Everything Announced at Google Cloud Next in 10 Minutes
CHAPTERS: 0:00 Gemini 2.5 Pro 0:05 Red de Área Amplia en la Nube 0:50 Google TPU: Ironwood 2:22 Gemini 2.0: Mejoras en la hipercomputadora de IA 3:44 Gemini en Google Distributed Cloud 4:25 Funciones de análisis, audio y flujo de trabajo 4:59 Lyria, Llama 4 y Vertex AI 6:29 Kit de Desarrollo de Agentes de Google 9:22 Suite de Interacción con el Cliente 10:00 Agentes de Datos: Ingeniería, Ciencia y Análisis 11:00 Agentes de Asistencia de Código de Gemini
Produced by: Samantha Harvey Edited by: Jacob Smith Animation: Charlotte Brown Additional Footage: Getty Images
Google Cloud Next introduced significant advancements in infrastructure and AI, featuring the Cloud WAN for global networking and the Ironwood TPU (7th gen) for massive compute power. The event highlighted new AI inference tools, Gemini integrations in Workspace and air-gapped environments, and a comprehensive ecosystem for building and deploying multi-agent AI systems.
Hello, everyone.
Welcome to Google Cloud Next.
Today, I'm pleased to announce that we
are making Google's global private network available
to enterprises around the world.
We call it Cloud Wide Area Network, or WAN.
Cloud WAN leverages Google's planet-scale network.
It's optimized for application performance
and delivers over 40% faster performance
while reducing total cost of ownership by up to 40%.
Companies like Citadel Securities and Nestle
are already using this network for faster, more reliable
solutions.
And it will be available to all Google Cloud customers
later this month.
Today, I'm proud to announce our seventh generation TPU,
Ironwood, is coming later this year.
Compared to our first publicly available TPU,
Ironwood achieves 3,600 times better performance,
an incredible increase.
It's the most powerful chip we have ever built
and will enable the next frontier of AI models.
In the same period, we've also become 29x more energy
efficient, and Amin will share more later today.
Today, we introduced our seventh generation of TPUs, Ironwood.
Ironwood TPUs are our largest and most powerful TPU pods
to date, more than a 10x improvement
from our most recent high-performance TPU
with over 9,000 chips per pod.
To meet the exponentially growing demands
of the most demanding thinking models, like Gemini 2.5,
this delivers a staggering 42.5 exaflops of compute per pod.
To give you a sense of the scale,
the world's number one supercomputer
supports 1.7 exaflops.
Ironwood pods offer more than 24 times that compute power.
Software is how we orchestrate and simplify access
to this powerful hardware.
And today, we're introducing three enhancements
for AI inference.
First, we're introducing new inference capabilities
in Google Kubernetes Engine, including
Gen AI-aware scaling and load balancing features,
which help reduce serving costs by up to 30%.
Tail latency by up to 60%, and increased throughput
by up to 40%.
Second, we're announcing that Pathways, Google's own
distributed ML runtime powering Gemini,
is now available for the first time for cloud customers.
Developed by Google DeepMind, Pathways
enables state-of-the-art multi-host inferencing
for dynamic scaling with high performance at optimal costs.
Now, you can scale out model serving
to hundreds of accelerators for the best
combination of batch efficiency and low latency.
Third, we're bringing VLLM to TPUs.
This allows customers who optimize PyTorch with VLLM
for GPUs to easily and cost-efficiently run
their workloads on TPUs.
All of these AI hypercomputer hardware and software
enhancements together enable us to deliver
more intelligence or useful AI output
at a consistently low price.
This is one reason why Gemini 2.0 Flash,
powered by AI hypercomputer,
achieves 24 times higher intelligence per dollar
compared to GPT-4.0 and five times higher
than DeepSeek R1.
Today, we are announcing that Gemini can run
on Google Distributed Cloud locally
in air-gapped environments,
as well as connected environments.
This all comes with the support
for NVIDIA's confidential computing
and Blackwell systems, DGX-B200 and HGX-B200 platforms
with Dell as a key partner.
This complements our Google Distributed Cloud
air-gap product, which is now authorized
for U.S. government secret and top-secret missions,
and on which Gemini is now available,
providing the highest levels of security and compliance.
Today, I'm pleased to announce three new innovations
with Gemini in Workspace.
