Google Cloud is rapidly expanding the Gemini developer ecosystem across several fronts: an agentic Gemini CLI (with custom slash commands and an extensions ecosystem) that now integrates directly with Google Data Cloud services, new Gemini CLI extensions for Cloud Run and security analysis, Gemini Cloud Assist additions (including investigations and root-cause analysis for Dataproc / Serverless for Apache Spark), and the Multimodal Live API tutorial and tooling that enable real‑time, streaming multimodal QA and inspection workflows — with Gemini models also pushed to hybrid/on‑prem via Google Distributed Cloud. (cloud.google.com)

Together these moves shift developer AI from isolated lab demos to integrated, end‑to‑end developer and ops workflows: CLI agents automate local dev tasks, extensions bridge terminals to BigQuery/AlloyDB/Cloud Run and automated security scans, Cloud Assist reduces mean‑time‑to‑resolution with root‑cause analysis for distributed jobs, and Multimodal Live enables real‑time streaming use cases (QA/inspection). The net effect is faster iteration, tighter cloud integration (Vertex AI / Google Data Cloud), and new governance/sovereignty options via on‑prem Gemini — but they also raise operational, security, and reliability questions as adoption scales. (cloud.google.com)

Google / Google Cloud (Gemini, Vertex AI, Google Distributed Cloud) is the central platform and vendor; developer relations and product leads (e.g., Taylor Mullen — Gemini CLI creator; Sujatha Mandava — Databases PM; Prithpal Bhogill — Group PM) are driving design and rollout; GitHub and the open‑source community are co‑players (CLI extensions, GitHub Actions); and enterprise customers and developer communities (discussed by Keith Ballinger and other Agent Factory guests) are early adopters and critics. (cloud.google.com)

Over the last two months Google Cloud has deeply integrated generative AI into BigQuery: BigQuery ML now supports Google’s Gemini embeddings plus over 13,000 open‑source embedding models and new T‑VF/rowwise functions for in‑place LLM inference, BigQuery throughput and reliability for gen‑AI jobs was dramatically increased (e.g., >100x text LLM throughput and >30x embedding throughput for first‑party models with provisions for Vertex AI Provisioned Throughput), new BigQuery & BigQuery ML operators for Vertex AI Pipelines were released to orchestrate training/inference, and complementary platform features landed such as Earth Engine raster analytics in BigQuery geospatial, BigLake GA for unified lakehouse access, CMETA and short‑query runtime optimizations, soft failover for Managed DR, and agent toolsets (ADK/MCP) to safely expose BigQuery to agentic apps. (cloud.google.com)

This wave of releases effectively brings model inference and embedding generation to the data warehouse (enabling semantic search, RAG, classification and agent tooling without large data movement), raises BigQuery from analytics engine to an AI inference platform at warehouse scale (hundreds of millions of rows per 6‑hour job with higher Vertex quotas), and reduces operational complexity by adding pipeline operators, default connections, global endpoints and disaster‑recovery controls — while also surfacing new governance, cost, and vendor‑choice considerations (Gemini vs OSS embeddings, Provisioned Throughput vs shared quotas). (cloud.google.com)

Google Cloud (BigQuery, BigQuery ML, BigLake, Vertex AI, Dataplex, Earth Engine) is the primary actor driving integration and platform changes; Gemini (Google’s family of models) is positioned as a first‑party model option inside BigQuery; open‑source embedding model providers (13K+ OSS models surfaced) are explicitly supported; customers and partners (e.g., Faraday cited for large embedding jobs) and third parties such as Databricks (strategic partner / ecosystem player) and the ADK/MCP community (agent/tooling ecosystem) are important stakeholders. (cloud.google.com)

Google Cloud is consolidating a production-ready stack for agentic AI on Vertex AI: Agent Builder (now surfaced as AI Applications / Vertex AI Agent Builder) provides a unified developer experience (Agent Development Kit, Agent Garden, Agent Engine) to design, ground, orchestrate and deploy multi-agent systems at enterprise scale; complementary platform updates — Model Garden with 200+ validated open models and improved multi-region/global endpoints for third‑party models like Anthropic’s Claude — plus infra/workflow integrations (BigQuery / BigQuery ML operators for Vertex AI Pipelines) and a new open commerce spec, the Agent Payments Protocol (AP2), together form an ecosystem intended to move agents from prototypes to production. (cloud.google.com)

