Google's decade-long bet on custom chips is turning into company's secret weapon in AI race

Nvidia has secured its position as the leading provider of artificial intelligence chips, catering to numerous major tech firms and achieving a staggering $4.5 trillion market cap. Google, a significant client of Nvidia, has been acquiring its graphics processing units (GPUs) to meet the soaring demand for AI computing power in the cloud. However, Google is evolving from being a mere consumer of Nvidia's technology to also becoming a formidable chip developer. On Thursday, Google unveiled its most advanced chip to date, named Ironwood, which is set to be widely available in the coming weeks. This marks the seventh iteration of Google's Tensor Processing Unit (TPU), a custom silicon that has been in development for over ten years. TPUs are specialized application-specific integrated circuits (ASICs) that play a pivotal role in AI by delivering highly efficient hardware tailored for specific tasks. Google asserts that Ironwood can handle demanding AI workloads—from training extensive models to operating real-time chatbots—and is over four times faster than its predecessor. Notably, AI startup Anthropic intends to deploy up to one million Ironwood chips for its Claude model. For Google, the advancement of TPUs provides a competitive advantage as major cloud service providers race to establish vast data centers, with the production of AI processors struggling to keep pace with demand. While other cloud companies are pursuing similar strategies, they lag significantly behind Google. Amazon Web Services (AWS) introduced its first cloud AI chip, Inferentia, in 2019, followed by Trainium three years later. Meanwhile, Microsoft announced its first custom AI chip, Maia, only at the end of 2023. Stacy Rasgon, a semiconductor analyst at Bernstein, noted, "Of the ASIC players, Google's the only one that's really deployed this stuff in huge volumes." He emphasized the challenges faced by other major players who require substantial time and resources to catch up. Initially developed for internal applications, Google's TPUs have been accessible to cloud clients since 2018. Recently, Nvidia expressed concern when OpenAI's cloud contract with Google was announced, prompting CEO Jensen Huang to engage in further discussions with OpenAI and its CEO, Sam Altman. Unlike Nvidia, which sells its chips as hardware, Google offers TPU access as a service through its cloud platform, which has become a significant growth area for the company. Alphabet, Google's parent company, reported a 34% increase in cloud revenue year-on-year, totaling $15.15 billion, surpassing analyst predictions. The company also reported a business backlog of $155 billion, indicating robust future growth potential. Google's CEO Sundar Pichai highlighted the significant demand for AI infrastructure products, including both TPU and GPU solutions. He stated, "It is one of the key drivers of our growth over the past year, and we expect strong demand to continue." While Google has not disclosed the exact size of its TPU business within its cloud segment, analysts from D.A. Davidson estimated in September that a division focused on TPUs and Google's DeepMind AI could be valued at around $900 billion, a notable increase from earlier estimates. Customization sets Google apart from competitors, as analysts point out the efficiency TPUs provide relative to other products. James Sanders from Tech Insights commented on the targeted nature of Google’s chips, designed for specific workloads. This efficiency is expected to become increasingly critical as infrastructure expands, with potential power supply becoming a major bottleneck. Additionally, Google launched Project Suncatcher, which investigates how a network of solar-powered satellites equipped with TPU AI chips could harness solar energy. The initiative plans to deploy two prototype satellites carrying TPUs by early 2027, paving the way for future computational advancements in space. In a significant move, Google recently expanded its agreement with AI competitor Anthropic, valued in the tens of billions, which is expected to bring over a gigawatt of AI compute capacity online by 2026. Google has also invested $3 billion in Anthropic, now serving as the core infrastructure provider for its next-generation Claude models. As competition intensifies, Google has secured a six-year cloud agreement with Meta, worth over $10 billion, although specifics regarding TPU use remain unclear. Furthermore, OpenAI is set to utilize Google Cloud while distancing itself from Microsoft, although they are not deploying GPUs at this stage. Alphabet's CFO Anat Ashkenazi attributed the growing momentum of Google's cloud services to rising enterprise demand for its comprehensive AI offerings. In the first nine months of 2025, Google signed more billion-dollar cloud contracts than in the previous two years combined. Amazon also reported robust growth in its cloud service, with AWS CEO Matt Garman noting that demand for the Trainium chip series is escalating. Google’s stock performance reflects investor confidence, soaring 38% in the third quarter, the best quarterly performance in two decades, while analysts remain optimistic about TPUs as a viable alternative to Nvidia’s offerings.