Fresh off re:Invent 2024, probably the most memorable takeaway from this year’s sessions has got to be the importance, for such a massive tech player as AWS, of investing broadly to create major impact in multiple arenas at once—including making a significant dent in the competition.
As I mentioned in my AWS re:Invent expectations blog, one of my main concerns going in was that this year’s re:Invent would be wall-to-wall GenAI.
But this year’s offerings were a pleasant surprise. Yes, GenAI was very present, but Amazon’s focus was both broader and deeper. Broader in that they spotlighted an extraordinary range of cloud and DevOps topics, while new GenAI capabilities deepened the field considerably. The focus was not only on core AI applications like text generation and image analysis, but also on using AI to support migration, enhance security, and more.
Beyond AI, Dr. Werner Vogels, Amazon’s VP and CTO, shared an important message for attendees in his keynote: AWS’ newest frontier addresses something so fundamental that we take it for granted: time.
As Vogels put it:
“We always wanted to give you the building blocks for building the next generation of systems. Those were originally compute, storage, databases, network, and security… but now, we’ve added a more fundamental building block to all of this. And the fundamental building block is time. I urge you all to look at the algorithms, and the mechanisms that you’re building in your applications, and whether synchronized time can actually help you to significantly reduce the complexity of your systems.”
I’ll come back to this in a minute as I share other major AWS releases and takeaways from my 13th re:Invent.
Watch: My 5 biggest AI takeaways from re:Invent 2024:
Migration Still a Major Theme
Years ago, with much less experience in the world of enterprise IT, I figured the transition to cloud, and especially the IaaS revolution, would take between 5 and 10 years. Turns out I was off—by decades. Only about 20% of enterprise workloads today are in the cloud.
That’s why Amazon is still staking its claim in the migration market, turning the spotlight on companies that have gone all in on AWS cloud like Experian, the credit data giant that processes 1.8 billion updates a month, and Booking.com, the global travel enterprise making heavy use of AWS GenAI offerings like Bedrock and SageMaker.
And now, generative AI is here to help enterprises with the transformation. In his CEO Keynote, Matt Garman announced Amazon Q transformation capabilities for VMware workloads. These help organizations migrate VMware-based workloads to AWS—and do it “80 times faster” than manual approaches, cutting the process from months to hours or weeks.
Amazon Q handles the challenges of cloud migration by automatically identifying application dependencies and then generating comprehensive migration plans. According to Garman, this cuts migration time and risk by automating the conversion of on-premises VMware network configurations to modern AWS equivalents.
Garman also announced another AI-powered transformation capability of Amazon Q Developer: migrating Windows apps to Linux. Q Developer uses agents to identify incompatibilities, generate transformation plans, and refactor source code in parallel across multiple applications. According to Garman, this will help modernize applications up to 4 times faster and save up to 40% in licensing costs.
Put together, maybe these moves will help speed up that transition to cloud and IaaS that I’ve been looking forward to.
Disrupting Major AI Model Providers
When Microsoft announced its strategic investment in OpenAI, starting with $1B in 2019, many people in the industry joined me in watching as OpenAI released multiple ChatGPT versions and added multimodal and multi-agent capabilities. Many people started to wonder where AWS stood in the game and what their next move might be.
Then came Bard from Google (now Gemini), further intensifying the competition. AWS was also going through some upheaval with the transition of Andy Jassy from AWS CEO to Amazon CEO and the appointment of Adam Selipsky as the new AWS CEO. All of which left AWS seemingly late to the generative AI race.
But Selipsky was soon replaced by then-AWS sales chief Matt Garman, who revitalized AWS and restored it to its renowned pace of innovation.
This year’s re:Invent was a testament to the resurgence Garman has led, with numerous important announcements, including Amazon Nova, a new generation of Amazon proprietary foundation models integrated into Amazon Bedrock. Nova helps companies leverage AI for customer service automation, personalization, content creation, and other tasks. The Nova series are “the fastest models in their respective intelligence classes in Amazon Bedrock.”
Examples like Nova Canvas, an image generation model, and Nova Reel for video creation now position AWS to directly rival parallel tools from brands like OpenAI (such as DALL·E and Sora) and Google DeepMind (such as Gemini models).
Amazon Bedrock Innovations
As I mentioned after the AWS GenAI Analyst Summit earlier this year, AWS had some big reveals up its sleeve when it came to Amazon Bedrock. Now, they’ve been unveiled to the general public.
The launch of the Amazon Bedrock Marketplace was among the most exciting reveals at re:Invent 2024. It opens up model choice and discovery, giving developers access to over 100 foundation models from providers like IBM, NVIDIA, and Stability AI. Marketplace simplifies the exploration, testing, and deployment of AI models tailored to various industries with built-in AWS security and governance.
Beyond Marketplace’s capabilities, other expansions of Amazon Bedrock include intelligent prompt routing, which predicts which model can optimally handle a request, while prompt caching reduces latency and costs by storing frequently used prompts. Together, these features offer a more streamlined AI experience, especially when it comes to scaling up genAI workloads.
Showing off the capabilities of Amazon Bedrock, AWS put together a demonstration that tapped into one thing most AI fans have in common: coffee.
