Today we're excited to introduce our new approach to confidential AI, the VAST DataEnclave. Okay, it's 2026. I think it's fair to say that we no longer need to question the intelligence, the automation, and the value that comes from AI models. Frontier model builders are pioneering new offerings that are nearing super intelligent levels. And the base weights that are used to create these models are some of the most valuable and protected datasets in the world. Enterprises are then taking these models and adapting them to their own unique requirements. These customized and fine-tuned weights have become the critical data assets that every enterprise and government must safeguard as their foundation to their AI-driven future. So as the value of AI continues to rise, so does the need to protect the most fundamental IP of these models. Here, we have a lot of work to do. Conventional approaches to encryption have been available long before the AI era, but accelerated computers and models that stretch across GPUs or even racks of GPUs introduce all new challenges. In this new world of accelerated computing, it requires the exact same standard of security, but with new approaches to protecting weights and other intermediate data as it moves into GPU memory and across GPUs over high-speed interconnects. And on top of that, the fact that the data is now in the open means that weights can be stolen or tampered by cloud administrators, exposed by compromised hypervisors, they can be accessed by malicious tenants, firmware exploits, or even physical attackers. As an industry, we have a gap. More importantly, this gap has introduced some very significant trade-offs that have limited the reach of AI, particularly in security-sensitive environments. For frontier model builders, the lack of data confidentiality means that their offerings have been constrained to a limited number of cloud platforms that these models can be deployed upon. And so many frontier models have no clear path to deploying their software into a customer's air-gapped data center. 50% of today's workloads and today's data resides on premises or in a customer's co-location facility. And that means that half of the world's data lives outside of where a frontier model can run. And the inverse is also true. If you think about an enterprise that has sensitive data sets, they can't easily ship this to some hosted AI service. Regulated industries and government agencies are legally restricted from exposing their data in certain environments. So many of them can't even compute outside their own sovereign or in certain cases air-gapped data centers. These constraints are just limiting the reach of artificial intelligence. 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. To start, VAST clusters will be able to create and deploy hardware isolated environments by taking confidential workloads and deploying them in secure virtual machines that leverage CPU level encryption. Now these VMs are very portable across most modern Linux environments and they leverage the trusted execution environments from Intel, AMD, and now NVIDIA's Vera data center processors. So we then bridge that security to NVIDIA's GPUs in confidential computing mode. And that allows us to protect both GPU memory and the traffic that goes across NVLink networks, thereby creating a secure and protected and multi-tenant enclave for your data. Now, a trusted runtime is only as good as its ability to enforce trust. And to build a platform that model builders and enterprises both trust, VAST clusters will support remote hardware attestation. This remote service works with the key vaults that weight owners maintain and allows them to validate the hardware platform that models run on as secure and tamper-proof. So this new cryptographic attestation layer helps in a few different scenarios. A model owner can host their own attestation service and that works with VAST clusters. And through some simple proxy service that can provide access into a secure air-gapped data center, we can verify hardware level encryption. Good, so now we can deploy models basically anywhere. Customers can also benefit from this approach. So let's suppose you have an enterprise customer that deploys VAST clusters in some AI cloud, maybe because they don't have liquid cooling and they want to use the newest processors. Whether they run open or closed weight models, all of the fine-tuning data and all of their other data can be processed in a secure enclave that is attested by and verified using the key vaults that our customers host and maintain in their own VPNs. So hosted environments can run open or closed weight models where the model builders have their IP protected, and the enterprise customers that they work with have trusted enclaves that protect their sensitive model weights as well as their sensitive data sets. Now, this solution works with multiple key management servers to ensure that all parties can enforce policy on their respective data sets. And if the platform can't prove it's trustworthy, whether the attestation fails or the service simply can't be reached, then the keys are never released. There's no attestation, there's no access period. Now as with everything we do, observability is a critical element baked into this new offering. Every action will be logged in a tamper-proof audit log that sits in the VAST DataBase. It's ready for high-speed queries to ensure simple, explainable, and governed AI. This new secure AI runtime, the VAST DataEnclave, unlocks the opportunity to deploy secure models and secure datasets in any public or private cloud data center. All while preserving the protections needed for IP owners to trust and verify the platforms that their AI is deployed upon. If you take a step back, it's important to recognize that this is more than a technology capability announcement. The reality is that VAST sits in the center of a very large ecosystem of model builders and platform providers. And today's news is an affirmation of VAST's big focus on partnership. And that is the part of the story that we are most excited to share today. We're bringing together a community of ecosystem partners that are joining us to ensure that intelligence is finally unbounded. Let's start with first the models. For years, VAST has been famous in our own little way for working with many of the world's largest AI model builders on the training and the inference infrastructure that they built. We've brought together several of these companies as early launch partners to get the data enclave off the ground, including Cohere, CrowdStrike, Deepgram, Factory AI, Fundamental, NVIDIA, and Twelve Labs. These AI pioneers will be launching their models into the data enclave to ensure a predictable, secure, and protected deployment platform. Now that platform of course needs hardware to run on. On this topic, when we explain what we're up to, so many of our server platform partners jumped at the idea that we could unlock access to leading AI models in the world's most secure data centers. And so to support this mission, we're happy to announce partnerships with Cisco and Supermicro. These organizations will join us as launch partners to deploy NVIDIA GPU reference architectures into secure enterprises and AI cloud data centers. Now speaking of AI clouds, it's also important to consider that not every data center is designed to support the computing machinery of tomorrow. For nearly half a decade, VAST has been instrumental in building out this new generation of AI clouds that so many model training initiatives and inference services are now built on. The VAST AI operating system powers a few million GPUs across this ecosystem of players, and their data centers are optimized for quickly deploying the newest generations of hardware of accelerated computers. That's why today's Data Enclave announcement aligns perfectly with the focus of our AI Cloud partners. And the secure attestation service we're enabling allows for model builders and enterprises to both deploy their model weights and data products in fully trusted environments where the builders and the customers both have full control over their core IP. So for scalable inference, customers can now leverage these NeoCloud environments to deploy accelerated computers, such as NVIDIA's Vera Rubin, in environments that they can't easily build themselves in their own data centers. Our cloud partners benefit by being able to offer new services and provide even more value to their enterprise and government customers. Today we're happy to announce a few of these cloud partners that are joining us for this initial launch, including Buzz, G42, NScale, and Sharon AI. So that's my news for today. And as you can probably tell, we're beyond excited to take this new step with our AI partners. And we look forward to removing the boundaries to trusted and secure computing for every customer.
