When artificial intelligence (AI) arrived on the scene a few years ago, it promised to change jobs forever. Companies were sold this idea that they could strike down the workforce, automate workflows, and increase their margins in just a few steps. 

Even before these could be rolled out to companies across the planet, the AI industry is already facing some major bottlenecks. Data centers needed to run AI aren’t straightforward to set up.

Energy to power them isn’t readily available, and there is much opposition to building large-scale data centers in various parts of the world. This is further pushing up AI pricing, which could effectively dull demand for the technology itself. 

With AI companies caught between challenges of scaling up and keeping prices low,  San Jose, California-based Frore Systems offers an opportunity to scale AI without actually building a large number of data centers.

Critical to this strategy are improvements in cooling technology used in data centers that can improve the efficiency of processing chips by as much as 35 percent. To understand how this works, we first need to understand what ails the AI industry at the moment: increasing costs of token generation. 

Tokens are the basic unit of AI output. Whether you ask an AI model a simple question or have it generate an image, summarize a large document, or process a spreadsheet, everything is first converted to tokens. 

The number of tokens needed by AI to complete a task depends on its complexity but also on the nature of the task.

Source link