AWS re:Invent 2025 – all the day two news and updates live from Las Vegas

Refresh

Get notified of updates

2025-12-03T18:12:42.738Z

And that’s a wrap – Sivasubramanian thanks us for coming, and we’re off to see more of the show floor, so come back later for more updates at re:Invent 2025.

2025-12-03T18:08:24.792Z

Sivasubramanian is back, and it’s the home straight for us.

“Everything we’re showing is just the beginning of what’s possible with AI,” he says, “with every step we are making it easier for anyone to build.”

2025-12-03T17:57:47.333Z

Time for another guest – Sivasubramanian wlecomes Colleen Aubrey, Senior VP, AWS Applied AI Solutions, to the stage.

She notes that transformation and agility aren’t opposed – they’re partners, and can be critical for business success.

“The real prize of AI is new products, services and business models,” she notes, but the journey is accelerating – with the ability to reimagine how we work the only limitation.

2025-12-03T17:54:57.321Z

Now, Sivasubramanian moves to look at the Frontier Agents announced yesterday – covering Kiro, DevOps and security with new, smarter services.

These agents are smart and more autonomous, he notes, allowing them to work alongside human workers much more effectively.

2025-12-03T17:53:32.547Z

2025-12-03T17:50:05.063Z

Nova Act provides a whole new way of agent training, Sivasubramanian notes, offering end-to-end training for everything an agent will ever interact with.

2025-12-03T17:48:17.577Z

How do we make this easier, Sivasubramanian notes – and a better model is the answer.

The new Amazon Nova Act is the solution – a new service to build and manage fleets of agents for automating production UI workflows at scale.

2025-12-03T17:45:20.317Z

Sivasubramanian is back, and wants to talk about relaibility – a crucial consideration for businesses.

Agents often struggle with being asked the same question repeatedly, he notes – so how can this be improved?

2025-12-03T17:37:31.716Z

Internally, Cook notes how AWS is using automated reasoning for tasks such as policy interpretation, storage, virtualization and more – “basically any place where failure is unacceptable”.

This approach, he says, is now being brought to agentic AI.

2025-12-03T17:36:27.257Z

Automated reasoning is the key to making AI agents most trustworthy, Cook notes, using algorithms to logically prove or disprove the behavior of a system.

2025-12-03T17:34:18.947Z

To explain more about how AWS is making its agents more trustworthy, Sivasubramanian introduces Byron Cook, VP, Distinguished Scientist, Automated Reasoning Group, AWS, to the stage.

Like giving a teenager access to your credit card, giving an agent access to your more precious information can be dangerous, Cook says – you might not get back exactly what you want.

2025-12-03T17:31:20.249Z

Sivasubramanian is back, and it’s time to talk about trust.

“As we hand over more tasks to AI agents…we need to be able to trust they will be able to perfrom as expected,” he notes.

“Agents introduce a whole new world of challenges which require radical new approaches.”

2025-12-03T17:29:42.352Z

Next, we have another customer story – Guillermo Rauch, CEO of Vercel is here.

He explains how Vercel Fluid Compute is powered by Amazon EC2 and AWS Lambda, helping the company create “fully self-driving compute infrastructure”.

2025-12-03T17:22:58.056Z

2025-12-03T17:21:26.797Z

Next, Sivasubramanian announces checkpointless training on Amazon SageMaker HyperPod, which allows users to recover training from faults in minutes acorss thousands of AI accelerators – what he calls “a paradigm shift” in training development.

2025-12-03T17:20:01.967Z

Sivasubramanian outlines how the newly-annoucned Amazon Nove Forge can go a long way to helping with this, providing businesses with a smoother and easier passage.

2025-12-03T17:17:56.897Z

There are a lot of challenges around customization, Sivasubramanian admits – and often this entails starting from scratch, which can be hugely expensive and inefficient.

2025-12-03T17:15:15.563Z

Next, there’s new model customization tools in Amazon SageMaker AI, which allows users to customize models in days instad of months.