This matters because enterprises now have a vendor-backed, end-to-end path to build, ground, govern, and scale agentic applications (design → tool/data connectors → runtime → observability → billing/commerce). The platform moves several risk/cost barriers (model discovery & infra, multi-region availability, data grounding, pipeline automation, and interoperable agent-to-agent commerce) toward standardized solutions — which accelerates adoption but raises new operational, security, and liability questions that industry stakeholders (cloud providers, payments networks, banks, retailers, and standards bodies) must address together. (cloud.google.com)

Primary players are Google / Google Cloud (Vertex AI, Gemini, Agent Builder/ADK/Agent Engine, Model Garden), model partners like Anthropic (Claude) and third‑party/open models (Qwen, Gemma, Llama variants), payments & commerce partners collaborating on AP2 (e.g., Mastercard, PayPal, American Express, Coinbase, Adyen, Etsy, Revolut and many others), developer community contributors (Gemini CLI creator Taylor Mullen and Agent Factory podcast participants; Google Cloud leaders such as Keith Ballinger appear in the developer narrative), and standards/implementation consumers like retailers, fintechs and platform vendors. (cloud.google.com)

Over the past weeks Google Cloud and partners have published a coordinated set of infrastructure, software and customer-case updates that together push high‑performance LLM inference into production: Google published an xPU / vLLM performance tuning guide (vLLM Performance Tuning, Aug 25, 2025) and a rightsizing/serving playbook for vLLM on GPUs and TPUs, released an upgraded XProf profiler and Cloud Diagnostics XProf library for deep xPU profiling (Sep 15, 2025), and continued rolling out AI Hypercomputer hardware/software stacks (Ironwood TPUs, A4/Blackwell GPU VMs and Dynamic Workload Scheduler) while customers and partners (Baseten, trading/finance benchmarks on C3 machines) published measurable cost/performance results (e.g., Baseten reports +225% cost‑performance for high‑throughput inference using A4/Blackwell + DWS). (cloud.google.com)

These developments matter because they address the three classic blockers to LLM production: (1) observability/profiling across heterogeneous accelerators (XProf gives per‑op, memory and CUDA‑graph visibility on GPUs/TPUs), (2) software/hyperparameter/operator tuning for inference engines (vLLM/xPU tuning guidance with concrete throughput/latency tradeoffs), and (3) economics and orchestration (AI Hypercomputer + Dynamic Workload Scheduler + partner optimizations that yield large improvements in tokens/sec and cost‑per‑request). Together they materially lower time‑to‑production, reduce TCO for reasoning/long‑context models, and shift the vendor competition toward tightly co‑designed hardware+software stacks. (cloud.google.com)

Key actors are Google Cloud (AI Hypercomputer, GKE Inference Gateway, TPU and A‑class VM rollouts, Cloud Diagnostics XProf), the vLLM open‑source serving community and its corporate integrators (vLLM authors / contributors), NVIDIA (Blackwell / HGX B200 on A4 VMs and associated SDKs), partners/customers demonstrating wins (Baseten’s case study), and the OpenXLA/XProf ecosystem that brings Google internal profiler advances to the community. Notable customers cited include Character.AI and Baseten in Google’s posts and case studies. (cloud.google.com)

Google Cloud is rapidly extending Dataproc and its Serverless for Apache Spark offering to tightly integrate AI-driven developer productivity and enterprise-scale analytics: in 2025 Google announced public-preview Gemini Cloud Assist Investigations that use Gemini to analyze failed and slow Spark/Dataproc workloads and recommend fixes, highlighted Dataproc’s Lightning Engine and other advanced Spark optimizations for analytics and AI, and launched Dataproc multi-tenant clusters to let notebook-driven data scientists share cluster compute with strong per-user isolation and service-account mappings. (cloud.google.com)

These moves matter because they address two major friction points for AI and analytics teams: (1) operational velocity — AI-assisted investigations reduce mean-time-to-resolution for Spark failures and performance problems, and (2) infrastructure efficiency and security — serverless Spark plus multi-tenant clusters aim to lower TCO, improve resource utilization for notebook-driven workflows, and provide stronger isolation and IAM-based access controls for enterprise data platforms. Collectively these features are positioned to accelerate model training/feature engineering and production analytics at lower operational cost. (cloud.google.com)

