AWS re:Invent 2023: All the day-three keynote announcements

The keynote theatre here at AWS re:Invent is filling up rapidly ahead of the opening session with Adam Selipsky. A torrent of attendees making their way into the hall. 

There certainly is a buzz in the air this morning. Anticipation ahead of arguably the biggest event of the year in Las Vegas. 

We're being treated to a heavy rendition of Thunderstruck by AC/DC. 

A fitting track given the sheer scale of this event so far in Las Vegas. Not long to go until Adam Selipsky takes the stage. 

While we're waiting, check out some of our coverage from AWS re:Invent so far this week. 

• AWS eyes AI-powered code remediation in Amazon CodeWhisperer update

• Amazon Detective offers security analysts their own generative AI sidekick 

PLUS: Why Amazon's Thin Client for business is an IT departments dream. 

It's likely we'll hear a lot about Amazon Bedrock today, the tech giant's LLM framework launched in April. 

Bedrock has been a roaring success so far, by all accounts, enabling customers to access a range of both in-house and third-party foundation models. 

Just months after launch, Bedrock had attracted "thousands of new customers", according to an AWS exec. 

There's a chance, albeit a small one, that AWS could unveil the launch of 'Olympus', its own 2 trillion-parameter large language model, this week. Speculation was rampant earlier this month amid claims the company was working on the model. 

If they were going to unveil Olympus, it would likely be here. 

AWS chief executive Adam Selipsky is on stage now, welcomed with a huge applause from delegates. 

Selipsky begins his keynote by thanking customers and partners, highlighting the broad range of organizations the cloud giant currently partners. 

Salesforce, in particular, just got a shoutout. AWS and Salesforce announced a huge expansion of their ongoing partnership yesterday. 

Salesforce's Einstein platform is now available on Amazon Bedrock. 

But it's not just large enterprises AWS is working with, the "up-and-comers" such as Whiz are working with AWS. 

Over 80% of unicorns globally run on AWS, Selipsky says. The cloud giant values its relationships with startups and dynamic young enterprises. 

"Enterprises and startups, universities, community colleges, non-profits, the cloud is for anyone," Selipsky says. 

"They count on us to be secure, to innovate rapidly."

So why are enterprises choose AWS? 

For a start, they're the oldest, most established cloud giant. They're the most secure. 

But they also 'reinvent', Selipsky says. The firm is continually exploring how it can take things to the next level and drive innovation. 

"We invented infrastructure from the ground up. Because we look at things differently, our globally infrastructure is fundamentally different from others."

AWS' infrastructure regions span 32 existing regions at present, and there are plans for more ahead. 

AWS has three times the number of data centers compared to the next largest cloud provider. Huge capacity and capabilities. 

AWS also offers 60% more services and 40% more features for customers. 

Selipsky highlights Amazon S3 storage as a prime example of its pattern of continuous innovation over the years. 

S3 Glacier Deep Archive storage was the next iteration, follows by Amazon S3 Intelligent Tiering. 

"Intelligent Tiering has already saved AWS customers over $2 billion," Selipsky says. 

Selipsky says there's still more to come for S3 storage, however. This is only the beginning for S3. 

We have our first announcement of the day here at AWS re:Invent. 

Amazon S3 Express One Zone is now generally available. Up to ten times faster than S3 standard storage, Selipsky said. 

This faster compute power will enable users to cut costs by up to 50% compared to S3 standard. 

We've moved on to general purpose computing. Selipsky points back to the launch of Graviton in 2018. 

This service has evolved rapidly in recently years. Today, more than 50k customers used Graviton to realize price performance benefits, he notes. 

And our second announcement of the day. The launch of Graviton 4. 

Selipsky says Graviton 4 is the most powerful and energy efficient chip we've ever built. 

40% faster for database applications compared to Graviton3 and 30% faster than Graviton3. Also 45% faster for large Java applications, Selipsky adds. 

We're onto generative AI now. Expect some bombshells here. 

"GenAI is the next step in artificial intelligence and it is going to redefine everything we do at work and at home."

"We think about generative AI as having three macro layers," Selipsky says. 

"The bottom layer is used to train foundation models and run them in production. The middle layer provides the tools you need to build and scale generative AI applications.

"Then at the top we've got the applications that you use in everyday operations,"

Selipsky points to AWS' long-standing relationship with Nvidia as a key example of how the company focused on the bottom infrastructure layer that underpins all generative AI innovation.

"Today, we're thrilled to announce we're expanding our partnership with Nvidia," he says. 

Nvidia CEO Jensen Huang has joined Selipsky on stage to give us an insight into this partnership expansion. 

Huang confirms the deployment of a whole new family of GPUs. The brand new H200, is the big one here. 

This improves the throughput of large language model inference by a factor of four, Huang says. 

"32 Grace Hoppers can be connected by a brand new NVLink switch," Huang says. "With AWS Nitro, that becomes basically one giant virtual GPU instance."

AWS and Nvidia are also partnering to bring the Nvidia DGX Cloud to AWS. 

"DGX Cloud is Nvidia's AI factory," Huang says. "This is how our researchers advance AI. We use our AI factories to advance our large language models, to simulate Earth 2 - a digital twin of the earth."

Huang says the company is building its largest AI factory. This is going to be 16,384 GPUs connected to one "giant AI supercomputer", he adds. 

"We'll be able to reduce the training time of large language models in just half of the time," Huang says. This will essentially reduce the cost of training in half each year. 

