AWS thinks custom models are its secret weapon in the AI wars

  • Amazon CEO Andy Jassy talked up the company's recently launched AI Custom Model Import tool

  • Analysts and industry experts alike have predicted a forthcoming wave of custom AI models

  • Amazon isn't the only one making moves on this front

It might not have seemed like such a big deal at the time, but Amazon CEO Andy Jassy argued the company actually dropped a bomb on its rivals with the launch of its custom AI model import tool a few weeks back.

Speaking during Amazon’s recent earnings call, Jassy said the debut of Custom Model Import for Bedrock was a “sneaky big launch as it satisfies a customer request we've heard frequently and that nobody has yet met.”

Basically, the tool allows customers to import custom models they’ve built in Amazon SageMaker into the hyperscaler’s Amazon Bedrock platform. This lets enterprises utilize AI investments they’ve already made while also leveraging Bedrock’s capabilities to scale their models and applications.

The import tool is still in preview, but Jassy indicated it’s already received a positive response.

“Customers are excited about this, and as more companies find they're employing a mix of custom-built models along with leveraging existing LLMs, the prospect of these two linchpin services in SageMaker and Bedrock working well together is quite appealing,” the CEO said.

Amazon, of course, has an incentive to toot its own horn but there is some truth to Jassy’s comments about demand for custom models.

Gartner has predicted spending on AI software will rise at a 19.1% CAGR to hit $297.9 billion by 2027. And on the Generative AI front, the research firm forecast more than half of GenAI models will be tailored to a specific industry or business function by 2027. That’s compared to 1% of GenAI models in 2023.

And we’ve heard from other industry experts about how the future will be all about custom models rather than monolithic large language models.

For instance, Red Hat CTO Chris Wright told us in January that there will be thousands of AI models to go along with the thousands of enterprise applications that are out there. He added these models will "be a combination of enterprise-built models as well as partner-provided commercial models."

But Amazon isn’t necessarily the only one who has picked up on this. Google Cloud allows enterprises to build and train their own models on its Vertex AI platform using a tool called AutoML. It also lets customers replace the default models in certain ML Kit APIs with custom TensorFlow Lite models.

Meanwhile, Microsoft allows customers to build and train their own models using AI Builder – and that includes pulling in models that were built outside of AI Builder using the Azure Machine Learning platform.

So, who will win out in the custom model race? We’ll have to wait and see.

Don't miss our upcoming Cloud-native 5G virtual event, which will feature speakers from AWS, Microsoft, and Google Cloud. Register for your free pass here.