Flexential CIO Jason Carolan Unpacks the Data Center Industry's NVIDIA Moment

For this episode of the DCF Show podcast, we interview Jason Carolan, Chief Innovation Officer at data center operator Flexential. He’s a 25-year expert in the enterprise IT industry, with experience leading companies through technological evolutions like the one we’re experiencing right now.

Carolan believes there is a bigger story to uncover from the sheer dollar amount of Nvidia’s recent blockbuster valuation. In response to Nvidia’s market dominance in AI and data centers, Carolan wanted to discuss larger trends that may follow from this specific news moment.

According to Carolan:

“Nvidia's earnings results and forecasts for a continued AI boom doesn't come as too much of a surprise with the volume of businesses that are increasingly testing and utilizing the technology.

Nvidia's data center business is a combination of GPU and their network technologies, which further showcases the importance of high performance architectures that can support next generation AI demands. The company is currently forecasted to ship 4-5 times more GPUs this coming year – indicating another trend line with little competition in sight.

As inference matures, we will see more diversity in chip suppliers but that is a ways off. The bottom line is that, now with accelerating AI rollouts, companies will need more compute capacity, ultra-high bandwidth and very low latency in order to succeed.”

Here's a timeline of the podcast's highlights:

1:30 - Carolan explains that Flexential is a US-focused data center and interconnect operator with 41 data centers across 19 markets, and increasingly those workloads are becoming AI workloads

3:37 - Carolan discusses the diversification in the stack of CPUs and GPUs and the thermal demand for both doubling or tripling in the next five years, and how data center operators are trying to support that cooling load. Carolan says operators need to explore and develop a diversified portfolio of cooling solutions that can cater to different types of data center environments, considering the need for SKUs and staged cooling upgrades based on hall-specific requirements.

5:07 - The conversation turns to the recent Cisco announcement with Nvidia and the trend of "incorporating more ASIC level or FPGA level services, which goes back to Tensor," as explained by Carolan. There is also a discussion on the diversification between training and inference and the industry's focus on training.

10:45 - Data Center Frontier asks about competition for Nvidia in supplying GPUs, and wonders about their competitive advantage through their highly integrated platform and vertical integration.

16:02 - The discussion lights on speculation that the next generation of AI innovation will be around and through IoT and dealing with unstructured data growth, and as such, resources need to be moved closer to manufacturing and logistics.

20:57 - Carolan opines on how the data center industry can address PR problems by having an ESG strategy, reducing local impact through design elements and clean energy, engaging with the community and partners early on, and creating a flywheel effect of bringing jobs to the region.

24:43 - Carolan says enterprise deployment will involve either investing in highly dense rack-level architecture with liquid cooling or sliding HGXs into existing environments. Large language models require more data access than inference, which needs a local data repository.