Simply mind boggling numbers here. And incredible compute power from Nvidia, all underpinned by AWS as part of this ongoing relationship between the duo. 

Now we're talking about capacity, Selipsky says. It's critical that customers have access to cluster compute capacity when training AI models. 

But fluctuating levels of demand means they need short-term cluster capacity, which is an issue that no other cloud provider has addressed so far - except from AWS, Selipsky says. 

Amazon EC2 Capacity Blocks for ML was announced a few weeks ago. This enables customers to scale hundreds of GPUs within one cluster. This will give them the capacity when they need it, and only when they need it, he adds. 

Costs are a major consideration for organizations training AI models at present. The costs incurred by firms during training have been rising rapidly over the last year or so. 

"As customers continue to train larger models, we need to keep pushing and scaling" Selipsky says. 

Another announcement. The launch of the new AWS Trainium 2 chip for generative AI and ML training. 

This is optimized for training FMs with hundreds of billions - to trillions - of parameters. It's also four-times faster than AWS Trainium. 

We've been through the underlying infrastructure work that AWS is currently focusing on. Now we're moving onto the tools. Expect some Bedrock announcements. 

"We know that many of you need it to be easier to access powerful foundation models and quickly build applications out with security and privacy in mind," Selipsky says. 

"That's why we launched Amazon Bedrock. 

"Customer excitement has been overwhelming," Selipsky says. More than 10,000 customers globally are using the Bedrock platform. 

Organizations spanning a host of industries. From Adidas to Schneider Electric and United Airlines and Philips. 

It's still early days though. There are new developments every day with generative AI, Selipsky says. This experimental stage at many enterprises is producing some great, tangible use cases of how generative AI can support organizations. 

The flexibility of Bedrock and the availability of models through the framework is the best setup for customers, Selipsky says. 

The events of the last ten days means that offering customers flexibility has never been more important. 

Bedrock gives access to foundation models from Anthropic, AI21 Labs, Meta's Llama 2, Cohere, and more. 

AWS' relationship with Anthropic has been a key talking point in recent months. 

The cloud giant has invested billions in the AI startup alongside Google. AWS is also Anthropic's primary cloud provider, underpinning all the AI innovation at the startup. 

Anthropic CEO and co-founder Dario Amodei has joined Selipsky on stage now. 

The duo are discussing the close relationship between the two firms. Hardware is a key talking point here. Anthropic is relying on Trainium for inferentia

Anthropic recently announced Claude 2.1, Amodei says. The token capacity in Claude 2.1 is seriously impressive compared to the model's previous iteration. 

Hallucinations have also been reduced by nearly half. This is a major breakthrough, Amodei says, providing customers with more peace of mind over potential AI risks. 

You can read more about Claude 2.1 in our coverage from last week. 

"We've put a lot of work into making our models hard to break," Amodei says. 

Safety and security is a major focus at Anthropic. Interpretability is also a key focus at the startup right now. 

Anthropic says the company views the current AI 'race' as a "race to the top" - but it's not what you think. 

This is about striking a lead in all the right ways, taking into account safety and responsibility. This enables Anthropic to set an example for the rest of the industry, he says. 

"The innovation around generative AI is explosive," Selipsky says. The company is excited about its own foundation models, the Titan models. 

"We carefully choose how we train our models, and the data we used to do so," Selipsky says. 

There are already multiple models in the Titan family. Titan Text Lite, Titan Text Embeddings Model, and Titan Text Express. 

All of these foundation models are "extremely capable" Selipsky says. But the critical factor here for organizations is deriving value from internal company data to maximize the use of generative AI tools. 

That's where Amazon Bedrock supports customers. It allows companies to access specific models based on their own specific business use cases. 

Delta Airlines is using Bedrock to drive AI innovation in customer support operations, Selipsky says. This is a prime example of a large organization leveraging the framework to create tools based on their unique individual needs. 

"Delta's been testing with different foundation models in Bedrock, including Anthropic Claude," Selipsky adds.  

"At the end of the day, you want FMs to do more than just provide useful and targeted information," Selipsky says. Fundamentally, we need generative AI tools to make decisions for use in some capacity. To act proactively. 

That's why AWS launched Agents for Amazon Bedrock recently. This is now generally available. 

We're onto security now. A key talking point in the last year amidst concerns over data leakage. 

Selipsky fires a cheeky broadside at OpenAI and Microsoft. Security is top of mind for AWS, he says. And fundamentally, the company has trust in both itself and its partners. 

The same can't quite be said about certain industry competitors...

On the topic of security and responsibility, Selipsky announces the launch of Guardrails for Bedrock. 

This enables users to safeguard generative AI applications. Users can easily configure harmful content filtering processes based on internal company policies. 

"We're approaching the whole concept of generative AI in a fundamentally different way," Selipsky says. Responsibility, security, and trust are top of mind for the cloud giant. 

On the topic of AI security, Pfizer chief technical officer Lidia Fonseca is live on stage now to run us through the pharma company's relationship with AWS and how the company is harnessing generative AI across its research operations. 

This has been a long-standing collaboration between the two companies dating back to 2019. Pfizer's move to the cloud in 2021 was supported by AWS. 

This saw the company move 12,000 applications and 8,000 servers to the cloud in just 42 weeks. One of the fastest, and largest, cloud migrations in history.

And all of this was done during arguably one of the most challenging periods in recent history - the pandemic.