Alphabet (GOOGL) built its artificial intelligence chips so Google would not have to depend entirely on Nvidia (NVDA). These chips have the potential to grow into a massive enterprise. Evercore ISI predicts $75 billion in 2027 and $113 billion in 2028 sales of Google’s TPUs to outside customers. As a result, technology that was initially intended to reduce Google’s own computing expenses would become one of Alphabet’s main new sources of income. Evercore analyst Georgia Anderson estimates TPU sales could represent roughly 35% of Google Cloud revenue in 2027, according to Investor’s Business Daily. She maintained an Outperform rating on Alphabet with a $450 price target. The significance is clear to investors. Google doesn’t need to outperform Nvidia in every area of AI computing. Alphabet can profit from the AI infrastructure boom that has made Nvidia one of the world’s most valuable companies by renting or selling Google’s chips. Additionally, Google controls the cloud where many of those processors will operate, in contrast to Nvidia. Google turned an internal cost saver into a product Google started creating TPUs more than 10 years ago because traditional processors became too costly for some of its biggest machine-learning tasks. At first, the chips were mostly developed for Google’s internal services. This tactic is evolving. Earlier in 2026, Alphabet announced that Google Cloud will begin providing TPU hardware to specific clients for use inside their own data centers. The company expected a small amount of revenue from those agreements in 2026, with the vast majority arriving in 2027. The opportunity is much increased by such action. In the past, customers who wanted Google chips had to use Google Cloud to consume them. Now, customers may purchase systems for their own facilities, rent TPUs via Google Cloud, or use them through infrastructure agreements, all of which allow Alphabet to generate revenue. For instance, Blackstone invested an initial $5 billion in a joint venture with Google that aims to launch 500 megawatts of TPU cloud capacity by 2027. Google will provide the chips, software and services. As a result, Google begins to resemble a true provider of AI infrastructure rather than just a cloud provider that also designs chips. Google’s AI chips are turning into a much bigger business. Bloomberg / Getty Images Nvidia finally has a different kind of competitor Google is not attempting to quickly replicate Nvidia’s operations. With a software environment that developers have spent years understanding, Nvidia continues to be the leading supplier of general-purpose AI accelerators. Google has a distinct edge. More Tech: Meta and Microsoft just sent employees a memo about Anthropic Microsoft cutting off popular software with no extension Oct. 13 Apple’s biggest AI advantage may face new privacy challenge The company controls the entire stack around its TPUs. Google creates the processors, runs massive data centers, controls Google Cloud, and creates Gemini models that use the infrastructure. Customers willing to optimize their software around Google’s framework may find TPUs more appealing. Evercore’s industry checks suggest TPUs can offer 20% to 40% lower costs for customers with experienced engineering teams, although software limitations and a shortage of specialized talent remain barriers to wider adoption. One client has already demonstrated the size that Google is pursuing. This year, Anthropic decided to increase the amount of Google TPUs it uses by several gigawatts. The extra capacity is anticipated to be online in 2027. The Claude developer will use both Google Cloud services and Google-built TPUs provided by Broadcom. It’s easy to overlook the importance. Any big AI firm needs a significant amount of processing power. Until recently, much of the investor discourse expected Nvidia to account for the majority of that expenditure. Google is showing that hyperscalers are becoming increasingly interested in the opportunity. Wall Street is starting to put a value on Google’s chips Evercore is not the only company seeing the emergence of a sizable enterprise. Piper Sandler estimates external TPU hardware sales could climb from roughly $8 billion in 2026 to $51 billion in 2027 and $104 billion in 2028. While BMO recently lifted its 2027 Google Cloud revenue prediction to $189 billion, citing improved cloud trends and the expanding semiconductor business, Wells Fargo has also increased its Alphabet forecasts due to external TPU sales. These estimations differ significantly, and investors should remember that none of these figures are certain. Google still has to produce the chips, get sufficient power for data centers, and persuade consumers that implementing its technology is worthwhile. Additionally, Nvidia has a significant installed edge. However, TPUs are important, even if Alphabet does not overtake Nvidia. Even with Evercore’s $75 billion projection, external chip sales would be a significant business inside one of the biggest corporations in the world. The economics go beyond chip sales, which is more significant. In addition to purchasing storage, networking, databases, cybersecurity, and artificial intelligence services from Google Cloud, a client drawn to Google’s TPUs may also purchase these services. This implies that a single chip transaction has the potential to increase expenditure inside Alphabet’s ecosystem. For many years, investors mainly saw Google as an advertising firm with a potential cloud business. But a third option is being made possible by artificial intelligence. Wall Street is beginning to place a lot of weight on Google, which may have developed one of the only reliable substitutes for the chips driving the AI growth. Alphabet’s next big venture may already be in its data centers if Evercore is even close. It was initially designed without external clients in mind. Related: Bank of America sees a $300 billion test for Google stock
CryptoAlerta — análise de criptomoedas e mercado em tempo real