On Wednesday, AI cloud service provider CoreWeave, Inc. (CRWV.US) announced two products at NVIDIA's (NVDA.US) Berlin GTC event — Physical AI Launchables for robotics and autonomous driving, and a Zero Trust AI Framework for the agentic AI era. Following the announcement, CoreWeave's pre-market shares rose more than 2%, providing temporary relief after a sharp pullback of over 25% last week.
These two launches may point to a simultaneous strategic intent: CoreWeave is attempting to extend itself from a pure "compute provider" toward the physical AI engineering layer and security control layer, positioning itself closer to customer workflows within the NVIDIA ecosystem.
Physical AI Launchables: Compressing "the pain before running" into a single command
Physical AI Launchables is not positioned as a new compute product or model tool, but rather a set of workload templates built and tested by CoreWeave engineers. Robotics, autonomous driving, and simulation teams can launch tasks with a single command, without needing to rebuild environments, configure storage, or wait for schedulers to be ready before each run. Richard Ahlfeld, CoreWeave's Senior Vice President of Physical AI, said: "Robotics and autonomous driving teams are doing some of the most demanding engineering work in this industry, and they should spend their time on that. Physical AI Launchables provide a tested starting point for the tools they are already using."
The first batch of templates covers five tools: NVIDIA Isaac Sim, CARLA, MuJoCo, NVIDIA AlpaSim, and NVIDIA Cosmos 3, spanning three key stages of the physical AI loop: data generation, evaluation and scoring, and policy training. Additionally, Launchables integrates agentic skills from Cosmos Data Factory for generative data augmentation of video, images, and text, automatic labeling, data curation and retrieval; each template also comes with optimized skills for coding agents such as Claude Code and Codex, enabling agents not only to "launch a run" but also to read run outputs and determine whether results are useful.
Notably, each tool also comes with a blank template, so teams are not limited to CoreWeave's pre-packaged workloads and can point to their own scenarios and data on the tested foundation. Nakul Arora, NVIDIA's Director of Product Management, said that NVIDIA's reference agentic skill designs for physical AI are designed to run on any cloud, while CoreWeave has built its own version optimized for its AI cloud, enabling these workflows to "run natively on its infrastructure."
Zero Trust AI Framework: Six layers of protection, keeping CoreWeave "outside the trust chain"
The Zero Trust AI Framework, also announced the same day, directly addresses the security paradigm shift brought by agentic AI. When AI agents retrieve context across multiple systems, invoke tools, and trigger actions, sensitive data no longer stays within a single application boundary but travels across prompts, embeddings, model weights, execution traces, and outputs. In traditional cloud models, customers have had to extend trust to cloud service providers — with providers managing encryption keys, holding privileged server access, and even operating proprietary models that may learn from customer interactions. Chen Goldberg, CoreWeave's Executive Vice President of Product and Engineering, said: "Security for AI workloads cannot depend on a provider's promise about its own behavior. The Zero Trust AI Framework replaces assurances with architecture, keeping critical controls in customers' hands."
According to the company, the framework consists of the following six layers of protection: Zero Key Access: through remote key encryption, customers continue to use their existing key infrastructure — whether software key managers, hardware security modules, or KMS deployed on other clouds — with key control never transferring; Zero Node Access: support access management makes provider access an explicit customer decision, with customers authorizing access within defined scopes; Zero Trust Infrastructure: establishing verifiable trust at the infrastructure layer based on hardware root of trust; Zero In-Use Access: plans to extend NVIDIA's third-generation confidential computing to CPU and GPU, covering data-in-use states, with related capabilities arriving with HGX Rubin NVL8 in the fourth quarter; Zero Model Data: CoreWeave explicitly does not train its own models on customer data; Zero Trust Network Encryption: ensuring inter-node communication through MACsec backbone encryption.
The overall design goal is summarized by CoreWeave as "keeping itself outside the customer's trust chain" — security no longer depends on provider promises but is delivered through customer-held keys, access authorization, and encryption architecture. Additionally, CoreWeave stated it intends to use NVIDIA's NVLink interconnect technology, including for its confidential computing efforts with NVIDIA HGX Rubin NVL8, which will launch in the fourth quarter.
Stock backdrop: Rebound after deep pullback, divergence remains
The two launches come at a moment of dramatic stock volatility for CoreWeave. Over the past week and a half, the stock has fallen more than 20% cumulatively, triggered by concerns over its approximately $50 billion debt burden, credit default swap spreads briefly touching distressed levels, and Tesla CEO Musk's "yikes" comment on social media amplifying market sentiment. However, Wall Street analysts have not formed a consensus bearish view: Bank of America reiterated its "Buy" rating and $140 price target the day before the announcement, arguing that CoreWeave's core advantage lies in "bringing active power capacity online at scale and speed," and that its five-year major customer contract structure provides high revenue visibility. In contrast, HSBC maintained a "Reduce" rating and $39 price target, citing the company's projected average return on invested capital of approximately 8.9% from 2026 to 2030, which is below its borrowing cost, meaning "the more it invests, the more value it destroys." According to LSEG data, of the 45 analysts covering CoreWeave, 32 still give "Buy" or "Strong Buy" ratings, with HSBC's view being in the minority.
Against this backdrop, the simultaneous release of Physical AI Launchables and the Zero Trust AI Framework provides CoreWeave with new material that goes beyond the "debt and compute" narrative: the former points to competition for the entry point of physical AI workflows, while the latter addresses the most sensitive security control concerns for enterprise customers in scaling agent deployments. Whether these two products can translate into verifiable revenue and customer retention will likely become a core focus for the market to test in the future.
Comments