The AI model is table stakes: Enterprise data is where the battle begins

Enterprises are prioritizing practical solutions for implementing artificial intelligence (AI) across their organizations. According to Nicola Tan, director of market development at Advanced Micro Devices Inc., the model itself is considered "table stakes" for enterprises, while the real value lies in their data. Junxia Zhou, senior product manager at Super Micro Computer Inc., and Mayank Gupta, director of solutions marketing at Nutanix Inc., emphasized during an exclusive broadcast on theCUBE that the integration of AI across an entire organization is the biggest challenge.

Zhou pointed out that pilots are designed for small data sets and a limited number of team members, but production requires the ability to support thousands of end users generating numerous inference requests. Additionally, factors such as token costs and fragmented data pose roadblocks to moving AI into production.

Supermicro and its partners have proposed a two-tier storage architecture to address these challenges. Clients can utilize flash memory (SSDs) to store active data sets and large-scale object storage for archival data. This approach balances high-performance data access with cost efficiency. Supermicro's focus is on optimizing the entire AI stack, as data access times to GPUs are becoming a critical aspect of AI pipelines.

Security and governance concerns have also emerged due to autonomous agents' extensive access to enterprise systems and data. Companies are seeking sovereign AI solutions to protect sensitive data and proprietary information, and to deploy models on-premises while connecting to hybrid and edge environments. Despite the benefits, Tan highlighted the importance of maintaining an "open strategy" that includes open-source solutions and standards, as well as flexibility to move between hybrid environments.