Primary players are Google Cloud (Dataproc engineering and product teams), the Gemini/Vertex AI family (providing the LLM and assistant capabilities), and the broader Apache Spark ecosystem (Apache Software Foundation and Spark users). Customers (enterprises, data engineering/science teams) and hardware/accelerator partners (GPU support is surfaced for Serverless Spark) are also central to adoption and integration decisions. (cloud.google.com)

Google Cloud is pushing an integrated ‘data cloud modernization’ stack for AI workloads that ties together BigLake (a storage engine that unifies data lakes and warehouses across object stores and clouds), a public commitment to Apache Iceberg as the open table format for lakehouses, new storage-first guidance for ML model artifacts (centralizing artifacts in Cloud Storage + high-performance access patterns), and a set of migration/transfer programs to reduce friction when moving on-prem or multicloud (Database Migration Program, managed Kafka patterns, and a no-cost Data Transfer Essentials offer for EU/UK customers). (cloud.google.com)

Together these moves lower technical and commercial barriers to modernizing data estates for AI: BigLake and Iceberg increase interoperability so teams can keep a single copy of data (avoid duplication and ETL), storage-focused MLOps patterns reduce model deployment latency and operational footprint, and migration/egress policy changes address vendor-lock-in and total-cost-of-ownership — all of which accelerate enterprises’ ability to train, serve, and govern large-scale AI on GCP and in multicloud settings. These are strategic because AI workloads amplify storage, governance, and ingress/egress costs and complexity. (cloud.google.com)

Primary players are Google Cloud (BigLake, BigQuery, Cloud Storage, Storage Intelligence, Data Transfer Essentials, Database Migration Program, Vertex AI), the Apache Iceberg community and ecosystem partners (Confluent, Databricks, dbt, Fivetran, Informatica, Snowflake cited as partners), plus ecosystem runtimes and integrators (Spark, Trino/Presto, TensorFlow, Kafka/Confluent and SI/partner consultancies). Competitive and regulatory actors — notably AWS and Microsoft Azure and EU/UK regulators implementing the EU Data Act — also shape the landscape. (cloud.google.com)

Google Cloud and its ecosystem are rolling out a rapid wave of AI security, governance, and compliance capabilities — combining platform controls (Security Command Center, Model Armor, Cloud Armor), new governance tooling (Compliance Manager, Data Security Posture Management, IAM role picker / Agentic IAM), partner integrations (security vendors embedding Gemini/Vertex AI) and operational patterns (centralized MCP proxy, agentic SOC) to both protect AI workloads and use AI to strengthen defenders; these announcements were highlighted across Google Cloud blog posts and the Security Summit (announced Aug 19, 2025). (cloud.google.com)

This matters because organizations face twin pressures: (1) an explosion of agentic AI use that expands attack surface and introduces novel risks (prompt injection, tool poisoning, session hijacking) and (2) stricter sovereignty/compliance requirements and operational complexity — prompting product-level AI protections, compliance automation, partner ecosystems (MSSPs, security ISVs), and prescriptive landing-zone patterns to close gaps and accelerate secure AI adoption. These CISO-level concerns and the expectation that AI will materially shift SOC automation were explicitly noted by APAC security leaders in Google Cloud’s CISO perspectives (Sep 15, 2025). (cloud.google.com)

Google Cloud is the primary platform and coordinator (Security Command Center, Model Armor, Cloud Armor, Agentspace/Agent Builder, Vertex AI/Gemini); strategic partners and vendors include Apiiro, Broadcom (Symantec), CrowdStrike, Trend Micro and many MSSPs; Mandiant provides frontline intelligence and incident response integration; industry analysts (e.g., BCG study cited by Google Cloud) and national/regulatory bodies (e.g., Canada’s CCCS / Protected B guidance) shape compliance patterns. Partner-built innovations and marketplace listings were called out in Google Cloud’s partner announcement (Sep 9, 2025). (cloud.google.com)

