Generative AI is having far reaching impacts on datacenters and new data shedding light on the nature and extent of the impacts. Analysts Melissa Incera and Dan Thompson return to the podcast to dig into the data and discuss the ramifications for enterprises, datacenter builders and operators and those putting AI to work with host Eric Hanselman. The computational requirements of generative AI are unique, taking demands similar to high performance computing or cryptomining and spreading them across a vast new community of users. The demands for GPU-based capacity are consuming more power, sometimes with order of magnitude increases. It’s also driving retrofitting of existing datacenters. The need for higher capacity and innovations like liquid cooling are causing a new surge. We’re at the beginning of the shift from a focus on model training to greater amounts of inferencing, changing where capacity is needed. 

Looking at money flows in this market offers additional nuance. Investments in AI startups have been massive. Capital investments from hyperscalers in their own infrastructure is also massive, but it remains to be seen if they’ll pay off in equivalent amounts of revenue. Is all of this sustainable? It’s an complicated question in the many aspects of the word.

 

Host: Eric Hanselman

 

Guests:

Melissa Incera

Dan Thompson

 

Links to show content:

Webinar replay (registration required)

 

For 451 Research clients:

https://clients.451research.com/reports/204284

https://clients.451research.com/reports/204175

 

451 Research commissioned report highlights:

New Report from S&P Global Market Intelligence and Vultr Provides Unique Glimpse into Path to AI Maturity

Full report (registration required)