Nvidia has solidified its position as the dominant force in AI chip production, amassing a market cap of $4.5 trillion and supplying many of the world’s foremost tech companies with its highly sought-after graphics processing units (GPUs). Among its major clients is Google, which is ramping up its acquisition of Nvidia’s GPUs to meet the burgeoning demand for AI computational power across its cloud infrastructure. However, Google is evolving from being merely a customer to a significant player in chip development as well.
Recently, Google unveiled its most advanced chip to date, the Ironwood, set to be widely released in the coming weeks. This marks the seventh generation of Google’s Tensor Processing Unit (TPU), custom silicon that has been under development for over a decade. Designed for specific applications, TPUs are application-specific integrated circuits (ASICs) that excel in handling AI workloads, such as training extensive models and supporting real-time AI agents. Remarkably, the Ironwood chip boasts more than four times the performance of its predecessor, with AI startup Anthropic planning to deploy up to 1 million units to power its Claude model.
While the race to establish expansive data centers intensifies among major cloud providers, Google’s proprietary TPUs provide a competitive edge, allowing the company to stay ahead of the curve. Other tech giants, like Amazon and Microsoft, are also entering the arena but lag behind Google’s advances. Amazon Web Services launched its first AI chip, Inferentia, in 2019, followed by Trainium in 2022, while Microsoft’s Maia chip was only introduced at the end of 2023.
“Google is the only ASIC player deploying these chips at massive scales,” stated Stacy Rasgon, a semiconductor analyst at Bernstein, highlighting Google’s significant progress compared to its competitors. Originally built for internal use, Google has offered TPUs to cloud customers since 2018, signaling a shift in the competitive landscape that has even raised concerns at Nvidia. Following OpenAI’s initial cloud partnership with Google, Nvidia’s CEO, Jensen Huang, reportedly sought further discussions with OpenAI’s leadership.
Rather than selling its chips outright, Google provides access to TPUs as a cloud service, which has emerged as one of its key growth sectors. Alphabet, Google’s parent company, recently reported a 34% year-over-year increase in cloud revenue, reaching $15.15 billion, fueled by heightened enterprise demands for AI infrastructure. Despite not publicly disclosing the size of its TPU business, analysts suggest that Google’s TPU division and its DeepMind AI operations could be valued as high as $900 billion.
Customization remains a pivotal strength for Google. According to analysts, the efficiency of TPUs allows customers to achieve better performance relative to competitors. James Sanders, an analyst at Tech Insights, emphasized that Google is producing “tightly targeted” chips tailored to the workloads they expect. Meanwhile, Rasgon noted that as infrastructure expands, power may emerge as the primary bottleneck, not chip supply.
In a groundbreaking initiative, Google recently announced Project Suncatcher, which aims to utilize interconnected solar-powered satellites equipped with TPUs to harness solar energy. This initiative plans to launch two prototype satellites by early 2027, aiming to significantly reduce terrestrial resource demands while testing hardware in orbit.
Overall, Google’s expansion efforts continue to pay off as evidenced by substantial deals, including a multi-billion-dollar agreement with Anthropic for TPU capacity and a six-year contract with Meta valued over $10 billion. The rise in demand for AI infrastructure underscores the prevailing shift towards leveraging TPUs alongside existing solutions from Nvidia and Amazon, positioning Google as a formidable leader in the ever-evolving AI landscape.
Original Source: https://www.cnbc.com/2025/11/07/googles-decade-long-bet-on-tpus-companys-secret-weapon-in-ai-race.html
Category :
Tags:
Publish Date: 2025-11-08 10:17:00