Using Bedrock’s multimodal processing and structured querying capabilities, AWS created a personalized virtual assistant that analyzed customer preferences and generated tailored recommendations. The AI agent handled complex customer inquiries, recommended coffee blends based on taste preferences, and tracked orders. This underscored key Bedrock features to build in operational efficiency along with customer engagement and trust.
Two more intriguing upgrades to Amazon Bedrock include model distillation, which lets developers tap into faster, more cost-efficient models with no degradation in performance, and multi-agent collaboration, enabling the deployment of AI agents that collaborate on multi-step, specialized tasks.
Finally, I’ve kept my eye on the problem of hallucinations since GenAI was first available to the public, so I was especially pleased to see the preview launch at re:Invent of Automated Reasoning, which aims to prevent factual errors and hallucinations as part of Amazon Bedrock Guardrails.
All these new features and capabilities will go a long way towards democratizing AI, although some of the challenges I mentioned after the AWS GenAI Analyst Summit still remain, like concerns about siloed efforts if businesses don’t adopt a carefully structured adoption strategy.
Chip-to-SaaS Value Chain
The Trainium 2 AI chip also made its debut at re:Invent 2024, while another proprietary chip, Trainium 3, was announced and is expected by late 2025. The Trainium 2 has four times the processing power of first-generation Trainiums and 96 GB of high-bandwidth memory (HBM); both chips are expected to enable AI model training and inference at a lower cost.
Of course, AWS isn’t the only one directly challenging Nvidia in the AI hardware space. Both Microsoft and Google were first to the market, with Maia and Cloud TPU, respectively. But it seems to me that AWS has not only put the most consideration into the scale and performance of running AI workloads but has also matured their play in this field to the point that they can start driving costs down to become even more competitive.
After last year’s re:Invent, I told you about the new $4B investment deal AWS had signed with Anthropic.
This year, we got to see that collaboration starting to bear fruit, as AWS and Anthropic unveiled Project Rainier, a next-generation UltraCluster supercomputer architecture powered by hundreds of thousands of Trainium chips. AWS has said that when finished, they expect it to be “the world’s largest AI compute cluster reported to date” and will be able to meet customers’ need to train and run massive AI workloads, including LLMs with trillions of parameters. I’ll definitely be keeping my eye on this, and I’m eager to learn more about actual use cases.
A New Key Primitive: Time
Last but not least, as teased by Amazon CTO Werner Vogels, comes the next big primitive: time.
In AWS-speak, “primitive” refers to cloud service “building blocks” like network, database, compute, and storage that can be combined and orchestrated with maximum flexibility. That’s where AWS has always focused its offerings. But as Vogels pointed out, there’s one primitive that’s been largely missing from the mix.
Maintaining accurate time synchronization in database systems distributed across multiple regions is crucial for ensuring transaction consistency and proper ordering. Traditional databases often rely on timestamps to order transactions, but in a globally distributed system, clock drift becomes a significant challenge.
As Werner explained in his keynote, it’s almost completely infeasible to guarantee that clocks in different regions are perfectly synchronized. This can lead to inconsistencies in transaction ordering and potential data corruption.
As use cases for global active-active databases offering high availability and concurrent transactions expand, the problem of time drift will grow increasingly acute.
To meet this need, AWS rolled out Amazon Aurora DSQL, an active-active distributed database to rival Google’s Spanner. Aurora incorporates the Amazon Time Sync Service for the atomic-level accuracy this architecture demands. Amazon rolled out its Time Sync Service in 2017, incorporating a hardware reference clock in every EC2 instance worldwide that synchronizes with satellite–connected atomic clocks.
Now, with Aurora DSQL, users get virtually unlimited scale, the highest possible availability, zero infrastructure management—and microsecond-level accuracy worldwide. And as Vogels pointed out, precise clocks reduce complexity significantly.
Time Sync allows Amazon to ensure strong data consistency across regions in both Aurora DSQL and its existing DynamoDB. Of course, while Time Sync helps mitigate the clock drift problem, other consensus algorithms like Paxos and Raft are also still necessary to maintain consistency and order in distributed systems.
Q: What is an atomic clock?
A: An atomic clock is a super-accurate clock that measures time using the vibrations of atoms, like a tiny, super-steady pendulum. It’s crucial for technologies like GPS, internet systems, and database systems, where precise time synchronization ensures data consistency and smooth operations.
Final notes
Every year, AWS re:Invent provides one of the deepest glimpses into where the cloud industry is heading. This year was no exception. I think that, like me, a lot of attendees went in jaded and dreading the specter of wall-to-wall AI and came away pleasantly surprised at how AWS manages to go beyond, not only leading the pack but also zig-zagging enough to keep things interesting.
Werner Vogels’ positioning of time as a primitive is a key example of that because most developers—though certainly frustrated with time drift—accepted it as a given when it doesn’t have to be. This willingness to challenge the givens of the industry is what keeps AWS in the lead.
Here I am at re:Invent with my amazing colleague, Alberto. We had a blast!
Need AWS Partner Network (APN) marketing content? Want to get through the AWS approval process faster? Schedule a free consult.
