Google Cloud Next 2026: Every Major Announcement — Agents, TPU 8, Virgo Network & More
- The One-Line Thesis: “The Agentic Enterprise Is Real”
- 1. Gemini Enterprise Agent Platform (Vertex AI Reborn)
- 2. TPU 8t & TPU 8i — Two Chips for the Agentic Era
- 3. Virgo Network — 1 Million TPUs in One Cluster
- 4. Agentic Data Cloud — Knowledge Catalog & Cross-Cloud Lakehouse
- 5. A2A Protocol v1.0 in Production at 150 Organisations
- 6. Workspace Studio — No-Code Agent Builder for Everyone
- 7. Agentic Defense — Security at Machine Speed
- 8. Gemini 3.1 Pro, 3.1 Flash & GLM 5
- 9. $185 Billion in CapEx — Google’s Infrastructure Bet
- 10. Customer Deployments: Walmart, NASA, and More
- What It All Means
- FAQ
⚡ Key Numbers at a Glance
The One-Line Thesis: “The Agentic Enterprise Is Real”
Thomas Kurian took the stage in Las Vegas on April 22 with a line that set the tone for everything that followed: “The experimental phase is behind us. The era of the pilot is over. The era of the agent is here.” Google Cloud Next 2026 was not a product showcase dressed up as a strategy event. It was a full rebrand of Google’s enterprise AI stack — consolidated, renamed, and repositioned around a single idea: the Agentic Enterprise.
In a pre-recorded video, CEO Sundar Pichai disclosed that 75% of all new Google code is now AI-generated — up from approximately 25% a year ago. Google’s total capital expenditure is expected to reach $185 billion in 2026, up from $31 billion just four years earlier. Over half of that compute investment is directed at the Cloud business. Gemini’s first-party API usage hit 16 billion tokens per minute, up from 10 billion last quarter — a 60% quarter-over-quarter jump that gives the infrastructure announcements in this keynote a concrete scale context.
Google’s competitive framing was deliberate: “Competitors hand you the pieces, not the platform.” Every announcement at Cloud Next 2026 was architected to reinforce a single pitch — that Google is the only cloud provider that owns the full vertical from custom silicon through frontier models through enterprise productivity software. Here is everything that was announced.
1. Gemini Enterprise Agent Platform — Vertex AI Reborn
The single most structurally significant announcement is the rebrand and consolidation of Vertex AI into the Gemini Enterprise Agent Platform — absorbing Agentspace into a unified product. This is not cosmetic renaming. It is a fundamental reorganisation of Google’s developer and enterprise AI surface area around agents as the primary unit of work.
The new platform includes several components, all newly announced or promoted to general availability:
Agent Studio is a visual flow canvas for building agent workflows, now in preview. Agent-to-Agent Orchestration lets agents hand off tasks to other agents, including third-party agents from Box, Workday, Salesforce, and ServiceNow. Agent Registry provides a discovery layer for enterprise agent inventories. Agent Identity gives each agent a verifiable cryptographic identity. Agent Gateway is the “air traffic control” for the agent ecosystem — enforcing MCP and A2A protocol policies in real time, with integrated Model Armor protection against prompt injection and tool poisoning. Agent Observability provides monitoring, anomaly detection, and a security dashboard across the full agent fleet.
The Model Garden now hosts over 200 models — including Anthropic’s Claude — giving enterprises a model-agnostic deployment surface within the Google Cloud platform. ADK (Agent Development Kit) hit v1.0 stable release in Python, Java, JavaScript, and Go.
Gemini Enterprise app also received significant upgrades: a new Agent Designer for building agents in natural language, an Inbox for managing agent activity across the enterprise, long-running agents for background tasks, Skills for modular agent capabilities, and Projects (coming soon) as a shared workspace for teams and agents.