Google Cloud has rolled out a tightly coupled set of agent- and model-driven offerings through late-August/September 2025: the Conversational Analytics API (public preview announced August 25, 2025) that lets developers embed natural-language, chat-based queries over Looker and BigQuery anywhere; a Conversational Commerce agent on Vertex AI (GA announced September 10, 2025) designed to drive product discovery and purchases (with pilot/launch partners such as Albertsons Companies); and a new open Agent Payments Protocol (AP2) (announced September 16, 2025) — an industry-facing, cryptographically-backed specification (backed by 60+ merchants, issuers and payments players) to let AI agents negotiate, mandate, and execute payments with auditable intent/cart/payment mandates. These platform releases are already being trialed in domain deployments (for example an AI assistant at Seattle Children’s Hospital described in a Google Cloud case post on September 18, 2025), and Google has published docs, SDKs and sample code to accelerate developer adoption. (cloud.google.com)

Taken together the announcements signal Google Cloud’s strategy to operationalize agents across analytics, commerce, and domain workflows: conversational analytics brings trusted semantic models and BigQuery context into chat-driven decisioning; Conversational Commerce couples discovery-to-cart agent flows with production-ready shopping experiences (Google cites a macro cost of poor discovery of roughly $2 trillion annually); and AP2 attempts to close the trust/liability gap for agent-initiated payments by standardizing verifiable ‘mandates’ and an auditable trail for issuers, merchants and wallets — potentially enabling a new class of autonomous personal-shopping and enterprise procurement agents while forcing banks, card networks and retailers to adapt or integrate with the protocol. These shifts have implications for fraud/risk models, UX patterns (human-present vs human-not-present flows), regulatory oversight, and payments interoperability. (cloud.google.com)

The primary orchestration and specification work is led by Google / Google Cloud (Vertex AI, Gemini, Looker and BigQuery teams, agent product leads), with early commercial / launch partners such as Albertsons Companies for Conversational Commerce and Seattle Children’s Hospital for a healthcare assistant. AP2’s initial ecosystem partners include major payment and platform players (reported as 60+ participants) such as Mastercard, American Express, PayPal and Coinbase (and other merchants, issuers and wallets who have publicly signaled interest), plus community/publishing of the AP2 spec and reference implementations on GitHub and the AP2 project site. Coverage and analysis have been published by press outlets (TechCrunch, Axios, PRNews) and community writeups. (cloud.google.com)

Over summer–fall 2025 Google Cloud published a concentrated set of practical learning resources and programs to accelerate generative AI and data science adoption: a technical blueprint collection ('101+ gen AI use cases with technical blueprints' published Aug 21, 2025), a curated set of 25+ enterprise how‑to guides (Jul 22, 2025), a new eBook 'A Practical Guide to Data Science on Google Cloud' (Sep 16, 2025), a no‑cost generative AI training and certification program for veterans (registration opened Sep 8, 2025), and a new suite of targeted AI training courses for intermediate/advanced learners (announced Sep 19, 2025). (cloud.google.com)

These releases bundle prescriptive blueprints, how‑to engineering guides, hands‑on eBook content, and public programs — signalling Google Cloud’s push to (1) lower the barrier to production for GenAI via Vertex AI/BigQuery/AI Studio integrations, (2) scale workforce reskilling through free and paid certification pathways, and (3) embed security and governance best practices into training — outcomes that could materially speed enterprise GenAI deployment and influence hiring/certification benchmarks. (cloud.google.com)

Primary actors are Google Cloud (product/engineering and Cloud Learning teams), Google Public Sector (outsreach and public programs), the Vertex AI and Gemini product families (underpinning the technical guidance), and Google Cloud Skills Boost / Google Cloud Learning as the training/certification delivery channels; these posts and programs are authored and promoted by Google Cloud leaders and teams (e.g., Richard Seroter, Andrea Sanin, Jeff Nelson, Erin Rifkin, Karen Dahut). (cloud.google.com)

Google Cloud is rapidly publishing a wave of customer success stories and AI case studies — spanning startups, financial services, manufacturing, healthcare, and public sector — that document real-world deployments of its AI and data stack (Vertex AI, Gemini, BigQuery, Dataplex, Manufacturing Data Engine, Analytics Hub and related services) to solve domain problems (e.g., DB Lumina for Deutsche Bank research, Tata Steel’s Manufacturing Data Engine for OT/IoT monitoring, Seattle Children’s Pathways Assistant for clinician workflows, and state DOT RAG pilots). These posts (Sept 5–24, 2025 and related items) emphasize measurable operational outcomes (time saved, fidelity of AI-generated drafts, reduced downtime, large-scale model usage by startups) and show both enterprise customers and startups building production AI on GCP. (cloud.google.com)

