AWS is close to a $100 billion revenue run rate, Amazon CEO Andy Jassy said on a Q4 earnings call – as retail shoppers and enterprise cloud services-guzzlers powered the company to $575 billion in 2023 revenue.
“The lion's share of cost optimization has happened… migrations have started to pick up again [and], larger new deals also accelerated” Jassy said of AWS – citing deals with BMW, NVIDIA, LG, Hyundai, and Merck.
(AWS revenues were $24.2 billion in Q4, up 13% year-on-year.)
Generative AI was an inevitable talking point on the earnings call and the company is snacking increasingly enthusiastically on its own dogfood.
AI, said Jassy, gives Amazon “the chance to reinvent so many of our customer experiences and processes, and we believe it will ultimately drive tens of billions of dollars of revenue… over the next several years.”
Amazon launches “Rufus”
Just this week, for example, Amazon launched Rufus, a generative AI-powered “expert shopping assistant… that represents a significant customer experience improvement.” Rufus was trained on Amazon’s vast corpus of content including product catalogues, customer reviews, community Q&As and web content. It launched in beta on February 1 to a “small subset of customers in Amazon’s mobile app” Amazon said.
“We're building dozens of Gen AI apps across Amazon's businesses, several of which have launched” Jassy said, “others are in development.”
But as customers are learning, getting this right is no mean feat.
“It’s a very iterative process and real work…”
“It's a very iterative process and real work, to go from posing a question into a chatbot and getting an answer, to turning that into a production-quality application at the quality you need for your customer experience and your reputation – and then also getting that application to work at the latency and cost characteristics that you need,” Jassy said.
(That’s an experienced emphasised by interviewees in a recent feature by The Stack on building out Retrieval Augmented Generation applications.)
See also: "You can end up in 'RAG hell'
On a Q&A with analysts, Jassy added: “Customers want choice… They want to experiment with all different sized models because they yield different cost structures and different latency characteristics…
“So Bedrock is really resonating with customers. [They want to] figure out what works best for them, especially in the first couple of years where they're learning how to build successful generative applications is incredibly important for them,” the CEO added – suggesting AWS was a more flexible partner than a Microsoft with its off-the-shelf Copilots.
Amazon’s approach to AI: Triple-threat
The company has highlighted a threefold approach to AI; essentially built around a) IaaS: Its own custom semiconductors as a managed service for training and inference of LLMs; 2) PaaS: Managed services like Bedrock, which let customers choose and run an LLM of choice in the cloud; 3) SaaS: Applications like its Amazon Q, et al; as announced at reInvent.
See also: AWS launches AI assistant “Amazon Q”, offers fine tuning for models on Bedrock
“Many [customers] are still thinking through [which three] layers of the stack they want to operate in… I think many of the technically capable companies will operate at all three,” Jassy said on the Q4 earnings call.
Other AWS AI news this week came with the integration of the Amazon Q AI assistant with AWS Glue, a data integration service. As Swami Sivasubramanian, Amazon VP, Database, Analytics and Machine Learning put it on LinkedIn, that integration lets customers use natural language to:
🟠 build data integration jobs faster
🟠 reduce the complexity of troubleshooting issues
🟠 get instant data integration SME help
He added: “Getting an end-to-end data integration job is as easy as asking Amazon Q to, “read data from S3, drop null records, and write the results to Redshift.” This is another step in our journey to simplify every phase of your building experience on AWS with analytics, AI/ML and generative AI.”
This will even mean that “once a job is authored, orchestrated, and deployed into a production system” Q can be used to help troubleshoot failures, and fix errors, using natural language, product documentation suggests. If performant, the integration could be hugely powerful for customers and render some AWS skillsets increasingly superfluous.
“You can ask Amazon Q for Data Integration to ‘read JSON files from S3, join on ‘accountid’, and load into DynamoDB’ and in response, it will return an end-to-end data integration job that can perform that action. You can review the generated job, test it against sample datasets, and move it to production” a product page suggests. The AI behind this has been “trained, and fine-tuned with domain knowledge base of AWS Glue.” Amazon is confident that it has got its chunking strategy right…