Gemini Enterprise Agent Platform: What’s New vs. What Was There Before
| Component | Previous State | Cloud Next 2026 Status |
|---|---|---|
| Vertex AI | Developer ML platform | Renamed → Gemini Enterprise Agent Platform |
| Agentspace | Separate enterprise search/agent tool | Absorbed into Gemini Enterprise |
| Agent Studio | Did not exist | New — visual workflow builder, in preview |
| Agent Gateway | Did not exist | New — real-time MCP/A2A policy enforcement |
| Agent Identity | Did not exist | New — cryptographic agent verification |
| Model Garden | ~50 models | 200+ models incl. Anthropic Claude |
| ADK | Beta (Python only) | v1.0 stable — Python, Java, JS, Go |
2. TPU 8t & TPU 8i — Two Chips for the Agentic Era
The chip announcement is the most technically consequential thing Google did at Cloud Next 2026. For the first time, Google’s eighth-generation TPUs ship as two distinct chips with specialised roles — a major architectural departure from previous generations which were general-purpose.
TPU 8t is the training powerhouse. It scales to 9,600 chips in a single superpod, delivers 121 exaflops of compute with two petabytes of shared, high-bandwidth memory, and achieves nearly 3x the compute performance of the previous generation Ironwood. It targets 97% “goodput” — meaning 97% of compute time is actually spent training, not idle — through real-time telemetry, automatic fault rerouting, and optical circuit switching for hardware-level failure recovery with no human intervention. TPU 8t is designed to turn months of model training into weeks.
TPU 8i is the inference and reinforcement learning chip, purpose-built for the low-latency demands of agentic workflows. It triples on-chip SRAM to 384 MB and increases high-bandwidth memory to 288 GB, keeping massive KV caches entirely on-chip — eliminating the memory-to-compute bottleneck that slows token generation under high concurrency. It delivers 80% better performance-per-dollar for inference than the prior generation, which Google says enables serving nearly twice the customer volume at the same cost. TPU 8i connects 1,152 chips in a single pod and introduces a dedicated Collectives Acceleration Engine that reduces on-chip latency by up to 5x.
Both chips run on Google’s own Axion ARM-based CPU hosts — meaning the full system (not just the chip) is optimised. Both are compatible with PyTorch for TPUs, native JAX, SGLang, and vLLM. General availability is expected later in 2026; interested customers can register now.
3. Virgo Network — 1 Million TPUs in One Cluster
The Virgo Network is a new megascale, purpose-built AI data center fabric announced alongside the TPU 8 chips. It is the networking layer that turns individual TPU pods into planet-scale training clusters.
Using a two-tier, all-to-all packet switch topology with extremely high port densities, Virgo can connect 134,000 TPUs into a single fabric within a single data center — and stitch multiple sites together into clusters of over one million TPUs. Google describes this as “transforming globally distributed infrastructure into one seamless supercomputer.”
Virgo is also available for NVIDIA Vera Rubin NVL72 systems — supporting up to 80,000 GPUs in a single data center and up to 960,000 GPUs across multiple sites. Storage was also upgraded: Google Cloud Managed Lustre now delivers 10 TB/s of throughput (a 10x improvement, up to 20x faster than other hyperscalers), and Rapid Storage improved from 6 TB/s to 15 TB/s. A new Smart Storage tier applies semantic meaning to unstructured data, forming the foundation for the Enterprise Knowledge Graph.
4. Agentic Data Cloud — Knowledge Catalog & Cross-Cloud Lakehouse
The Agentic Data Cloud is Google’s answer to the question that determines whether agentic AI can actually work in enterprise environments: where does the data live, and can agents reason over it in real time?
Three components were announced. The Knowledge Catalog grounds agents in trusted business context across an organisation’s entire data estate — providing a semantic layer that lets agents query and cite internal data sources accurately. The Data Agent Kit gives data scientists a Gemini-powered authoring environment for building and deploying data agents. The Cross-Cloud AI-Native Lakehouse provides seamless access to data regardless of which cloud platform it lives on — removing the siloed-data bottleneck that breaks most agentic workflows.
