Volume 343, Issue 3 · 15 articles from 2 issues
BigQuery ObjectRef is now GA, letting teams analyze images, video, audio and documents alongside structured data in SQL.
Google Cloud has made BigQuery ObjectRef generally available.
Google says analyzing unstructured data such as images, videos, audio and documents has historically required separate pipelines and external ML workloads.
Teams then had to stitch inference outputs back into the data warehouse, creating operational overhead.
ObjectRef lets teams build multimodal data platforms and conversational analytics entirely in SQL within BigQuery.
At PyTorch Conference 2026, Google open-sourced TorchTPU, its native PyTorch stack for TPUs, along with new integrations.
At PyTorch Conference 2026, Google announced new products and integrations for running PyTorch on Google Cloud or in customers' own infrastructure.
TorchTPU, Google's native PyTorch stack for TPUs, is now open source on GitHub.
Google says TPUs now join GPUs and CPUs as a native PyTorch option.
Users can train, serve and deploy models on TPUs with tools such as vLLM, SGLang, llm-d, TorchTitan, Ray and Hugging Face.
Google Quantum AI shows magic state cultivation working on its Willow processor, cutting errors about 40x.
Google Quantum AI reports that magic state cultivation works in practice on its Willow processor, reducing errors roughly 40x compared with running without it.
The result, published in Science, targets a long-recognized bottleneck: producing the high-quality, error-corrected magic states needed for universal quantum computing, which has been considered extremely resource-heavy.
Google calls the demonstration a crucial step toward scaling quantum computers to practical use.
Google announces llm-d 1.0, an open-source Kubernetes-native stack for distributed LLM inference, released at PyTorch Conference.
Google says llm-d 1.0, an open-source, Kubernetes-native stack for distributed LLM inference, was released at PyTorch Conference.
The CNCF Sandbox project was co-founded by Google with Red Hat, IBM and others and has more than 500 contributors from over 25 organizations.
It offers cache-aware routing, multi-tenant traffic control, tiered KV cache and high availability.
Google says it is production-ready on Google Cloud TPUs and GPUs.
Google Cloud and RemotiveLabs describe how automakers can build full virtual vehicles in the cloud, from Android Automotive to MCUs.
Google Cloud and RemotiveLabs explain how automakers and Tier 1 suppliers can build complete virtual vehicles on Google Cloud.
The authors note that software-defined vehicles increasingly rely on cloud-native development, CI/CD and AI.
A gap remains between cloud-based development of high-level systems like Android Automotive OS and lower-level embedded controllers.
The post describes an approach that virtualizes the full vehicle, down to the smallest microcontroller.
Google launches Gemini 4 Argon, its new frontier model, via API, Antigravity, the Gemini app and AI Mode, at $2/$10 per million tokens.
Google introduced Gemini 4 Argon, its latest frontier model, now globally available to developers and enterprises through the API in AI Studio and the Gemini platform, to Google AI Ultra subscribers in Google Antigravity, and to Ultra and Pro subscribers in the Gemini app and AI Mode in Search.
Google says Argon excels at deep, multi-step work in coding, finance, legal, cybersecurity and science, ranking #1 on the Vals Index.
Introductory pricing is $2 per million input tokens and $10 per million output tokens.
VPC Service Controls gains folder-based perimeter membership and guided remediation to protect data from agents by default.
Google Cloud announced enhancements to VPC Service Controls aimed at easier perimeter creation and maintenance, including the ability to add folders to perimeter membership so contained projects inherit policy.
Admins can also manage granular exceptions and resolve denials with guided, scoped remediation.
Google frames the change around AI agents, a distinct class of non-human identity that requests data across projects at machine speed.
BigQuery's token-based cost management for generative AI functions is GA, with daily quotas to cap AI token spend.
BigQuery's generative AI functions, such as AI.GENERATE, AI.IF, AI.CLASSIFY and AI.SCORE, bring LLMs into SQL but are billed by the token, and Google has announced general availability of token-based cost management for them.