2025-12-03T17:10:23.697Z

These are the big questions, Sivasubramanian says, unveiling Reinforcement Fine-tuning in Amazon Bedrock, which promises to improve model accuracy with advanced customization.

2025-12-03T17:05:34.094Z

Over-training can be a bad thing, he notes, as the model may become so focused on specifics, it loses focus on what made it effective in the first place.

2025-12-03T17:03:53.480Z

Sivasubramanian returns, and it’s time to talk about efficiency – the next big question, especially when it comes to latency and scale.

Model customization is the answer, he notes – and AWS wants to help turn your vision into reality.

2025-12-03T16:55:53.157Z

Now, we’re taking a look at how Blue Origin is using AWS for its space exploration work. It’s certainly a break from the normal use cases, as the company explains how it uses AWS AI technology for its automated landing and takeoff procedures.

2025-12-03T16:53:03.418Z

What you need is not just the memory of the past, but also the context of the current interaction, Sivasubramanian notes.

He reveals the new AgentCore Memory episodic functionality, which he says can build agents which truly understand user behavior and recognize patterns across similar situations.

“The more your agents experience, the smarter they become,” he notes,

2025-12-03T16:50:54.027Z

Now, Sivasubramanian wants to talk about what’s next for AgentCore, with new capabilities being announced here at re:Invent 2025.

Memory is the key differentiator going forward, Sivasubramanian notes, but can be something agents struggle with – especially when it comes to the when, or the why of decisions.

2025-12-03T16:47:38.599Z

Amazon Bedrock Agentcore looks to solve many of these problems, Sivasubramanian notes, providing the tools and services that keep everything running smoothly, while you focus on buidling breakthroughs.

He uses the examples of identity authentication, which can be a hugely complex process, but with agents to take much of the heavy lifting, developers can focus on other key tasks.

2025-12-03T16:46:13.413Z

With so much activity going on, how can companies see true results when it comes to development and PoCs?

Getting agents to production is still too hard, Sivasubramanian notes, and complexity slows down innovation.

2025-12-03T16:42:39.201Z

Sivasubramanian announces new expansions for Strands – support for TypeScript and Edge devices, hopefully unlocking new agentic capabilities.

2025-12-03T16:41:34.775Z

“In this new world, we believe how you build agents should be really simple,” Sivasubramanian says.

He highlights AWS’ Strands Agents SDK, which offers a whole new way at developing agents that saves time and stress across the board.

2025-12-03T16:39:57.843Z

Agents require a strong model, effective code and useful tools to bring them to life, Sivasubramanian says.

The true power of AI is making all of these work together effectively – but this is often a labor-intensive and tricky process, he notes.

2025-12-03T16:38:08.558Z

Surprise surprise – it’s agents!

Agents can go above and beyond what simple chatbots and services can offer, Sivasubramanian says, proactively searching for solutions and assistance.

2025-12-03T16:36:33.154Z

“The reality is, building and scaling these systems can be harder than the problems they solve, ” Sivasubramanian says.

So what’s the answer?

2025-12-03T16:36:01.425Z

We are no longer limited by the syntax of language, or the parameters of past systems, Sivasubramanian says.

“We are living in times of great change,” he says. “Agents give you the freedom to build without limits.”

He highlights customer use cases from ocean clean ups to brain analysis to show how effective AI agents can be in a wide variety of use cases.

2025-12-03T16:34:26.185Z

Sivasubramanian asks us to think back to earlier technological successes, and how this made us feel.

The possibility of building anything, of limitless creativity, is what’s really important, he says.

AI agents are helping builders around the world realise this feeling even more than ever, he says.

2025-12-03T16:31:47.938Z

Here we go – after a short video once again outlining the potential AI technology can bring to the world, the lights go down and Swami Sivasubramanian, AWS Vice President, Agentic AI, takes to the stage.

2025-12-03T16:21:49.713Z

So what should we expect from this morning’s keynote?

