With an eye on a trend in artificial intelligence processing, Google () announced on Wednesday a chip dedicated to AI inference. Although the new model signal for lower processing costs, the company continues with foot on infrastructure investments.
Also read:
“We are making great investments for now and for the future. In 2025, our capex It will be $ 75 billion, ”said Google CEO, Sundar Pichai, at the opening of the Google Cloud Next 25 ′ event in Las Vegas.
Named Ironwood, the company’s new inference chip is the first TPU, a tensor processing unit, developed specifically for this purpose.
Inference is the process by which already trained intelligence models react to commands. They are chips dedicated to the speed of answers to the questions asked by users of an AI chat or an action requested from an assistant.
Also read:
Continues after advertising
Google Cloud customers will be able to hire two different computing capabilities based on new models: 256 chips and 9,216 chips. The company states that the new model has significant performance gains, more than 3,000 times higher than the company developed by the company in 2018, with an energy efficiency of 29 times.
Bold investments in infrastructure for data centers They have been one of the main ones at a time when they try to expand their.
For the company, the new chip is a step towards what has been called the “Inference Age”, in which PENAORATIVE agents will react proactively and create data to deliver answers and ideas collaboratively. Morgan Stanley analysts predict that 75% of computational demand and energy centers energy in the coming years will be centered on inference.
Continues after advertising
Also read:
This is what has motivated companies like Google Cloud itself and competitors such as AWS, from Amazon, to develop their own versions of inference chips, a market in which Nvidia has been dominant.
*The reporter traveled to Las Vegas invitation from Google Cloud
