
Episode 102 - The training to inference passageway with Sumti Jairath, SambaNova
From DCD Zero Downtime: The Bi-Weekly Data Center Show by DatacenterDynamics
April 30, 2026 · 42 min
About this episode
This episode features an interview with Sumti Jairath discussing the shift in the AI industry from training to inference and the challenges in the semiconductor sector.
With the AI industry fueling a shift away from training and towards inference, and claiming agentic AI as its next frontier, the need for hardware to support workloads that require high-compute and low-latency performance has never been greater. On this episode of Zero Downtime, DCD talks to Sumti Jairath, chief architect at SambaNova, about some of these driving forces, how the semiconductor industry has changed over the last decade, and what some of the biggest challenges facing chip companies right now are
People in this episode
Host: DatacenterDynamics
Guest: Sumti Jairath
Topics covered
- AI industry
- inference
- hardware performance
- semiconductor industry
- chip challenges
Keywords
- AI
- inference
- hardware
- semiconductors
- chip companies
- low-latency
- high-compute
Mentioned in this episode
Organizations: SambaNova
More episodes of DCD Zero Downtime: The Bi-Weekly Data Center Show
- Episode 103 - Networking at hyperscale with Pradeep Vincent, Oracle Cloud Infrastructure · May 14, 2026 · 29 min
- Bonus episode - The telco build out with Patrick Halley, Wireless Infrastructure Association · May 7, 2026 · 37 min
- Episode 101 - Unpacking clean energy procurement with Nayel Brihi, BloombergNEF · April 16, 2026 · 30 min
- Episode 100 - It's good being Mr Wonderful, with Kevin O'Leary · April 2, 2026 · 31 min
- Episode 99 - A hyperscale competition with Damien Lucas, Scaleway · March 19, 2026 · 29 min
- Bonus episode - Can enhanced geothermal power the AI boom? With Cindy Taff, Sage Geosystems · March 12, 2026 · 28 min
Explore listener stats, chart rankings, contacts and more on the DCD Zero Downtime: The Bi-Weekly Data Center Show podcast page.