Four daily quotas, managed through the Google Cloud IAM quota system, let administrators set self-serve guardrails on token consumption.
The post walks through practical examples and resources for getting started.
Google Cloud's Agentspace and Vertex AI put agents into daily operations, with customers such as Telus, Suzano and Danfoss.
Analytics Insight describes how Google Cloud has built its strategy around agentic AI that plans and executes tasks.
Agentspace gives employees one place to search across Drive, SharePoint, Jira, Confluence, Salesforce and Box, including images, video and audio, and offers a no-code Agent Designer, role-based results and prebuilt connectors so smaller firms can build agents.
Vertex AI underpins custom development, alongside investment in AI infrastructure and shared protocols for agents to work across platforms.
Customers including Telus, Suzano and Danfoss report faster operations and time savings, though the article notes these figures come from Google itself.
Google Cloud explains how to build MCP Apps that render live, interactive UIs inside Gemini Enterprise chats.
Google Cloud explains how developers can implement MCP Apps in Gemini Enterprise.
MCP Apps (SEP-1865) is an open Model Context Protocol extension that pairs a tool with a live UI.
In Gemini Enterprise, that UI renders in a sandboxed iframe in the chat and can expand to fullscreen or a side panel.
Developers host their MCP server on a private Cloud Run service while Gemini supplies sign-in, governed data access and the Default Assistant.
Google Cloud customers can now select Anthropic's Claude Opus 5.5 and Claude Sonnet 5.5 in the Google Antigravity development platform.
Google says Claude Opus 5.5 and Claude Sonnet 5.5 are now available in Google Antigravity for Google Cloud customers.
Claude Opus 5.5 targets complex work, while Sonnet 5.5 is a faster, lower-cost complement.
Once an administrator enables them, developers can pick the models directly from the model picker.
Administrators can also enable Google's Gemini models, including Gemini 3.8 Flash and Gemini 4 Argon.
At AGNTCon + MCPCon, Google Cloud launches Agent Compute Engine and new agent governance capabilities derived from scaling Gemini.
Google Cloud says scaling Gemini taught it that securely running autonomous agents is fundamentally an infrastructure challenge.
At AGNTCon + MCPCon, the company launched Agent Compute Engine along with expanded agent governance capabilities.
The offerings are designed to support massive numbers of concurrent agent sessions with execution control and security.
Google says these are the same infrastructure capabilities behind Gemini for business, now delivered to customers.
Google details new Google Distributed Cloud air-gapped capabilities bringing frontier models and hardened security to sovereign environments.
Google says agentic systems are moving from pilots to production, but organizations with strict sovereignty, defense or critical infrastructure mandates often cannot use public cloud.
Building an AI stack inside a disconnected perimeter can take months.
The latest innovations in Google Distributed Cloud air-gapped aim to deliver frontier models, hardened security and high performance where sensitive data lives.
Google calls the update a major leap forward for these organizations.
Confidential Computing on Google Cloud G4 instances with NVIDIA RTX PRO 6000 Blackwell GPUs is now generally available.
Google Cloud has made Confidential Computing generally available on its G4 instances powered by NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs.
Google says production AI agents process sensitive data and valuable IP in real time, which requires protection while data is in use.
The release integrates NVIDIA Confidential Computing directly into G4 instances.
It is designed to protect proprietary model weights, user prompts and generated outputs.
Google Cloud adds OpenAI models to its Model Garden, saying it is now the only hyperscaler offering all three leading closed LLMs.
Google Cloud is launching OpenAI models in its Model Garden.
Google says it is now the only hyperscale cloud provider offering access to all three leading closed-source LLMs.
These sit alongside more than 200 first-party, third-party and open models on a unified enterprise AI platform.
The company positions the move as giving customers flexibility to balance cost, capability and performance for agentic workloads.