Google’s chip business is on a faster track than Nvidia’s AI boom

Nvidia (NVDA) is investigating its own CEO. Google is building the chips that could take its customers.

Nvidia is down more than 5% premarket, trading at its lowest level since April 7, as its board examines CEO Jensen Huang’s relationship with an alleged Chinese spy. Alphabet (GOOGL) is up about 2%. It’s the latest chapter in Wall Street’s first $4.5 trillion CEO scandal.

Google on Monday introduced two new tensor processing units, or TPUs – its internally developed AI chips. The TPU 8tx and TPU 8i Max join the 8t and 8i unveiled in April, marking the first time Google has introduced four TPU systems in a single year. The two new chips will ship next year.

Think of it as a chip menu. The 8tx keeps data encrypted for governments, banks and other regulated businesses, paired with AMD (AMD) processors. The 8i Max packs two chips into one package for queries with long pop-ups – the amount of information an AI model can juggle at once.

The forecasts are breathtaking. Bloomberg Intelligence projects that TPU sales will grow from about $12 billion this year to more than $135 billion by 2028, not including chips leased through Google Cloud. That’s more than 11 times in two years.

Nvidia’s data center business grew eightfold in the first two years of the AI ​​boom, from $15 billion in fiscal 2023 to $115 billion in fiscal 2025.

Column chart comparing growth over two years. Nvidia's data center revenues grew nearly 8-fold, from $15 billion in fiscal 2023 to $115 billion in fiscal 2025. Estimated Google TPU sales grew more than 11-fold, from $12 billion in 2026 to $135 billion in 2028.
Nvidia, Bloomberg Intelligence, Yahoo Finance

The chips are already flying. Epoch AI estimates that the computing power of TPU shipments doubled in a single quarter this year, to around 800,000 chips.

Anthropic is the headliner. Its IPO filing shows about $161 billion in rents tied to TPUs, which Raymond James says makes it the largest custom chip customer of Google design partner Broadcom (AVGO). Meta (META), OpenAI and Oracle (ORCL) are also buyers.

Google was deploying approximately one data center TPU per year. Now there’s one for every task.

Column chart of data center TPU systems that Google announced each year from 2016 to 2026: one or no years most of the time, two in 2023, and four in 2026.
Google, Yahoo Finance

Nvidia still has the bigger engine.

Its next-generation Rubin chip is rated at 35 petaflops — a measure of raw computing speed — compared to about 10 for Google’s 8i and about 20 for the 8i Max, according to Yahoo Finance calculations.

Google sells on price. SemiAnalysis found that Google’s current Ironwood TPU delivers up to three times more AI output per dollar than Nvidia’s Blackwell chips when running AI agents, although Nvidia’s best system still wins when speed is paramount.

Bar graph of maximum AI computing power per chip in petaflops. Nvidia Rubin 35, Google TPU 8i Max 20, Nvidia Blackwell Ultra 15, Google TPU 8t 13, Google TPU 8i 10 and AMD MI355X 10.
Google, NVIDIA, AMD, Yahoo Finance

Forecasts aren’t sales, and Nvidia still sells far more AI chips than anyone else.

The next big moment for Nvidia will come on November 17, when it reports its third-quarter results, a day before a House-requested closed-door briefing on CEO Jensen Huang. Investors will want to know if customers are hedging their bets. As Stacy Rasgon of Bernstein wrote to clients: “Jensen is no ordinary CEO; it IS NVIDIA.”

In-house AI tools helped analyze interview transcripts, research, and write this article. Humans reviewed, edited and fact-checked the material before publication.

Jared Blikre is the global markets and data editor for Yahoo Finance. Follow him on @SPYJared or send him an email to jaredblikre@yahooinc.com.

Gn headline

Scroll to Top