Help me analyze in Google Sheets,
which guides you through your data
to complete expert-level analysis.
Audio overviews in Google Docs,
where you can interact with Docs in an entirely new way
by creating high-quality audio versions of your content.
And Google Workspace Flows to help you automate
time-consuming, repetitive tasks
and to make decisions with more context.
Today, we're making Lyria available on Google Cloud
to transform text prompts into 30-second music clips
and we're the first hyperscaler to offer this capability.
Let's hear a clip from Lyria.
We are excited to announce the general availability
of Metasllama 4 on Vertex AI.
And last week, we announced that AI2's full portfolio
of open models are also accessible
on the Vertex AI Model Garden.
With Vertex AI, you can be sure your model has access
to the right information at the right time.
You can connect to any data source
or any vector database on any cloud.
And announcing today, you can build agents
directly on our existing NetApp storage
without requiring any data duplication.
You can connect to a broad range of applications,
including Oracle, SAP, ServiceNow, and Workday.
And for model factuality,
we offer the most comprehensive approach
to grounding on the market today,
combining grounding with Google Search,
grounding with your own enterprise data,
Google Maps, and third-party sources.
Now, with Google Cloud, starting today,
you can build and manage multi-agent systems
with Vertex AI and our new agent development kit.
You can scale the adoption of agents
across your enterprise
with our newly released Google Agent Space.
And you can accelerate deployment with packaged AI agents
that are ready for use today.
Now, following the introduction
of Vertex AI Agent Builder last year,
we're announcing today a new agent development kit.
It is a new open-source framework
that simplifies the process
of building sophisticated multi-agent systems.
Now you can build sophisticated Gemini-powered agents,
help them use tools, do complex multi-step tasks,
including reasoning or thinking.
You can also discover other agents,
learn their skills,
and enable agents to work together
while maintaining precise control.
Agent development kit supports the model context protocol,
which provides a unified way for AI models
to access and interact with various data sources and tools
rather than requiring custom integrations
for each and every one.
We're also introducing a new agent-to-agent protocol
that allows agents to communicate with each other
regardless of the underlying model and framework
they were developed with.
This protocol is supported by many leading partners
who share a vision to allow agents
to work across the multi-agent ecosystem
and with agents built on other agent frameworks,
including Landgraf and Crew AI.
Today, we're putting AI agents
in the hands of every employee with Google Agent Space.
Employees using Google Agent Space
can now find and synthesize information
from within their organization,
converse with AI agents,
and have these agents take action
on their behalf for their enterprise applications.
Google Agent Space combines
Google quality enterprise search,
conversational AI or chat,
and Gemini and third-party agents.
It also includes a broad set of tools,
including purpose-built connectors to search and transact
with documents and databases,
as well as SaaS applications
with advanced security and compliance
to protect your data and your intellectual property.
Today, we're announcing our next generation
of customer engagement suite,
which will include human-like voices, comprehension,
and the ability to understand emotions
so agents can adapt better during the conversation.
Streaming video support so virtual agents
can interpret and respond to what they see
in real time through customer devices.
AI assistance to build custom agents in a no-code interface.
And the ability to use a variety of tools
through API calls to interact
and perform specific tasks for your application,
like look up products, add to cart, or check out.
And today, we're very excited to announce
specialized agents for every member of your data team.
Now, for data engineering teams,
we deliver agents for all aspects
of the data engineering lifecycle,
from catalog automation to metadata generation
to maintaining data quality to data pipeline generation.
And for data science teams,
our AI agent acts as a comprehensive coding partner
in your data science notebook,
accelerating every step of your workflow
from data loading and feature engineering
to predictive modeling.
And for data analysts and business users,
our conversational analytics agent
performs powerful, trustworthy analysis
entirely in natural language.
And you can also embed this agent in line
in your own web or mobile application.
And today, we're announcing new code assist agents
to help with everything from modernizing code
to helping with the full software development lifecycle.
Developers can interact with our agents
on a Kanban board, which provides a real-time display
of the tasks Code Assist is working on,
as well as the ability for developers
to interact with our agents.
Code Assist also has integrations
with dozens of partners, such as Atlantean,
Sentry, Snyk, and many more coming soon.
At Economy Media, your opinion matters to us.
Subscribe and let us know what you think
in the comments below.