Renen Hallak: AI Inference Puts Data at the Center
If you’d like to join me – and peers – for deeper conversations on innovation and leadership, get on this list for Fortt Knox Executive Communities, launching soon: mba.fortt.com.
This is an AI-assisted summary of my Fortt Knox Update with VAST Data CEO Renen Hallak. View the full interview here:
At CES in Las Vegas, Jon Fortt sits down with Renen Hallak, founder and CEO of VAST Data, to unpack why the AI era is rapidly shifting from compute-centric to data-centric infrastructure. Hallak explains how VAST’s early work with NVIDIA BlueField DPUs allows its software platform to run directly on ARM processors embedded in smart NICs—eliminating the need for x86 servers at key layers and enabling ultra-low-latency, always-on inference workloads.
The conversation draws a sharp contrast between AI training and inference. Training is episodic and checkpoint-driven; inference is production-critical, distributed, and must be highly available near end users. As AI agents emerge, Hallak argues, data systems must support “long-term memory” at massive scale, pushing storage and data orchestration to the center of the stack. VAST positions itself as the software “operating system” beneath next-generation AI applications, sitting atop the new Vera Rubin chips NVIDIA talked about at CES and integrating closely with model builders, neo-clouds, and enterprise IT.
Hallak also delivers a striking growth update: VAST’s annualized revenue growth has accelerated beyond 3x, making it one of the largest consumers of flash storage is an emerging NAND/SSD supply crunch. With data volumes exploding due to multimodal AI and production inference, flash, not GPUs or power, is becoming the next bottleneck. Enterprises, from banks to retailers, are moving decisively from experimentation into regulated, auditable production deployments—fueling demand that continues to outstrip supply.
“We are the plumbers. If what we do works, you shouldn’t know we even exist. But we’re enabling this next revolution and making it widespread. It’s always been new hardware that makes something possible, and then the software infrastructure layer that makes it easy for everybody to use. That’s the phase we’re at now with AI.”
Shift from compute-centric to data-centric AI infrastructure
VAST software now runs inside NVIDIA BlueField-4 DPUs (ARM-based)
Inference workloads demand always-on, low-latency, distributed storage
AI agents require persistent “memory,” driving massive data growth
VAST evolving into a full software infrastructure “operating system”
Tight partnerships across model builders, neo-clouds, and hardware OEMs
Revenue growth accelerating beyond triple-year-over-year
Flash/NAND emerging as the next major AI supply bottleneck
VAST’s efficiency reduces required physical storage capacity
Enterprises moving from AI pilots to regulated, auditable production
If you’d like to join me – and peers – for deeper conversations on innovation and leadership, get on this list for Fortt Knox Executive Communities, launching soon: mba.fortt.com.