while allowing the system to use historical exploration records to improve search branching, resource allocation, and stopping strategies, then feed improved strategies back into subsequent exploration rounds. In VGG16 and LayerNorm kernel engineering tasks, iteration counts to achieve comparable performance dropped to approximately 1/2.43 and 1/1.79 of the control group respectively. This represents self-improvement at the exploration strategy level, providing empirical support for "AI helping improve the AI development process" and opening space for faster software optimization and lower computing costs.
Furthermore, if next-generation Gemini stably delivers stronger reasoning, coding, and computer operation capabilities into commercial products, Google has the opportunity to drive a new round of value revaluation through improved task success rates, expanded enterprise adoption, cloud and subscription revenue growth, and upward profit estimate revisions. AGI and RSI discussions can elevate market attention, while Google's existing product entry points, exclusive TPU computing infrastructure, and profitability base provide what appears to be the clearest shareholder return transmission path on Wall Street for this technological progress.
AI Application Cash Cow Combined with TPU Monetization—Wall Street Bets on Google's Dual-Engine Growth
Berkshire Hathaway's favored artificial intelligence stock appears to be an extremely sensible bargain-buying choice. Missed the "first act" of the AI megatrend? The "second act" could be 15 times larger. Most investors believe they missed the AI ship by not buying Nvidia at historically low levels in 2005. But according to some Wall Street analysts, we are currently only at the end of the "first act"—the AI prototype development phase. The "second act" is global application deployment.
Berkshire Hathaway's massive position in Alphabet and its determined accumulation can be viewed as a significant vote of confidence. In Q2, Berkshire purchased more than 48 million shares of Alphabet's two share classes, with combined holdings now representing approximately 12.6% of its equity portfolio. Alphabet is Berkshire's third-largest stock holding, reflecting immense confidence in the tech giant.
Although legendary investor Warren Buffett no longer occupies the CEO office, he initiated the first purchase of Alphabet before the end of his Berkshire Hathaway CEO term. Over the weekend, he also announced he would step down as chairman to become honorary chairman. However, current management is continuing his investment tradition, with recent moves demonstrating strong confidence in Alphabet.
Google parent Alphabet has not concentrated all its AI bets on a single path. Instead, it is deploying broadly. First and foremost, Alphabet's business is anchored by the Google search engine; by integrating AI Overviews into search results, it provides users with a convenient AI entry point. Most people will encounter generative AI through this feature and enter the broader Google ecosystem. Alphabet's large language model family, represented by Gemini, also ranks among the best-performing AI models, offering premium subscription versions that generate additional revenue.
If Alphabet's models are surpassed by competitor products, it remains a cornerstone investor in Anthropic, which developed the Claude family of AI models. Anthropic has also become one of the best-performing options and could go public in the near future. Its management's pursuit of a $2 trillion valuation, if achieved, would generate substantial returns for Alphabet.
Finally, Alphabet owns Google Cloud, an infrastructure platform handling workloads for numerous AI companies. It rents out substantial computing power to customers, and given that existing capacity is insufficient to meet current demand, this business will continue growing for a considerable time as more AI computing capacity comes online.
Alphabet is positioned to benefit from AI in multiple ways. Not everything it does will ultimately become a successful solution, but if most of its bets succeed, it will emerge as one of the biggest winners in the AI era.
On Wall Street, analysts' average 12-month price target for Alphabet stands at $428.41, with a high target of $515, corresponding to potential upside of approximately 22.56% and 47.34% respectively. Calculations use Class A shares (GOOGL) based on the September 18 regular trading close of $349.54; targets derive from StockAnalysis-disclosed S&P Global Wall Street analyst consensus, with a composite "Strong Buy" rating. The highest target comes from Citizens JMP analyst Andrew Boone, who reiterated a $515 target with an "Outperform" rating. His bullish thesis centers on AI expanding the scope of searchable questions and enhancing user engagement, injecting renewed growth momentum into Google's core advertising revenue engine.
Tigress Financial analyst Ivan Feinseth raised his target to $485 on September 17, maintaining a "Strong Buy" rating, corresponding to approximately 38.75% upside from the same baseline. His core logic includes: AI expands search activity and improves ad effectiveness; Google Cloud's accelerating growth and margin expansion strengthen profitability; and Gemini, proprietary TPUs, and AI services open new commercialization channels. He emphasized the dual-growth logic of "AI-enhanced search engine leading AI application cash generation, while expanding revenue streams through next-generation models and AI computing services"—with a massive commercial user base and proprietary AI computing infrastructure allowing Google to capture AI demand across multiple segments.
Evercore ISI's Mark Mahaney raised his target to $450, with field research showing AI products significantly improving B2B and C2 user experiences, behavioral improvements supporting revenue growth, and upward earnings revisions supporting a higher target. Evercore ISI's survey found 78% of respondents in August 2026 named Google as their preferred search engine, improving for four consecutive quarters from a prior low of 70%. Based on this, Mahaney raised 2026–2028 search revenue growth forecasts, with 2028 total revenue and operating profit projections approximately 4%–5% above market consensus.
Comments