Latest / Neural Newscast / HPE and Hydra Host Scale Enterprise AI Infrastructure [Model Behavior]
Transcript
- [00:00] Nina Park: From Neural Newscast, this is Model Behavior, [00:02] Nina Park: AI-focused news and analysis on the models shaping our world. [00:12] Nina Park: I'm Nina Park. Welcome to Model Behavior, a neural newscast editorial segment where we examine [00:18] Nina Park: how AI systems are built, deployed, [00:20] Nina Park: and operated in real professional environments. Today is June 16th, 2026. On [00:27] Announcer: this episode, we're looking at Hewlett-Packard Enterprise's massive expansion into on-premises [00:32] Announcer: AI infrastructure designed specifically for autonomous agents. alongside a major funding round [00:38] Announcer: for a startup attempting to democratize GPU access across the globe. [00:42] Thatcher Collins: Thanks Nina, the news from the HPE Discover event in Las Vegas today marks a significant pivot [00:48] Thatcher Collins: toward what they are calling the private cloud AI factory while much [00:52] Thatcher Collins: of the recent agentic AI discussion has focused on large scale cloud based models [00:57] Thatcher Collins: HPE is doubling down on the hardware and software needed to run thousands of autonomous agents [01:03] Thatcher Collins: behind a corporate firewall [01:05] Thatcher Collins: they are heavily leveraging their partnership with Nvidia to do it exactly. [01:09] Announcer: The centerpiece of this announcement is the new proliant compute dl3944 gen12 server [01:16] Announcer: It is expected in early 2027, and will be among the first systems to feature [01:20] Announcer: Nvidia's new Verrus CPUs. These chips are purpose-built for AI agents, [01:26] Announcer: specifically handling the rapid tool calls and complex orchestration that these systems require. [01:31] Announcer: "Thatcher, [01:32] Announcer: it is interesting to see them focusing so specifically on providing deterministic performance [01:37] Thatcher Collins: for agents." "It is, because agents are inherently more unpredictable than a standard chatbot. [01:42] Thatcher Collins: If an agent has the power to call tools or access sensitive data, latency [01:46] Thatcher Collins: and reliability become security issues. I noticed HPE is also integrating Zerto software here. [01:53] Thatcher Collins: The claim is that if an agent goes off the rails, the software can immediately detect it [01:58] Thatcher Collins: and rewind its actions [01:59] Thatcher Collins: to a clean state does that actually solve the trust problem for enterprise leaders Nina?. [02:04] Announcer: It is a bit of both by combining that with Nvidia Confidential Computing they create a chain [02:10] Announcer: of trust where data is protected even while it is unencrypted for processing they're also including [02:16] Announcer: the Nvidia Agent Toolkit which uses Nematron models and the Open Shell Secure Runtime. [02:21] Announcer: This essentially turns the server into an agentic operating system. For regulated industries, [02:27] Announcer: this move to keep everything [02:29] Announcer: on premises is about mitigating the risk of data leaks inherent in cloud-based inference. [02:34] Thatcher Collins: While HPE builds the physical factories, there is another side [02:38] Thatcher Collins: to this infrastructure puzzle — capacity. We saw news yesterday from HydraHost, [02:44] Thatcher Collins: a startup based in Boulder. They just raised $100 million in a Series A round led by [02:50] Thatcher Collins: Kindred Ventures, with participation from Kindred Ventures, Made By Nvidia, and ArcInvest. [02:56] Thatcher Collins: They aren't building data centers, Nina. They are building a marketplace to manage them. [03:02] Announcer: That is right, HydraHost is tackling the GPU crunch [03:06] Announcer: through their broker AI factory operating system. Instead of raising billions to build facilities [03:12] Announcer: as we saw recently with Oracle, they aggregate GPU supply from data centers globally. [03:18] Announcer: Their network is already in over 50 data centers across the Americas, Europe, and Asia Pacific. [03:25] Announcer: It is an off-take network meant to make renting GPUs more flexible for developers and enterprises. [03:31] Thatcher Collins: I want to push back a little on the marketplace model. We have seen these types [03:36] Thatcher Collins: of aggregators before. The challenge is often consistency and latency when you are pulling resources [03:42] Thatcher Collins: from a distributed network of different providers. HydraHost claims their operating system handles [03:48] Thatcher Collins: the provisioning and orchestration to make it seamless but can a distributed model [03:53] Thatcher Collins: really compete with the massive unified clusters being built by the hyper-scalers Nina. [03:59] Announcer: according to CEO Aaron Ginn the goal is higher utilization by aggregating supply [04:05] Announcer: they can offer capacity in a way that hyperscalers might find difficult to match in terms [04:10] Announcer: of flexibility they aren't just for small developers they're serving frontier labs [04:15] Announcer: and major inference platforms. The one hundred million dollars will specifically go toward expanding [04:20] Announcer: their data center footprint [04:22] Announcer: and growing the software team to convert megawatts into tokens more efficiently. [04:27] Thatcher Collins: It is a fascinating contrast on one hand you have HPE building the literal boxes for private [04:34] Thatcher Collins: on premises agentic work on the other HydraHost is trying to optimize the global pool of compute [04:40] Thatcher Collins: that is already out there both seem to be responding to the same pressure need the sheer capital [04:47] Thatcher Collins: and technical complexity required to make AI agents a functional reality in the enterprise. [04:52] Announcer: The infrastructure layer is clearly where the battle for agentic AI is being fought right now. [04:58] Announcer: Thank you for listening to Model Behavior, a Neural Newscast editorial segment. You can find [05:03] Announcer: more details and our full coverage at mb.neuralnewscast.com. Neural Newscast [05:09] Announcer: is AI-assisted, human-reviewed. View our AI transparency policy at neuralnewscast.com. [05:16] Nina Park: This has been Model Behavior on Neural Newscast. Examining the systems behind the story.