Latest / Neural Newscast / HPE and Nvidia Build Agentic AI Factories [Model Behavior]
Transcript
- [00:00] Announcer: From Neural Newscast, this is Model Behavior, [00:02] Announcer: AI-focused news and analysis on the models shaping our world. [00:12] Nina Park: I'm Nina Park. Welcome to Model Behavior. This program examines how AI systems are built, deployed, [00:19] Nina Park: and managed within professional environments. Today is Tuesday, June 16th, 2026. We are observing [00:27] Nina Park: significant movement in the hardware sector today, specifically regarding [00:31] Nina Park: how enterprises are retooling [00:33] Nina Park: their infrastructure to accommodate the shift toward autonomous agents and agentic workflows. [00:38] Thatcher Collins: That's right, Nina. Silicon Angle is reporting from the HPE Discover event in Las Vegas where [00:44] Thatcher Collins: Hewlett-Packard Enterprise has expanded its private cloud AI portfolio. [00:49] Thatcher Collins: The centerpiece of this update is the integration of NVIDIA's new Vira chips. [00:53] Thatcher Collins: It appears the hardware layer is being re-engineered specifically to handle [00:57] Thatcher Collins: the high-frequency orchestration required by autonomous agents rather than standard inference. [01:03] Nina Park: The hardware specifics are notable. The new HP E Proliant Compute DL3944 Gen12 server slated for [01:11] Nina Park: early 2027 will be among the first to feature these NVIDIA Vira CPUs. These are not [01:18] Nina Park: general-purpose processors, they are purpose-built to facilitate the rapid tool calls [01:23] Nina Park: and multi-step reasoning typical of autonomous agents. HPE is also bundling [01:28] Nina Park: the Nvidia Agent Toolkit, which integrates Open Source Nemotron models [01:32] Nina Park: and the OpenShell secure runtime directly into the stack. Nina, I want to dig into [01:37] Nina Park: that private cloud focus. We frequently hear about agents operating in the public cloud [01:42] Nina Park: for the sake of scale. Why is HPE making such a significant bet on on-premises deployments [01:47] Thatcher Collins: for these [01:48] Nina Park: autonomous systems? [01:49] Thatcher Collins: Is this primarily a technical decision regarding latency for tool execution [01:54] Thatcher Collins: or is there a more fundamental concern regarding [01:56] Thatcher Collins: how these agents interact with sensitive corporate data? [01:59] Nina Park: It really comes down to compliance and security, Thatcher. In regulated sectors like finance, [02:05] Nina Park: the risk of data leakage in a public cloud remains a significant barrier. To address this, [02:10] Nina Park: HP is using Xerto software to create a resilient safety net. If an agent begins to deviate from [02:16] Nina Park: its parameters, the system can detect it and immediately revert the environment to a clean state. [02:22] Nina Park: They are also deploying [02:23] Nina Park: Nvidia Confidential Computing to ensure data remains encrypted throughout the process. [02:27] Thatcher Collins: Beyond specialized hardware, we are also seeing a significant capital injection [02:33] Thatcher Collins: into the broader compute market. HydraHost, a startup based in Colorado, announced a $100 million [02:40] Thatcher Collins: Series A round yesterday with participation from Nvidia. HydroHost is not building [02:46] Thatcher Collins: its own data centers instead they have created a marketplace and an operating system [02:51] Thatcher Collins: called Broker to aggregate idle GPU capacity from existing providers across the globe. [02:57] Nina Park: This serves as a strategic response to the global compute capacity crunch. [03:02] Nina Park: Instead of raising the billions required for physical infrastructure, HydroHost leverages software [03:08] Nina Park: to deliver GPU capacity in a flexible way. Their broker operating system is already operational in [03:14] Nina Park: over 50 data centers across the Americas, Asia Pacific, and Europe. They function as an [03:20] Nina Park: off-take network for frontier labs [03:22] Nina Park: and enterprises looking for performance GPUs outside of traditional hyperscalers. [03:28] Thatcher Collins: I have to offer some skepticism here, Nina. We recently saw Oracle stock price impacted [03:33] Thatcher Collins: because their capital expenditures on AI infrastructure exceeded expectations. [03:38] Thatcher Collins: If the industry giants are struggling with the costs of building these AI factories, [03:43] Thatcher Collins: can a distributed marketplace like HydraHost really guarantee the same reliability? [03:47] Thatcher Collins: Coordinating complex workloads across disparate data centers owned by third parties seems [03:52] Thatcher Collins: like it would introduce significant networking and latency hurdles. Those hurdles are [03:57] Thatcher Collins: the primary friction point thatcher. However, HydraHost argues that their distributed model [04:02] Thatcher Collins: increases hardware utilization in ways that rigid centralized data centers cannot. [04:08] Thatcher Collins: The fact that NVIDIA and Founders Fund are participating suggests a strong belief [04:12] Thatcher Collins: that the infrastructure pipeline needs to be more than just a few massive companies. [04:17] Thatcher Collins: It provides a necessary pressure valve for the industry as the demand for inference tokens [04:22] Thatcher Collins: continues to skyrocket. It highlights the two distinct directions the industry is taking. [04:28] Thatcher Collins: On one side, you have HPE and Nvidia building highly specialized secure on-premises environments [04:34] Thatcher Collins: for specific agentic tasks. On the other, HydraHost is attempting to commoditize [04:39] Thatcher Collins: and distribute the underlying compute power. Both models are chasing the same goal, [04:44] Thatcher Collins: making AI infrastructure more accessible [04:47] Thatcher Collins: and resilient as we move into the era of complex autonomous systems. [04:52] SPEAKER_03: It marks a fundamental shift in the AI stack, moving from general compute [04:57] SPEAKER_03: toward purpose-built architecture. Thank you for listening to Model Behavior, a Neural Newscast [05:04] SPEAKER_03: editorial segment. You can find more analysis and the full story list at mbnneuralnewscast.com. [05:14] SPEAKER_03: Neural Newscast is AI-assisted, human-reviewed. [05:18] SPEAKER_03: View our AI transparency policy at neuralnewscast.com. [05:23] Announcer: This has been Model Behavior on Neural Newscast, examining the systems behind the story.