In a word (or two, depending on how pedantic you want to be) – more AI. Sivasubramanian is one of AWS’ top minds when it comes to all things AI, and often structures his keynotes around themes on how the technology can do good and benefit workers of all levels.

We’re also expecting more news and announcements today – possibly more from Kiro?

2025-12-03T15:54:07.239Z

We’re in and seated – passing by a Guiness World Record attempt out in the lobby for the world’s largest game of Rock, Paper, Scissors (which isn’t a sentence I ever thought I’d type…)

2025-12-03T14:29:17.385Z

Good morning from day two of AWS re:Invent 2025!

It’s another early start for us, so we’re off to get coffee and breakfast before this morning’s keynote from Dr Swami Sivasubramanian, Vice President, Agentic AI – starting at 8.30am PT.

2025-12-03T02:47:15.756Z

That’s a wrap on day one – but be sure to tune in tomorrow for even more news and updates as they happen.

2025-12-02T18:09:59.370Z

And with that salvo of news, it’s a wrap on the opening keynote of AWS re:Invent 2025! It’s been a back few hours, so we’re off to digest everything we heard, and take a trip around the show.

2025-12-02T18:08:58.226Z

It got a bit tight at the end, but Garman rattles through all 25 announcements in just a snifter under 10 minutes – an impressive end to all the news being revealed here.

As promised – here’s everything he covered…

2025-12-02T18:00:57.794Z

Garman is talking so fast, I’m not even going to try and keep up with everything being announced and talked about here – I’ll just try and get a photo of the sum-up slide at the end.

Garman is being kept honest by a shot clock – so eyes on the prize…

2025-12-02T17:58:58.480Z

“This is a big leap forward in the journey towards unlocking the value of AI,” he notes.

We’re onto the final stretch now, as Garman switches to everything else being announced at the show.

He’s planning to quickly rush us throught 25 new launches – in just 10 minutes.

Pray for the fingers of us live-bloggers…

2025-12-02T17:51:42.739Z

Finally, there’s AWS DevOps Agent, looking to simplify, resolve and proactive spot incidents before they happen.

The always-on service is able to continuously monitor your code and sources, giving in-depth insights and alerts to spot issues.

2025-12-02T17:49:25.044Z

There’s also a new AWS Security Agent, which helps users build applications that are secure from the start across AWS, multicloud, and hybrid environments.

Garman notes how AI-accelerated development can greater lengthen security strain, but the new agent can proactively scan code for vulnerabilites, and even carry out penetration testing, taking time-intensive tasks away from over-stretched security workers.

2025-12-02T17:45:16.494Z

Garman reveals Frontier Agents – a new kind of agents which are autonomous, scalable and long-running.

These include the new Kiro autonomous agent – which looks to help developers create faster and more effectively.

2025-12-02T17:38:43.529Z

That’s not all though – as Garman says there is a chance for all software development to be reimagined.

The company announced Amazon Kiro earlier in 2025, aiming to make coding simpler and more effective – and now, there’s some new expansions coming.

Garman reveals start-ups can now get a year of Kiro for free, and that Amazon itself is now relying on Kiro as its own in-house development platform.

2025-12-02T17:35:17.474Z

Garman is back, and it’s time to talk about developers – and especially, technical debt.

This is something costing companies billions of dollars every year – so how can AI help?

Garman reveals some new tools for AWS Transform, aimed at supporting all kinds of modernization, giving custom options to create code transformation agents for any code or API framework.

2025-12-02T17:32:06.556Z

Our final major customer case study is from May Habib, Chief Executive Officer and Cofounder, WRITER, who outlines how AWS is helping it boost and develop its AI agents.

The company is now bringing Amazon Bedrock guardails to its platform, giving customers much more oversight and control over their entire AI stack.

2025-12-02T17:28:34.280Z

Customer service is another key area Amazon knows a lot about, Garman notes – with its Amazon Connect platform offering a range of contact center services.

2025-12-02T17:21:47.002Z

Garman returns, and moves on to AWS’ own agentic solutions.