The strategic implication is pointed directly at Snowflake and Databricks. Google is explicitly positioning BigQuery and its data fabric as context engines for agents, not just storage and compute surfaces. The value proposition is shifting from “where the data lives” to “how AI reasons over it” — a reframe that challenges the business model of the entire modern data stack.
5. A2A Protocol v1.0 in Production at 150 Organisations
The Agent-to-Agent (A2A) protocol — Google’s standard for inter-agent communication across platforms — reached a significant milestone at Cloud Next 2026: v1.0 is in production at 150 organisations, not in pilot, routing real tasks between agents built on different platforms.
A2A is now governed by the Linux Foundation’s Agentic AI Foundation and has reached version 1.2, with signed agent cards using cryptographic signatures for domain verification. Microsoft, AWS, Salesforce, SAP, and ServiceNow are running A2A in production environments.
The relationship between A2A and Anthropic’s MCP is complementary, not competitive. MCP handles how an agent connects to tools and data sources. A2A handles how agents communicate with each other across organisational and platform boundaries. Google adopted MCP across its own services in December 2025, launching fully managed remote MCP servers for Google Maps, BigQuery, Compute Engine, and Kubernetes Engine. Apigee is being positioned as an API-to-agent bridge — enabling existing REST APIs to become callable tools for agents without rewriting integration code.
Also announced: Project Mariner, a web-browsing agent that lets Gemini-powered workflows navigate and interact with live websites as part of multi-step agent tasks.
6. Workspace Studio — No-Code Agent Builder for Everyone
Workspace Studio is the most consumer-facing announcement from Cloud Next 2026. It is a no-code platform that lets any business user — not just developers — build and deploy AI agents across Gmail, Docs, Sheets, Drive, Meet, and Chat by describing automations in plain language.
A user can type “every Friday, remind me to update my tracker” and Gemini builds the automation. Workspace Studio connects to third-party applications including Asana, Jira, Mailchimp, and Salesforce, and can call external APIs via webhooks or run custom logic through Apps Script.
Alongside this, Workspace Intelligence hit general availability — providing real-time, personalised contextual awareness across all Google Workspace apps. It understands complex semantic relationships between data in Gmail, Docs, and other apps, your active projects, your collaborators, and company-specific information. The system does the information-gathering so users don’t have to, surfacing what matters at the moment it’s needed.
A separate migration announcement that will catch Microsoft’s attention: Rapid Enterprise Migration is now in preview and claims to be up to 5x faster for moving an entire organisation from M365 to Google Workspace — built on significant interoperability improvements between the two ecosystems.
7. Agentic Defense — Security at Machine Speed
The cybersecurity announcement at Cloud Next 2026 carried an arresting statistic: the time between initial attacker access and handoff to secondary threat groups has collapsed from eight hours to 22 seconds. Traditional security operations that rely on human-speed investigation cannot respond at that pace. Agentic Defense is Google’s answer.
The platform combines Google’s Threat Intelligence and Security Operations with Wiz’s Cloud and AI Security Platform — following Google’s $32 billion acquisition of Wiz — into an integrated autonomous protection layer. New agents for threat detection, detection engineering, and remediation operate continuously. Wiz’s AI Application Protection Platform (AI-APP) adds protection specifically for AI workloads: guarding against prompt injection, tool poisoning, sensitive data leakage, and reasoning drift across multicloud, hybrid, and AI environments.
The Agent Gateway (part of the Gemini Enterprise Agent Platform) doubles as the security control plane — providing centralised, real-time policy enforcement for every agent interaction, with anomaly detection that flags suspicious tool use or unauthorised data access before it impacts operations.
8. Gemini 3.1 Pro, 3.1 Flash, GLM 5 & What’s Next
Cloud Next 2026 is not a Gemini 4 announcement event — DeepMind CEO Demis Hassabis confirmed that team is “focusing on Gemini 4 this year,” with a likely reveal at Google I/O in May. But the existing Gemini 3.x line received significant updates.