This matters because the case studies collectively demonstrate Google Cloud’s strategy of offering an integrated, domain-anchored AI + data platform (compute, silicon choices, model access, data governance and RAG tooling) to accelerate time-to-value and reduce integration friction — enabling rapid pilots (week-long sprints), domain-grounded RAG systems, predictive maintenance at scale, and new product launches by startups. The stories also surface implications for procurement and partnerships (cloud credits/accelerators for AI startups), governance and responsible-use guardrails for public-sector deployments, and vendor competition around startup ecosystems and model/compute supply. (cloud.google.com)

Key players include Google Cloud (Vertex AI, Gemini, BigQuery, Dataplex, Manufacturing Data Engine and partner programs), customers highlighted in the case studies (Deutsche Bank, Tata Steel, Indiana DOT, Oklahoma DOT, Seattle Children’s Hospital), system integrators and partners (e.g., North Highland, Litmus, ClearBlade), and dozens of startups and ISVs using GCP (examples listed in the startups roundup such as Afooga, Anara, Inworld AI, Krea.ai and many others). These stakeholders (platform provider, enterprise/agency adopters, partners, and early-stage builders) are demonstrating both technical patterns and business outcomes for AI on GCP. (cloud.google.com)

Google Cloud is building out an end-to-end AI infrastructure stack — from subsea capacity and specialized cables to high‑performance instance types, TPU v5p-based Hypercomputer improvements, Kubernetes/GKE dataplane advances, and service‑mesh/networking features like Traffic Director’s proxyless gRPC — to create a tightly integrated "AI backbone" that reduces latency, increases throughput for distributed training/serving, and simplifies operational management for ML workloads. (cloud.google.com)

This matters because modern generative AI and large‑scale ML push networking, hardware, and orchestration to new limits — Google’s investments (subsea cables, multi‑core fiber, Jupiter/Andromeda fabric, TPU v5p pods, C3/Titanium instances, eBPF/Cilium dataplane and proxyless gRPC) are intended to deliver lower P50/P99 latencies, multi‑region throughput scale, predictable high IOPS and message rates, and simpler secure service meshes — which together change how organizations design, cost, and operate ML training and inference at cloud scale. (cloud.google.com)

Primary players are Google Cloud (AI Hypercomputer, TPU v5p, Jupiter/Andromeda fabric, GKE Dataplane V2, Traffic Director, subsea cable investments), hardware and cable partners such as NEC and SubCom (MCF and cable builds), ecosystem partners like Aeron/Adaptive for low‑latency messaging and benchmark work, and open‑source/network projects (Kubernetes, Cilium/eBPF, Envoy and gRPC/xDS) that enable the dataplane and proxyless service mesh capabilities. (cloud.google.com)

Over summer 2025 Google Cloud moved to make open models and "run-anywhere" model endpoints a practical enterprise reality: Vertex AI published a step-by-step guide (July 25, 2025) to take open models from discovery through fine-tuning to production endpoints using the Vertex AI Model Garden (200+ validated models) and managed inference features; Anthropic’s Claude models were given a new global endpoint on Vertex AI to improve availability (generally available July 28, 2025); Google announced general availability of Gemini on Google Distributed Cloud (on‑prem/air‑gapped and connected options) (August 28, 2025); and Google launched a no‑cost Data Transfer Essentials offering for EU & U.K. customers (September 10, 2025) to remove outbound egress friction for in‑parallel multicloud workloads. (cloud.google.com)

Taken together these moves lower the three major practical barriers to enterprise adoption of large models and multicloud architectures: (1) friction of discovery, fine‑tuning and production serving for open models (managed Model Garden, PEFT notebooks, optimized serving stacks); (2) availability and resilience across regions (global endpoints for third‑party models like Claude); and (3) data sovereignty and multicloud portability (on‑prem Gemini via Google Distributed Cloud and zero‑cost multicloud transfer in the EU/UK), with regulatory alignment to the EU Data Act and clear competitive implications for AWS and Azure. These changes accelerate hybrid/multicloud patterns, reduce vendor‑lock‑in friction, and shift cost/operational tradeoffs for enterprises building production AI. (cloud.google.com)

