Warren Buffett's Successor, Greg Abel, Buys an AI Stock That Poses a Threat to Nvidia

Warren Buffett took control of Berkshire Hathaway (NYSE: BRKA) (NYSE: BRKB) in 1965. Under his leadership, the company grew into a multinational conglomerate worth more than $1 trillion, and the stock returned about 6,100,000%. Over the same period, the S&P 500 (SNPINDEX: ^GSPC) returned about 46,100%.

After six decades, Buffett stepped down as Berkshire's CEO in December 2025 and handed the reins to Greg Abel, who previously served as vice chairman of noninsurance operations. Abel has already overseen major changes to Berkshire's stock portfolio, including his decision to triple the company's stake in Alphabet (NASDAQ: GOOGL) (NASDAQ: GOOG) in the first quarter.

Berkshire now has nearly 7% of its $332 billion portfolio invested in Alphabet stock. That could be a brilliant move, given that the company has positioned itself as a serious threat to Nvidia (NASDAQ: NVDA). Here's what investors should know.

Image source: Getty Images.

Alphabet's Google has been building custom AI chips for over a decade

Alphabet began developing custom artificial intelligence (AI) accelerator chips called Tensor Processing Units (TPUs) in 2015. They were "designed for a single, specific purpose: running the unique matrix and vector-based mathematics that's needed for building and running AI models," according to the company.

Alphabet, which recently launched its 8th generation TPUs, first made its custom AI chips available to Google Cloud customers in 2018. They have since become an important source of revenue, drawing demand from several large AI companies, including Anthropic and Meta Platforms.

• Anthropic has agreed to buy or rent up to a million TPUs in a deal valued at "tens of billions of dollars." VentureBeat reports that about 600,000 of those chips will be leased through traditional cloud computing contracts, while 400,000 will be sold directly to Anthropic.
• Meta Platforms has agreed to rent TPUs in a deal worth "billions of dollars," according to The Information. In addition, Meta is reportedly talking to Alphabet about buying TPUs to deploy in its own data centers.
• Alphabet partnered with Blackstone to create a new cloud computing company that will exclusively use Google hardware (including TPUs) and software. The unnamed company will bring 500 megawatts of computing capacity online by 2027.

Alphabet has historically restricted access to TPUs to Google Cloud customers, but the company is changing tack in an effort to take market share from Nvidia. "As TPU demand grows from AI labs, capital markets firms, and high-performance computing applications, we will begin to deliver TPUs to a select group of customers in their own data centers in a hardware configuration to expand our addressable market," CEO Sundar Pichai told analysts in April.

Google TPUs are unlikely to overtake Nvidia GPUs as the dominant AI accelerator

Nvidia invented graphics processing units (GPUs) in 1999, and they have become the industry standard in AI accelerators. The company holds more than 80% market share, and it has a key competitive moat in its CUDA platform. The CUDA ecosystem comprises hundreds of code libraries and pretrained AI models, as well as application frameworks that streamline development across a broad range of use cases.

Custom chips like TPUs lack robust software ecosystems. "Mainstream AI development relies primarily on PyTorch, an open-source [machine learning] framework that can be tuned for CUDA," writes VentureBeat. Alphabet only recently added PyTorch support to TPUs, and its software stack still falls woefully short of matching CUDA's breadth.

In addition, Nvidia GPUs are designed as general-purpose accelerators, which means they can run far more algorithms than Google TPUs, which are specifically built for certain AI tasks. That makes TPUs more cost-efficient in certain cases, but it also limits their utility. VentureBeat explains, "If a new AI technique is invented tomorrow, a GPU will run it immediately." The same cannot be said of TPUs.

Even so, some Wall Street analysts see a long runway for growth. Gil Luria at DA Davidson estimates TPUs could capture 20% of the AI accelerator market over time, which would make custom chips a $900 billion business for Alphabet, reports Bloomberg. But even in that scenario, Nvidia would still dominate the market.

Wall Street says Alphabet stock is slightly undervalued

Alphabet reported impressive financial results in the first quarter. Revenue increased 22% to $109.8 billion, the fourth straight acceleration, driven by strong demand for TPUs and AI models. Meanwhile, operating income soared 39% to $39.6 billion, driven by margin expansion in the cloud computing segment.

Wall Street expects Alphabet's earnings to grow at 16% annually over the next three years. That makes the current valuation of 29 times earnings look reasonable. Most analysts think the stock is slightly undervalued. The median target price of $430 per share implies 11% upside from the current share price of $386.

Trevor Jennewine has positions in Nvidia. The Motley Fool has positions in and recommends Alphabet, Berkshire Hathaway, Blackstone, Broadcom, Meta Platforms, and Nvidia. The Motley Fool has a disclosure policy.