Gemini 3.1 Pro, the most advanced reasoning variant, is available in preview. Gemini 3.1 Flash delivers a 15% improvement in overall accuracy over Gemini 2.5 Flash and is optimised for high-frequency agentic workflows and real-time processing. Gemini 3.1 Flash Image (codenamed Nano Banana 2) and Veo 3.1 Lite (for high-volume video generation) and Lyria 3 Pro (audio/music generation) round out the media model updates. A new experimental model, GLM 5, targets complex systems engineering and long-horizon agentic tasks through the Model Garden.
Gemini 3.2 is expected to be formally announced during the conference, with an expanded context window beyond one million tokens and optimised parameter counts for reduced inference latency. The Gemini 2.5 generation is scheduled for retirement in October 2026 in favour of the 3.x line.
Also announced: Gemma 4 open models released under Apache 2.0 licensing — continuing Google’s open-weight strategy for developers who want to fine-tune and deploy locally.
9. $185 Billion in CapEx — Google’s Infrastructure Bet
Sundar Pichai’s video message to the Cloud Next audience included a figure that puts the scale of Google’s AI investment in context: total capital expenditure for 2026 is expected to reach approximately $185 billion — up from $31 billion just four years ago. More than half of that compute investment is directed at the Cloud business.
For context on what that means in practice: Anthropic — Google Cloud’s largest TPU customer — generates 80% of its revenue from enterprise API usage running on Google Cloud TPUs. Anthropic secured 3.5 gigawatts of next-generation TPU capacity through Broadcom and Google, starting 2027. Every performance improvement in TPU 8t and 8i announced at Cloud Next directly determines Anthropic’s inference cost structure — which flows through to Claude’s per-token pricing, which feeds into Anthropic’s competitive position ahead of its anticipated October 2026 IPO. The infrastructure and the competitive AI landscape are more directly linked than they appear on the surface.
Google also confirmed a chip co-development deal with Marvell Technology, building on Broadcom’s confirmed TPU 8 design codenames (Sunfish for training, Zebrafish for inference) as part of an extended 2031 supply chain roadmap.
10. Customer Deployments: Walmart, NASA, Team USA & More
Cloud Next 2026 featured a heavier-than-usual emphasis on production deployments rather than concept demos. The customer stories were a deliberate counterpoint to “pilot wearing logo costumes” criticism that has dogged enterprise AI announcements.
Walmart has deployed Gemini Enterprise internally to equip store and supply chain team leaders — using Google Pixel Pro Fold devices — to connect to live business data faster. A customer-facing Gemini-integrated shopping tool is also in rollout. NASA / Artemis II used Google Cloud agentic AI for flight readiness review. Team USA at the 2026 Winter Olympics used Google Cloud and Gemini with a DeepMind tool to analyse snowboarding footage — mapping athlete wireframes and tracking rotational velocity, flight dynamics, and tuck compression in real time (demonstrated on stage by three-time Olympic gold medallist Shaun White). Deutsche Telekom‘s Industrial AI Cloud in Germany — built on Google’s AI infrastructure — was highlighted as a blueprint for sovereign AI deployment in Europe.
🎯 Watch the Full Keynote
The Google Cloud Next 2026 opening keynote is available to stream for free. A free digital pass provides access to keynotes and select sessions on demand after the conference.
Read the Official Recap → Watch Sessions On Demand →What It All Means
Google Cloud Next 2026 is the clearest signal yet that the enterprise AI market is no longer competing on model quality alone. The battleground has shifted to the execution layer — the infrastructure, orchestration, governance, and application integration stack where agents actually run. Google’s thesis is that no competitor can match a vertically integrated stack that spans from custom silicon (TPU 8) through frontier models (Gemini) through developer tools (Gemini Enterprise Agent Platform) through enterprise apps (Workspace) and security (Agentic Defense).
The A2A protocol’s adoption — now at 150 organisations in production with Microsoft, AWS, Salesforce, SAP, and ServiceNow all participating — is the most strategically significant under-the-radar development. A2A is becoming the lingua franca for inter-agent communication across platform boundaries. Whoever defines and governs that protocol has structural influence over the entire multi-agent ecosystem, regardless of whose model is doing the reasoning.
