AWS Re:Invent 2023 unpacks new moves, directions and announcements from AWS that round off a lot of areas for the Cloud major as it chases the new AI-led future
One CEO was wearing a black leather jacket and another a dapper suit, and both signified the personalities that these companies underline in the game of AI. When Jensen Huang, CEO, Nvidia came on stage to share some announcements with Adam Selipsky, CEO, Amazon Web Services at the latter’s keynote for Re:Invent 2023- one thing was pretty clear- the future is about handshakes and snow-globes, not boxing rings or race-strips- hopefully.
The current state of the industry is a far cry from the usual model where each big player is driving in its own lane to win the race. Looks like there is a shift towards collaboration, integration and co-builds – making everyone come together in a ‘sphere’, which, incidentally or not so, is the latest attraction at Las Vegas.
New Rabbits from AWS’s hat
A string of announcements, panels and conversations at this event hint strongly at that shift. Specially as AWS expands its partnership with Salesforce on one front and with Nvidia on the chip side. The collaborative stance is also visible in the way AWS has been working together with models from AI21, Anthropic, Cohere, Meta, and Stability AI.
Also remarkable is how AWS is trying to nail every vertical possible with enterprise-grade Generative AI offerings (As seen with industry use-cases shared from Pfizer, Kone, BMW and more at the keynote). AWS is also polishing rough corners on many areas like foundational models, processors, custom silicon, infrastructure layers of AI and more as we can see in the key launches shared at this event. Most launches are available for preview and will be generally available in near future.
AWS would be expanding partnership with Nvidia for Gen AI workloads, Selipsky said; followed by Huang saying how AWS was the first Cloud to recognize the importance of GPUs and how Nvidia would be bringing its DGX Cloud (its own AI factory like thing) to AWS.
“With AI, there will not be one provider or model. What would matter is choice. And the events of the past few days have made it very clear. We want to make it easy to move between a range of foundational models.” Selipsky quipped.
He was also later joined by the CEO of Anthropic talking about training Machine Learning workloads.
Hinting at some big names on the fragile issue of security, Selipsky also underlined how AWS will make sure customer’s data is not used for underlying models and how it will strengthen the spirit and letter of responsible AI.
AI- Both bones and flesh
AWS has announced capabilities and ambitions that show how it will attack both the hardware as well as infrastructure and tools aspects of AI for gaining competitiveness.
It launched Graviton4 – calling it the most powerful and energy-efficient AWS processor to date. As claimed- Graviton4 provides up to 30 per cent better compute performance and 75 per cent more memory bandwidth than current generation Graviton3 processors- for performance and energy efficiency for a broad range of workloads running on Amazon Elastic Compute Cloud (Amazon EC2).
It also unboxed its second-generation Trainium2 chips for four times faster training than first-generation Trainium chips. These will be able to be deployed in EC2 UltraClusters of up to 100,000 chips, making it possible to train large language models (LLMs) and other foundation models (FMs) and do so in a fraction of the time, while improving energy efficiency almost twofold, the company states.
What’s interesting is the part of customisation for different needs. AWS has announced that customers have more options to customise models in Amazon Bedrock with fine-tuning support for Cohere Command, Meta Llama 2, and Amazon Titan multimodal models, with Anthropic Claude coming soon.
Among other things, AWS also announced Amazon Q—a new type of gen AI assistant that is specifically for work and can be tailored to a business.
Then, there was the buzz about Guardrails for Amazon Bedrock, which AWS says, will help customers implement safeguards customized to their generative AI applications and aligned with their responsible AI principles.
Earlier at the event, Peter DeSantis, senior vice president of AWS Utility Computing unveiled new serverless innovations like Elastic Cache serverless, Redshift serverless with AI-driven scaling and optimisation. “At AWS, it’s always day One”, he stressed, as he also shared some interesting work happening on error correction of qubits with AWS’s quantum computing efforts.
The Helicopter Gaze
Interestingly, Selipsky began his keynote saying- “Reinventing is in our DNA, that’s how we started with Cloud Computing. Today we have 3 times more data centres than the next largest Cloud Provider; and 60 per cent more services, 40 per cent more features. No other cloud provider gives the kind of footprint of availability zone for every region, with the level of separation, that we do. We were the first Cloud provider to start with something like Graviton in processors. We were the first one to bring GPUs to the Cloud. We reinvent so that our customers can.”
The event – which leaned heavily towards AI-related stuff (and not so much about Cloud) also entailed some demos to show new use-cases in works- like predicting moves in sports, enabling easy conversational interfaces for managing drones in various industries, and building AI apps with Gen AI ease.
AI has been the flavour of the year, and of the last few days, for sure. It’s no surprise that AWS is betting big on AI on so many fronts.
Jacket or suit- what was common for both CEOs were the choice of footwear – they both wore sneakers. Something that corresponds to the pace and hunger with which these players are running towards AI.
As DeSantis said, “When we are pushed in a new direction, our usual response is ‘let’s go’.”
Let’s see how far and fast they sprint, and in what direction. For now, AWS has set the ‘sphere’ rolling.