Google is making a concerted effort to expand its presence in the artificial intelligence (AI) chip market, taking a page from Nvidia’s playbook in an industry long dominated by the graphics processing unit (GPU) leader. The Alphabet-owned company is leveraging its financial resources and technical expertise to offer its own tensor processing units (TPUs) as an alternative to Nvidia’s hardware, seeking to capture a larger share of the rapidly growing AI computing market.

Central to Google’s strategy is an AI data center cluster called Lake Mariner, located near Niagara Falls in western New York. Google has backed the project with a $3.2 billion financial guarantee, enabling developers to rent computing power from thousands of TPUs. One major customer of this facility is Anthropic, an AI developer aligned with Google, illustrating its efforts to build partnerships akin to Nvidia’s ties with OpenAI. Sources indicate the approach mirrors Nvidia’s model of providing financial guarantees that help data centers secure cheaper financing, with some of that capital cycling back through chip purchases.

Until recently, Nvidia has maintained a near-monopoly on AI chips, with an estimated market share exceeding 90 percent. Its GPUs have been favored for training and running AI models, aided by an ecosystem including proprietary hardware connectors and its CUDA programming library. Nvidia CEO Jensen Huang has publicly downplayed Google’s challenge, asserting Nvidia’s broader market reach and skeptical of TPUs’ cost advantages. “Our market reach is far greater than any TPU or ASIC can possibly have,” Huang said during a podcast in April.

Despite this, Google’s Cloud unit has intensified its chip-related efforts amid recent leadership changes and increasing demand for AI compute. In May, Google announced it would sell TPUs directly to customers and introduced its first TPU designed specifically for AI inference workloads, which applies to real-time query servicing. Mark Lohmeyer, vice president of AI and computing infrastructure at Google Cloud, noted that these advancements are attracting customers who had previously not considered TPUs, citing Citadel Securities as an example. Citadel’s chief technology officer, Josh Woods, reported that TPUs have enabled some workloads to run up to four times faster and at 30 percent lower cost.

The intensifying competition has also prompted Google to partner with Blackstone on a $5 billion cloud-services venture aimed at competing with Nvidia-backed providers like CoreWeave and Nebius, which offer cloud platforms primarily built around Nvidia technology.

Industry analysts see Google’s moves as a response to unprecedented demand for AI computing power that no single supplier can fully satisfy. Stacy Rasgon of Bernstein highlighted Google’s more aggressive monetization of its hardware capabilities as a departure from its prior approach.

Nevertheless, Nvidia remains a formidable competitor with entrenched customer loyalty and a comprehensive hardware and software stack. Some so-called “neo-clouds” hesitate to diversify beyond Nvidia, concerned about losing access to its chip allocations if they deviate—a phenomenon referred to as “Jensen jail” by venture investor Adam Fisher. Huang has stated publicly that Nvidia welcomes customers purchasing its products either fully or selectively.

Google’s push into the AI chip market signals a broader shift in cloud and AI infrastructure, as well-capitalized firms vie for leadership in what many see as a critical element of future technology development. Whether Google can break Nvidia’s stronghold remains an open question as both companies scale their offerings to meet escalating market demands.