Google Unveils Trillium Chip At I/O Event, Promising Breakthrough In AI Computing

Alphabet, Google’s parent company, has introduced its latest innovation: the Trillium chip. CEO Sundar Pichai, in a briefing call with reporters, hailed the Trillium chip as a significant leap forward in AI computing technology. 

According to Pichai, the demand for machine learning computing power has skyrocketed, growing by a factor of one million in the past six years, with Google well-positioned to meet this demand after over a decade of pioneering AI chip development.

The Trillium chip is touted as nearly five times faster than its predecessor and offers a remarkable 67 per cent improvement in energy efficiency. These performance gains put Google in direct competition with Nvidia, the current market leader in AI data center chips, which commands roughly 80 per cent of the market share. Google’s Tensor Processing Units (TPUs) have already secured a significant portion of the remaining 20 per cent of the market, thanks to their integration with Google's software and cloud computing platform.

Unlike Nvidia, which sells its chips directly, Google rents access to its custom chips through its cloud computing platform. This strategy has allowed Google to expand its reach and attract a diverse range of customers seeking powerful AI computing solutions. The Trillium chip, designed to power text and media generation from large models, is expected to be available to cloud customers in late 2024.

Google’s engineers have achieved additional performance gains by increasing the chip’s high-bandwidth memory capacity and overall bandwidth. These enhancements address crucial bottlenecks in AI model processing, further boosting performance. The Trillium chips are designed to be deployed in pods of 256 chips, providing scalability to meet diverse computational needs.

The Trillium chip is expected to hit the market later this year.