Google is quietly building a serious challenge to Nvidia's dominance in artificial intelligence chips, and the tech world is starting to notice. While Nvidia still controls 81% of the AI chip market, Google's custom Tensor Processing Units (TPUs) have landed multibillion-dollar deals with Meta Platforms and Anthropic, signaling a major shift in how companies are sourcing the hardware that powers their AI systems.

What Are Google's TPUs and Why Do They Matter?

Google's Tensor Processing Units are custom-designed chips built specifically for artificial intelligence workloads, not general computing tasks. The company first deployed TPUs internally in 2015 to handle its own AI needs. Since then, Google has iterated through seven generations of the technology, with the latest version called Ironwood, announced in November 2025.

Ironwood represents a significant leap forward. Google claims the chip delivers a 4x performance increase compared to the previous generation for both training large AI models and running inference, which is the process of using a trained model to make predictions on new data. The company also offers Axion central processing units (CPUs), which are Arm-based processors that Google says deliver twice the price-to-performance value compared to chips built on the x86 architecture by Intel and Advanced Micro Devices.

What makes this particularly noteworthy is that Google has reportedly closed the performance gap with Nvidia's flagship Blackwell processors, the chips that have dominated the market for the past several years.

How Is Google Winning Major Customers Away From Nvidia?

• Anthropic Partnership: AI company Anthropic announced in October 2025 that it will purchase up to 1 million TPUs from Google to build 1 gigawatt of computing capacity in 2026, in a deal worth tens of billions of dollars. Anthropic specifically cited TPUs' "strong price-performance and efficiency" as reasons for the choice.
• Meta Platforms Deal: In February 2026, it emerged that Meta had signed a multibillion-dollar agreement to rent Google's TPUs for running AI workloads, marking a major win for Google in the hyperscaler market.
• Apple's Existing Use: Apple has already used Google's TPUs to train AI models for Apple Intelligence, and reports suggest the company is considering TPUs again to train a more advanced version of Siri.

These deals represent a fundamental shift. For years, companies building AI infrastructure had limited options; Nvidia's GPUs (graphics processing units) were the default choice. Now, major tech companies are actively choosing Google's alternative because it offers better economics and performance for their specific needs.

Alphabet CEO Sundar Pichai revealed last year that the company is witnessing "substantial demand" for AI infrastructure products, and the company believes TPUs will continue to see strong demand going forward.

Could Google Really Become the Second-Largest AI Chip Player?

According to DA Davidson analyst Gil Luria, if Google were to sell its TPUs to third parties at scale, the company could capture 20% of the AI chip market in the long run. That would make Google the second-largest player in the space, behind only Nvidia. Luria also estimated that TPUs could become a $900 billion business for Google over time.

To meet this anticipated demand, Google is taking aggressive steps. The company is diversifying its supply chain and has reportedly brought in Marvell Technology, a specialized chip designer, to help build more TPUs. This suggests Google is serious about scaling production to meet customer demand.

The broader context matters here. The global AI chip market is expected to reach $1 trillion in revenue by 2030, according to market projections. That's an enormous opportunity, and it's large enough that multiple companies can grow substantially even as Nvidia's market share declines.

What Does This Mean for Nvidia and the Broader Market?

Nvidia remains in a strong position despite the emerging competition. The company has estimated it will sell $1 trillion worth of chips based on its Blackwell and Vera Rubin architectures in 2026 and 2027 alone, dwarfing the $100 billion in AI chip revenue that Broadcom anticipates for the same period. Nvidia is also preparing for the next phase of AI, focusing on agentic applications, which are AI systems that can take independent actions to accomplish goals, and inference workloads.

However, Nvidia's dominance is no longer unquestioned. If Google captures even a fraction of the 20% market share that analysts predict, combined with gains by AMD, Broadcom, and other competitors, Nvidia's current 81% market share could shrink meaningfully over the next several years.

The real story here is not that Nvidia is in trouble, but that the AI chip market is maturing. Companies are no longer willing to accept a single-vendor solution. They want options, they want competition, and they want chips optimized for their specific workloads. Google's TPUs represent exactly that kind of alternative, and the fact that major companies like Meta and Anthropic are choosing them signals that the era of Nvidia's unchallenged dominance may be ending.