For enterprises, the clearest takeaway is the one Kurian opened with: the experimental phase is over. The companies building toward the agentic execution layer now — with proper governance, data readiness, and agent security — are positioned for what Google is calling the defining opportunity of 2026. Those still running pilots are running out of runway.
❓ Frequently Asked Questions
When and where is Google Cloud Next 2026?
Google Cloud Next 2026 runs April 22–24, 2026 at the Mandalay Bay Convention Center in Las Vegas, Nevada. Approximately 32,000 people attended in person. The opening keynote was livestreamed globally and is available on demand via Google Cloud’s YouTube channel.
What is the Gemini Enterprise Agent Platform?
It is a comprehensive rebrand and consolidation of Vertex AI, absorbing Agentspace into a unified end-to-end platform for building, scaling, governing, and optimising AI agents. It includes Agent Studio, Agent Gateway, Agent Registry, Agent Identity, Agent Observability, over 200 models in the Model Garden, and ADK v1.0 stable in four languages.
What are TPU 8t and TPU 8i?
Google’s eighth-generation TPUs ship for the first time as two specialised chips. TPU 8t is the training chip — 9,600 chips per superpod, 121 exaflops, 3x the compute of the prior generation. TPU 8i is the inference and RL chip — 80% better performance-per-dollar than previous gen, 3x more on-chip SRAM, designed for low-latency agentic workflows. Both run on Google’s Axion ARM-based CPU hosts.
What is the Virgo Network?
Virgo is Google’s new purpose-built AI data center networking fabric. It can connect 134,000 TPUs within a single data center and over 1 million TPUs across multiple sites into a single training cluster. It also supports NVIDIA Vera Rubin NVL72 systems (up to 960,000 GPUs across sites).
What is the A2A protocol and why does it matter?
Agent-to-Agent (A2A) is Google’s open protocol for communication between AI agents across different platforms and organisations. It reached v1.0 in production at 150 organisations including Microsoft, AWS, Salesforce, SAP, and ServiceNow. It complements Anthropic’s MCP (which handles agent-to-tool connections) — together they form the emerging standard infrastructure for multi-agent enterprise systems.
What is Workspace Studio?
Workspace Studio is a no-code platform that lets any business user build and deploy AI agents across Google Workspace apps (Gmail, Docs, Sheets, Drive, Meet, Chat) using plain-language descriptions. It connects to Asana, Jira, Mailchimp, Salesforce, and external APIs via webhooks or Apps Script.
Is Google announcing Gemini 4 at Cloud Next 2026?
No. Demis Hassabis confirmed the DeepMind team is focused on Gemini 4 for 2026, but it is not expected to be announced at Cloud Next. The likely reveal is at Google I/O in May 2026. Gemini 3.2 may receive a formal announcement during the conference.
How much is Google spending on AI infrastructure in 2026?
Alphabet’s total capital expenditure for 2026 is expected to reach approximately $185 billion, up from $31 billion four years ago. More than half of the compute investment is directed at the Cloud business. This is the largest infrastructure bet in Google’s history.
Latest Articles
Browse our comprehensive AI tool reviews and productivity guides
Musk v. OpenAI Trial: The Case That Could Reshape the Entire AI...
Musk v. OpenAI Trial: The Case That Could Reshape the Entire AI Industry
Musk called himself "a fool" on the stand. Altman appeared by prerecorded video from AWS while being sued. The judge reprimanded both sides. And the AI industry's most consequential legal battle is just getting started.
Big Tech Q1 2026 Earnings: The $665 Billion AI Bet — Winners,...
Big Tech Q1 2026 Earnings: The $665 Billion AI Bet — Winners, Losers, and What It Means
Five tech giants reported Q1 2026 earnings in 48 hours. Combined AI capex: $665 billion — 75% more than 2025. Alphabet and Amazon won. Meta spooked investors. Here's every number that matters.
