Google DeepMind and Google Cloud collaborate on private AI compute for Gemini models

Amazing work from Wei Huang, Tiziano Santoro, and team graduating this groundbreaking research project. Private inference tech like this will be key to making privacy-preserving and safe AI possible and is an important step to democratizing AI at the global scale.

Like what Apple did last year with Apple Private Cloud Compute, Google today also announced their Private AI Compute. Private cloud compoute instances extend on-device AI capabilities to the cloud while retaining their privacy assurance. https://lnkd.in/g65bW-TY

As a pioneer in responsible AI and privacy-preserving technologies, we’re taking our next step with Private AI Compute. Private AI Compute extends on-device privacy assurances to the cloud, while unlocking our most capable Gemini models to bring you faster and more helpful AI experiences. Most importantly, it’s built on one integrated Google tech stack, powered by our custom TPUs and Titanium Intelligence Enclaves (TIE). This makes this technology and the experiences it can enable unique to Google. This advancement is the result of a collaboration across Google DeepMind, Google Cloud, Google Research, Android, and numerous AI and security teams across Google that have all been focused on scaling helpful AI, responsibly. Read more about how Private AI Compute advances AI privacy: https://lnkd.in/guhB3gar

Private AI Compute is a cutting-edge solution designed to deliver powerful artificial intelligence capabilities through cloud technology while ensuring data privacy for users. By leveraging advanced computing techniques, this approach enables individuals and organizations to benefit from AI tools without compromising sensitive information. The system prioritizes user confidentiality, allowing for secure and efficient processing of data in a private environment. This innovative framework aims to enhance the accessibility of AI applications while fostering trust and security among users. With Private AI Compute, organizations can harness the potential of AI to drive insights and decision-making, all while maintaining control over their data integrity and privacy. This solution represents a significant advancement in balancing AI power and user privacy in cloud computing.

A Fortune 100 CISO told me they discovered a vulnerable open-source ML model was being used in their org... from an engineer's LinkedIn post. This is the reality of "AI Sprawl." Security teams are flying blind. AI assets are fragmented across code, cloud, containers and 3rd party apps. Visibility is lagging far behind adoption. CISOs and security leaders don’t have a clear and accurate answer to “what AI is running in my enterprise?” You can't secure what you can't see. That's why we have built the AI Asset Inventory in Repello platform. It's the single source of truth to: 🔍 Discover every AI asset, from models and agents to datasets. 🗺️ Catalogue them across your entire environment (code, cloud, infra). 🔗 Trace the relationships between all components. Now you can answer “what AI is running in my enterprise?” confidently, with the right insights. Stop finding out about risks when it's too late. Get complete visibility. See it in action -> https://lnkd.in/gHzCPheq

Google introduced Private Compute AI, a new cloud-based platform that enables its Gemini AI models to handle complex tasks for users while maintaining on-device-level privacy. The system ensures that personal data remains inaccessible even to Google itself. It achieves this through a protected execution environment built on custom TPUs and Titanium Intelligence Enclaves, where data is transmitted over encrypted links, isolated by hardware-level controls, and processed within ephemeral virtual machines that are destroyed after each session. The design creates a secure bridge between device and cloud, powering features such as context-based suggestions and multi-language transcription on the Pixel 10, while ensuring that user data is never stored, reused, or exposed.

Any application, any model, any hardware, any cloud. Learn more. Red Hat has launched its strategic AI 3.0 platform, designed to be the foundational "Linux of Enterprise AI." The goal is clear: provide a unified platform for Any Model, Any Hardware, Any Cloud, ensuring stability and control. And Red Hat’s new AI strategy is transforming enterprise IT by enabling GPU-as-a-Service (GPUaaS) and Model-as-a-Service (MaaS).  Read how Red Hat is solving complexity and cost for mission-critical AI workloads.

…key message – 5 (AI Infrastructure) Google just redefined trust in the cloud. Their new “private AI compute” service built on AMD-based Trusted Execution Environments claims that cloud-processing can be as secure as local processing. Here’s the kicker: even the data-centre operator (or Google itself) can’t access your data or decrypt your memory while it’s being processed. The host separation means your workload executes in a hardware-rooted enclave with isolation from the infrastructure. This is a mental shift for private cloud and sovereign cloud regulated models, compute you can’t physically access, but you can trust. 🔗 https://lnkd.in/d_ujAk_H #strategicadvice #nofilter #GoogleCloud #PrivateCloud #SovereignCloud #Endava #AMD

This Month in Datadog we Spotlight the new Instance Explorer, where you can compare the cost and performance of instances across AWS, Azure, and Google Cloud. We also cover ensuring teams are informed during outages with External Provider Status and Status Pages, ingesting OTLP metrics via the OTLP Metrics API, how we combined our security and SRE teams, surpassing 1,000 Datadog integrations, and much much more. Watch now! https://lnkd.in/edch8XpD

Google has introduced Private AI Compute, a cloud-based processing system that aims to enhance AI capabilities while ensuring data privacy. This platform integrates Google’s advanced Gemini models with robust privacy measures, reflecting a commitment to responsible AI development. Similar to Apple’s Private Cloud Compute, it addresses the challenge of balancing computational power with user privacy expectations. Private AI Compute allows for faster, more personalized AI experiences without compromising sensitive data, which remains inaccessible even to Google engineers. Utilizing a secure infrastructure powered by custom Tensor Processing Units and Titanium Intelligence Enclaves, the system ensures data protection through encrypted connections and zero access assurance. Enhancements in AI functionalities, such as improved suggestions and transcription capabilities, signal a new era of AI tools that prioritize user control and transparency in data usage.