The company focused on areas where it feels it can offer real expertise, he notes – as a massive, multi-national business, it can provide real-world experience.

2025-12-02T17:12:27.993Z

“We’re clearly witnessing a golden era of creativiy, with AI amplifiying and enabling people to take the next step,” Narayen says, “we’re constantly pushing the boundaries of what’s possible.”

“When it comes to AI for creativity, we’re reimagining every stage of the process.”

Firefly and Photoshop are taking a leading role in amplifying AI, he notes, all backed up by AWS tools such as Amazon S3 and EC2.

2025-12-02T17:09:05.309Z

Time for another customer case study – and it’s a big one.

Shantanu Narayen, Chair and Chief Executive Officer, Adobe, takes to the stage to talk about how his company is using AI.

2025-12-02T17:08:05.551Z

2025-12-02T17:02:36.442Z

Garman likens building agents to raising children – you want to make sure they grow and prosper, but don’t get into trouble, and use the right tools.

Customers can struggle with this, but luckily AWS has the answer – Policy in AgentCore, which provides real-time, deterministic controls for how your agents interact with your enterprise tools and your data, making sure they stay in bounds.

2025-12-02T16:57:36.525Z

Garman returns, and it’s time to talk about agents.

“Agents are exciting, because they can take action and get things done,” he notes, moving on to talk about Amazon Bedrock Agentcore.

But with agents only set to play a bigger role in the future, what is next for AWS in this space?

2025-12-02T16:51:36.101Z

AWS and Sony have been working together since 2010, connecting up to 129 million gamers to connect and experience “Kando” together, he notes – with creators also given more tools and insight.

This Sony engagement platform brings fans and creators closer together, and needs tools to harness the wealth of data within the company’s data ocean using AWS platforms.

AI is also playing a key role, with Sony’s internal generative AI tools running on Amazon Bedrock.

2025-12-02T16:49:18.327Z

Now it’s time for our first major customer talk – John Kodera, Chief Digital Officer and Corporate Executive Officer, Sony Group Corporation, comes to the stage.

He wants to talk about “Kando” – a term which covers a range of definitions, but includes the feelings one has when watching a film or playing a game.

Sony has a proud history of reinventing itself, he notes, and is no longer just a hardware company, but so many areas of entertainment.

2025-12-02T16:45:53.379Z

Garman reveals Amazon Nova Forge, which lets customers build their own frontier model with access to “open training” models, mixing your data with Amazon-curated datasets, before deploying on Bedrock.

2025-12-02T16:41:30.777Z

Now we move on to your data – what makes your business special, Garman notes, and having the right model can make all the difference.

Customization is key, he notes – but training your own model can be super-expensive, and building from an open weights model can work, but these can have limits on how effective they can be.

So what’s the answer?

2025-12-02T16:33:44.128Z

Amazon Nova, which provides foundation models, is also getting an upgrade – an all new Amazon Nova 2, available in three tiers – scaling from everyday workloads up to more complex work, and a new speech-to-speech humanlike conversations.

Garman says Nova 2 stands up well in offering fast, cost-effective reasoning models for everyday workloads against the competition.

2025-12-02T16:31:20.334Z

Model choice is critical when building, Garman notes – and Bedrock offers a whole bunch to pick from – including the likes of DeepSeek, Meta, OpenAI and Writer.

But there’s more to come! New models from Google, Nvidia, Kimi and Minimax are making their debut – including some new offerings from Mistral.

2025-12-02T16:29:03.546Z

Now, we’re moving on to Inference platforms – a critical consideration for AI development.

“We’re seeing nearly every single application in the world being reinvented by AI,” Garman notes, with inference playing a key role.

Amazon Bedrock is playing a major part for customers across the world, he notes – so what’s new for the platform?

2025-12-02T16:26:31.355Z

Garman also reveals AWS is hard at work on Trainium 4, which will provide more performance, bandwidth and memory capability – but there’s no news on a possible release date just yet.