AI Is Replacing Developers — The Real Numbers (2026)
AI Is Replacing Developers — The Real Numbers (2026)
Snap fired 1,000. Google generates 75% of new code with AI. Entry-level developer jobs fell 20%. But 1.3M new AI roles were created and India's AI hiring surged 59.5%. Here's what's actually happening.
I Used Claude Free for 3 Months Instead of ChatGPT and Gemini...
I Used Claude Free for 3 Months Instead of ChatGPT and Gemini — Here’s What Happened
I launched and grew NivaaLabs on Claude's free tier for 3 months. I also used ChatGPT and Gemini. Here's the honest, task-by-task breakdown of what each AI actually does well — and which one I'd recommend for someone building something real on a $0 budget.
Runway Gen-3 Turbo: Real-Time Video Tested (2026)
Runway Gen-3 Turbo: Real-Time Video Tested (2026)
Runway Gen-3 Turbo's real-time video generation capabilities are put to the test, examining quality, speed, and value.
Best AI Coding Tools 2026: Every Major Tool Ranked — Cursor, Claude...
Best AI Coding Tools 2026: Every Major Tool Ranked — Cursor, Claude Code, Copilot, Windsurf & More
85% of developers now use AI coding tools daily. AI writes 46% of all new code. The market has 10+ serious tools and most developers end up using two or three. Here's how every major AI coding tool in 2026 ranks — with real benchmark data, honest pricing, and a verdict for every workflow type.
DeepSeek V4 Review: V4 Flash & V4 Pro — Almost Frontier, a...
DeepSeek V4 Review: V4 Flash & V4 Pro — Almost Frontier, a Fraction of the Price (April 2026)
DeepSeek V4 arrived April 24, 2026 — one year after R1 shook Silicon Valley. V4 Pro is the world's largest open-weight model at 1.6 trillion parameters. V4 Flash is cheaper than GPT-5.4 Nano. And both run on Chinese chips. Here's everything you need to know.
GPT-5.5 vs Claude Opus 4.6 (2026): Which AI Model Wins for Your...
GPT-5.5 vs Claude Opus 4.6 (2026): Which AI Model Wins for Your Work?
OpenAI's GPT-5.5 arrived April 23 claiming to be the smartest model yet. Anthropic's Claude Opus 4.6 still holds the top Chatbot Arena ELO. Both cost real money. Which one actually wins for your workflow? Here's the full data-driven comparison.
GPT-5.5 Review: OpenAI’s Smartest Model Yet — Agentic Coding, Computer Use &...
GPT-5.5 Review: OpenAI’s Smartest Model Yet — Agentic Coding, Computer Use & More (April 2026)
GPT-5.5 landed April 23 — seven weeks after 5.4. OpenAI calls it a "new class of intelligence for real work." It's faster per token, stronger at agentic coding, computer use, and scientific research, and comes with the strongest safety guardrails yet. Here's everything you need to know.
Project Glasswing: Anthropic’s “Too Dangerous to Release” AI and the Cybersecurity Reckoning
Project Glasswing: Anthropic’s “Too Dangerous to Release” AI and the Cybersecurity Reckoning
Anthropic built an AI so capable at hacking that they won't release it publicly. Claude Mythos Preview found a 27-year-old OpenBSD zero-day for under $50. Project Glasswing is what happens next.
Google Cloud Next 2026: Every Major Announcement — Agents, TPU 8, Virgo...
Google Cloud Next 2026: Every Major Announcement — Agents, TPU 8, Virgo Network & More
Google Cloud Next 2026 just happened. Here's everything: new 8th-gen TPUs, the Gemini Enterprise Agent Platform, A2A protocol in production at 150 orgs, Workspace Studio for no-code agents, and a $185B infrastructure bet. One article, all the details.
OpenAI ChatGPT Ads Review: The $100 Billion Bet That’s Already Getting Messy
OpenAI ChatGPT Ads Review: The $100 Billion Bet That’s Already Getting Messy
OpenAI launched ads in ChatGPT, faced user backlash, fired back at Anthropic's Super Bowl jabs, and just flipped to cost-per-click pricing. Here's what's actually happening.