Meet the Models Coming to Your Datacenter

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Meet the Models Coming to Your Datacenter · Channel: VAST Data · 04:11 · Watch on YouTube

Topics: AI agents, AI governance and regulation, AI infrastructure and hardware, Local and open-source models, Privacy and data protection

Summary

The video presents the VAST DataEnclave, a capability intended to bring encrypted AI model weights into a customer’s data center so models can work with sensitive data without moving it elsewhere. It frames confidential computing, attestation, and secure, on-premises or air-gapped deployment as ways to address customer concerns about privacy, security, and governance. The speakers describe a future in which agents support many stages of software development and business operations, while humans guide their governance and policies. VAST says the DataEnclave is planned to enable secure models and datasets in public or private cloud data centers.

Key points

Customers in sensitive or regulated contexts want AI that keeps proprietary information private.
The video emphasizes bringing compute to data rather than transmitting sensitive data to another location.
Confidential computing is presented as a way to bring encrypted model weights into a customer’s data center.
VAST previews the DataEnclave, a capability it says will ship over the next few months.
The speakers highlight attestation and access controls as important elements of trust.
The VAST AI Operating System and DataEnclave are described as supporting secure, air-gapped deployments.
The speakers envision agents being used throughout software development and across business and government.
Humans are described as retaining an important role in setting governance and policies for agents.
VAST says the announcement enables deployment of secure models and datasets in public or private cloud data centers.

Chapters

Start	Chapter	Summary
00:00	The need for private AI	Speakers describe the risks of using AI with proprietary data and customers’ demand for private, encrypted solutions.
01:04	Introducing the VAST DataEnclave	VAST previews the DataEnclave and explains how confidential computing can bring encrypted models to customer data without copying that data elsewhere.
02:06	Secure deployment and agent governance	The discussion turns to air-gapped deployments, agent use in regulated environments, and the human role in setting governance and policy.
03:09	Trust and the next phase of AI	Speakers emphasize trust and auditable engineering, then describe the DataEnclave’s opportunity to deploy secure models and datasets across data centers.

Quotes

You want to take the compute to the data. — 00:30

It’s called the VAST DataEnclave. — 01:04

I think for the first time, really, we don’t have to make this false choice between fast and safe. — 01:36

Humans become more relevant than ever by driving the governance and policy layer that agents are ultimately beholden to. — 02:38

Our mission is to make AI ubiquitous — 03:40

Concepts and entities

VAST Data, VAST DataEnclave, confidential computing, VAST AI Operating System, air-gapped, agents, regulated industries

Full transcript (original language)

00:00 Feels more comfortable. Sound speeds. Take one. The information is too proprietary and too confidential. This is a really challenging and complex problem. There’s a lot of customers that really need to have private AI. When you have agents taking actions inside of the code base, there’s a whole new world of risk that is introduced. The most important key piece is confidential computing.

00:30 Customers want to have an encrypted solution. Some countries, they want to make sure they host everything on-prem and nothing is on the cloud. The highest leverage is really around their most sensitive data. I want to bring the AI to where my data already sits. If you want these highly sensitive applications to have the option to say, “Hey, I don’t want to transmit my data.” You want to take the compute to the data. Customers are looking for models that they can keep their information private for themselves.

01:04 They don’t want to share their proprietary information with everyone else. Growing our state-of-the-art models and agent system that understand video holistically to where customers’ data live. VAST’s exabyte-scale data platform already lives where the customer is. Till today, the best AI models could not securely reason over sensitive data. And we want to change that. So today, we’re happy to preview a new capability that we’re shipping over the next few months. It’s called the VAST DataEnclave.

01:36 Confidential computing allows us to bring encrypted model weights into the data center. Right there, without being required to copy their data out of it, without needing a separate enclave. I think for the first time, really, we don’t have to make this false choice between fast and safe. The platform can’t prove it’s trustworthy. There’s no attestation, there’s no access, period. That control is critical in an era where AI is becoming more important than ever in the typical day-to-day activities of a business. As we expand our use of AI,

02:06 safety and security take center stage. Customers that are sitting on top of massive amounts of video content in regulated industries, they are frankly constrained. And a lot of that has to do around security and governance and having a completely secure stack. That allows you to deploy into these secure environments fully air-gapped. We can do it fast with local compute. We can do it safe with the VAST AI Operating System and the DataEnclave. A year from now, a regulated software team

02:38 will be able to use agents in every stage of their software development lifecycle. A company today has 1,000 employees, and their brains collectively represent the way that company operates, that same company will have a million agents. Most businesses and government agencies around the world today are looking for agentic solutions for highly secure, regulated, and highly governed. Humans become more relevant than ever by driving the governance and policy layer that agents are ultimately beholden to.

03:09 From prompt to usable insight and artifacts generation is now end-to-end automated process. And you guys are that ideal platform, right? To be able to bring these confidential enclaves for all the model vendors to bring their models securely to the edge. So when you think about what this ecosystem is, we obviously have a role to play. There’s a lot of trust that has to happen for customers to really not only get the full benefits of AI, but also build a stack that they can rely on. It’s trust built upon engineering excellence

03:40 that is auditable and bulletproof. I think this is the craziest time to be alive, to be honest. unlock, right, that is happening is fantastic. Truly a magical experience that our customers enjoy. Our mission is to make AI ubiquitous, to enable its widespread adoption and the development of more and more AI applications. Today’s DataEnclave announcement unlocks the opportunity to deploy secure models and secure data sets in any public or private cloud data center.

04:11 We are excited to build the next phase of this AI revolution, intelligence becomes unbounded.