Primary actors are Google Cloud (Vertex AI, Google Distributed Cloud, Data Transfer Essentials), model providers and partners such as Anthropic (Claude models) and NVIDIA (hardware partnerships for Gemini on GDC), customers and system integrators (examples cited include Replicate and government customers such as GovTech Singapore/CSIT), and regulators in the EU/UK implementing the EU Data Act; competitors (AWS, Microsoft Azure) and the broader open‑model ecosystem (Qwen, Llama, Gemma, etc.) are also central to the market dynamics. (cloud.google.com)

In Aug–Sep 2025 Google Cloud published a coordinated set of technical posts and case studies that quantify per-query energy, emissions and water for Gemini-era inference and lay out practical engineering patterns to reduce cost and energy at scale — for example a Google Cloud measurement paper estimates a median Gemini Apps text prompt uses ~0.24 Wh, 0.03 gCO2e, and ~0.26 mL of water, while companion posts and product updates (GKE Inference Gateway / AI Hypercomputer, model-streaming, Cloud Storage artifact strategies) and partner case studies (Baseten) show software + hardware co-optimization (new A4 VMs / NVIDIA Blackwell, model compilation with TensorRT-LLM, prefix-aware routing, disaggregated serving, Dynamic Workload Scheduler) can deliver large improvements in cost-performance and startup time for inference. (cloud.google.com)

This matters because per-query efficiency gains (reported by Google as dramatic reductions year-over-year) both lower monetary cost for inference and reduce marginal energy/carbon per request, enabling wider deployment of generative AI — but the same efficiency gains can drive higher overall energy demand as usage scales (the Jevons/rebound concern), so operational patterns (right-sized accelerators, fast model loading, disaggregation, workload schedulers) and transparent measurement frameworks are now central to AI sustainability and TCO decisions for enterprises. (cloud.google.com)

Key players include Google Cloud (engineering and product teams publishing the measurement methodology, GKE Inference Gateway, AI Hypercomputer, storage and model-artifact guidance), infrastructure partners and ecosystem projects (NVIDIA hardware and software stacks such as TensorRT-LLM and Dynamo; open-source/projects like vLLM and llm-d), and customers/ISVs (Baseten’s case study) — along with independent reporters and researchers scrutinizing claims (coverage and critique from outlets like Ars Technica and The Verge). (cloud.google.com)

Google Cloud is accelerating developer productivity and AI‑ops by releasing and integrating tools across profiling, incident support, setup, and container workflows: XProf (moved under OpenXLA) and the Cloud Diagnostics XProf library (announced Sep 15, 2025) bring Google’s internal ML profiler to the community and a managed TensorBoard hosting path for fast large‑profile analysis; Google Cloud Setup (launched Aug 1, 2025) provides guided, exportable (Terraform) foundation flows for proof‑of‑concept, production and enhanced‑security environments; Skopeo guidance (Aug 27, 2025) adds daemonless, scriptable image operations with Artifact Registry to simplify CI/CD and image migration; and the Cloud Support API (documented/applied since Apr 4, 2022) enables automated “red‑button” / break‑glass case creation for critical incidents integrating with support workflows. (cloud.google.com)

Taken together these developments reduce friction across the ML development lifecycle—faster profiling and memory/roofline views (XProf) speed model optimization and hardware tuning, guided foundation setup shortens secure onboarding and enforces baseline controls (including CMEK and Security Command Center), registry/CI improvements (Skopeo + Artifact Registry) reduce image transfer and promotion cost/latency in pipelines, and programmable support APIs let organizations automate incident escalation—shifting the bottleneck from tool plumbing to model and product iteration, while raising operational, security and vendor‑control considerations. (cloud.google.com)

Primary players are Google Cloud product teams (authors and maintainers of Cloud Setup, Cloud Support API, Cloud Diagnostics XProf integration and the blog posts), the OpenXLA community (XProf’s external hosting and framework neutrality), TensorBoard/TensorFlow tooling teams (visualization frontend), the container/OSS ecosystem around Skopeo (open‑source project used with Artifact Registry), and enterprise customers/DevOps/SRE teams who must adopt these flows; secondary players include hardware/software vendors (NVIDIA, TPU/GPU/ xPU ecosystem authors referenced in XProf), and third‑party CI/CD and security tools that integrate with Artifact Registry and Binary Authorization. (cloud.google.com)