2025-12-02T16:25:47.431Z

Garman unveils one of the new racks on stage – it’s certainly…tall? Thr crowd goes wild for the new hardware though.

2025-12-02T16:24:16.620Z

There’s also a shout-out for Trainium, which Garman says is becoming increasingly popular for inference work, including Perplexity’s Claude AI service.

He also gives us a look at Project Rainier, where over 500,000 Trainium 2 chips are helping power and scale Claude’s AI development.

But that’s not all – Garman also reveals the new step-up – Trainium 3 – with new ultra servers now available, offering 4.4x compute over the previous generation.

2025-12-02T16:21:11.073Z

Garman says AWS wants to take this further – and announces AWS AI Factories. Essentially acting as a private AWS region, the new offering allows customer-specific AI infrastructure to be deployed directly into a customer’s data centers.

2025-12-02T16:20:58.730Z

Garman starts with a note on hardware – namely Nvidia GPUs.

He notes the two companies have worked together for 15 years, and AWS is “the best place to run Nvidia GPUs.”

“Nothing is too small for us to focus on – those details really matter,” he says.

A new P63 GPU offering will bring Nvidia’s GB200 and GB300 will bring the biggest organizations more power and efficiency – including OpenAI, which is using it to run and improve OpenAI.

2025-12-02T16:15:12.471Z

“Giving all of you the freedom to keep inventing is why we’re here today…every single industry is in the process of being reinvented,” he says – with AI playing a key role.

“But the true value of AI has not yet been unlocked,” he says – with AI agents now playing a key role.

“I believie the advent of Ai agents has brought us to an inflection point,” he says, “this will have as much impact on your business as the Internet or the Cloud.”

But doing all of this will require an all-new tech stack, Garman says.

2025-12-02T16:13:32.175Z

“So what motivates us?” Garman asks.

“We wanted to make it possible for every developer…to access the technology, infrastructure and capabnilities to build whatever they want.”

Garmn notes developers often struggle to have the time to get all of this together, and less time building – and this is what AWS wants to help solve.

2025-12-02T16:10:22.128Z

“But at Amazon, everythign starts with the customer,” Garman says, “the largest enterprises in ever single industry and vertical, are running their businesses on us.”

He has a particular fondness for his start-up customers, and it’s time to hear from some of them.

2025-12-02T16:05:49.791Z

Garman starts with a recap of the last year for AWS, which has been a big year – just the $132bn, in fact.

Over 500 trillion objects are now stored in Amazon S3, and Bedrock and other systems also saw a huge boost.

There’s also a shout-out for Ocelot, it’s quantum chip prototype.

This is all backed up a global resilient network, with 38 regions, with three more to come soon, Garman says.

2025-12-02T16:03:50.784Z

Apparently this keynote is being streamed on Fortnite…who would have expected that?

2025-12-02T16:02:22.113Z

Here we go! After a prety fun little light show, and brief video outlining how (surprise surprise) AI is going to help change the world, and AWS is powering this, the lights go down and AWS CEO Matt Garman takes to the stage.

2025-12-02T15:54:42.081Z

2025-12-02T15:21:23.174Z

So what should we expect from Matt Garman’s keynote? There will doubtless be a number of new product announcements, and also some special guests.

We’ve been told top execs from Sony and Adobe will be here, plus some surprises, we have no doubt…

2025-12-02T15:19:58.644Z

The opening keynote is one of the high points of the show, and there’s huge queues to get in – but we are in and seated now!

As always, there’s loud pumping tehcno to welcome us in, courtesy of DJ Kara…

2025-12-02T14:13:15.314Z

Good morning from Las Vegas! It’s day one of AWS re:Invent 2025, and we’re raring to go.

Time for some breakfast and coffee before Matt Garman’s keynote starts at 8am PT.

2025-11-30T14:06:22.127Z

Hello from Las Vegas! We’re all ready for a packed few days of AWS news and announcements.

Things will be starting off tomorrow with a keynote from AWS CEO Matt Garman, beginning at 8am PT, so tune in then to see all the updates as they happen.