Google's "secret weapon" - TPU will support a $900 billion super track?

Investors are increasingly convinced that the artificial intelligence chips under Alphabet may become an important source of revenue for its parent company in the future. These custom chips, seen as Google's "secret weapon," known as Tensor Processing Units (TPUs), not only drive growth in its cloud computing business internally but also have the potential to open up a nearly trillion-dollar new market by selling to third parties.

Driven by this expectation, Alphabet's stock price surged 31% in the fourth quarter, making it the tenth best-performing component of the S&P 500 index. The market's optimism about the commercialization prospects of TPUs stems from a series of key developments in recent weeks. At the end of October, Alphabet announced it would provide AI startup Anthropic PBC with chips worth tens of billions of dollars, which led to a more than 6% increase in its stock price within two days.

A month later, The Information reported that Meta Platforms Inc. was negotiating to invest billions of dollars in TPUs, which again triggered a surge in stock prices. DA Davidson's head of technology research, Gil Luria, estimated that if Alphabet can seriously advance external sales of TPUs, it could capture 20% of the AI chip market in the coming years, making it a business worth about $900 billion.

Wall Street generally believes that TPUs are a core advantage for Alphabet in the fierce AI competition. Even though Alphabet has never sold a chip externally, more efficient self-developed chips mean better performance and lower costs for cloud services. However, once it begins to seriously sell TPUs to external customers, its business landscape will undergo a disruptive change.

Seeking Alternatives Beyond Nvidia

In the current AI chip market dominated by Nvidia, TPUs provide an attractive alternative for companies seeking supply chain diversification.

DA Davidson's head of technology research, Gil Luria, stated:

“If companies want to reduce their dependence on Nvidia, TPUs are a great option, which gives us plenty of reason to remain optimistic.”

TPUs are classified as "Application-Specific Integrated Circuits" (ASICs), specifically designed to accelerate machine learning workloads. This characteristic makes them less versatile than Nvidia's chips but also brings significant cost advantages. Mark Iong, a portfolio manager at Homestead Advisers, pointed out that in the current environment where investors are becoming increasingly cautious about AI-related spending, cost-effectiveness is a real advantage:

“Nvidia's chips are more expensive and harder to obtain, but if you can use ASIC chips, Alphabet is there and far ahead in the market.”

The Imagination Space of Nearly One Trillion Dollar Market Value

Analysts have begun to quantify the enormous commercial value that external sales of TPU could bring.

Gil Luria estimates that if Alphabet can seriously advance the external sales of TPU, it could capture 20% of the AI chip market in the coming years, making it a business worth approximately $900 billion.

Morgan Stanley analyst Brian Nowak has also observed a "nascent TPU sales strategy." He cited forecasts from the company's Asian semiconductor analysts, predicting that TPU procurement is expected to reach 5 million units by 2027, about 67% higher than previously expected; and 7 million units by 2028, 120% higher than previously expected. In a report to clients on December 1, Nowak wrote that while most of the demand may come from Alphabet's own use and Google Cloud Platform, this also "reveals the potential for Alphabet to sell more TPUs."

According to Morgan Stanley's estimates, every sale of 500,000 TPUs to third-party data centers could increase Alphabet's revenue by approximately $13 billion in 2027 and contribute 40 cents to its earnings per share. Based on analysts' forecasts, Alphabet's revenue in 2027 is expected to be around $447 billion, and an increase of $13 billion would mean nearly 3% sales growth. According to data compiled by Bloomberg, market consensus expectations for the company's 2027 revenue have been raised by more than 6% over the past three months.

Synergistic Effects of Full-Stack AI Layout

The value of TPU lies not only in its potential as an independent product but also in its deep synergy with Alphabet's entire AI ecosystem.

Google's recently released latest AI model, Gemini, has received high praise from the outside world. This model has been optimized to run efficiently on TPU, further confirming the intrinsic value of TPU. Mark Iong commented, "Alphabet is the only company that has a leading position at every layer of AI," pointing out that in multiple areas, from the Gemini model to Google Cloud and TPU, "this gives it an incredible advantage."

Nevertheless, it remains unclear how determined Alphabet is to sell chips to third parties on a large scale, but its internal advantages have already paved the way for its future business decisions.

Excessive expectations for the TPU business could also lead to disappointment for investors in the future, especially against the backdrop of Alphabet's significant stock price increase.

Currently, Alphabet's stock price is about 27 times expected earnings, the highest level since 2021, and well above its ten-year average. However, even so, its valuation is still lower compared to other large tech stocks like Apple, Microsoft, and Broadcom.

Allen Bond, a portfolio manager at Jensen Investment Management, recently took the opportunity of the stock price increase to reduce some holdings. However, he remains optimistic about the company's overall position and prospects, believing that "the path for TPU to become a revenue driver is credible." He stated:

"Alphabet is demonstrating real strength and progress in the AI field. Although this is increasingly recognized by investors, its valuation still appears reasonable considering growth expectations. A company with strong momentum in AI that is trading at a lower price than Microsoft and Apple means it remains a core holding."