your123 - stock.adobe.com
Amid ferment in the fast-growing market for generative AI technology, AWS is out with a new service that aims to make it easier for enterprises to access foundation models for text and images.
The cloud giant on April 13 introduced Amazon Bedrock, a service that gives enterprises access to foundation models from startups such as AI21 Labs, Anthropic and Stability AI through an API. Bedrock also enables enterprises to access Amazon Titan foundation models, including two new large language models.
In addition, AWS is steering customers to train foundation models with the vendor's silicon chip processors. Amazon EC2 Trn1n instances powered by AWS Trainium and Amazon EC2 Inf2 instances powered by AWS Inferentia2 are now generally available.
An expected need for the market
AWS' new service was expected, said Mike Gualtieri, an analyst at Forrester Research.
"This isn't just one AWS model that [enterprises will] have to use," Gualtieri said. Instead, the vendor is highlighting that users can also make their own derived models -- new models created from already established models.
"They're saying, 'Hey, you can stay if you want to do all of this cool AI stuff. You can stay right here and do it,'" he continued.
While Google and Microsoft enable enterprises to create their own derived models, the AWS competitors often don't emphasize that in their product releases. Rather, they focus more on the APIs being embedded in their applications and less on enterprises being able to make their own.
AWS is helping to fill a need in the generative AI market, according to Gartner analyst Arun Chandrasekaran.
Mike GualtieriAnalyst, Forrester Research
"The generative AI ecosystem is evolving quite rapidly," he said. "Companies building their own foundation models are striving for differentiation through domain specificity, transparency, cost, performance, privacy and accuracy. A broad suite of models enables providers to offer flexibility and choice rather than going with a one-model-fits-all approach."
Moreover, while Google and Microsoft appear to have an advantage in part because they're able to embed their generative AI APIs into business applications such as Microsoft Teams and Dynamics 365 CRM or Google Workspace, the demand for generative AI services is too high for two cloud providers to meet it all, Chandrasekaran said.
"There's a misnomer out there about all of these large language models that there's only going to be a few massive ones," Gualtieri said. "There's going to be thousands of them."
However, AWS might face some problems when integrating the third-party services from the smaller vendors.
"The challenges are more around the depth of integration for the third-party services and if they can make it as seamless as first-party services," Chandrasekaran said.
"Also, how AWS will position these AI models and companies relative to each other to create clearer swim lanes will be interesting to watch," he continued, referring to AWS integrating the third-party technologies without them clashing with each other.
Another issue is that AWS might not be ready to release a service like Bedrock.
AWS did not release pricing for Bedrock, but the vendor claims that the cost of generative AI technology is a barrier to entry for some enterprises.
"It looks to me like they may have rushed this out, with primarily third-party AI, no pricing, no info on model size, no info on training data sets," added Karl Freund, founder and analyst at Cambrian AI.
In addition, running Bedrock on its own AI chips might be challenging for AWS because AWS is less competitive than Nvidia and Google in AI hardware, he said.
"We will have to wait and see how well they inference with Inferentia and Trainium," Freund said.
Bedrock is available now in limited preview. Besides Bedrock, AWS also revealed that its AI coding assistant, CodeWhisperer, is now generally available for free to individual developers.
Esther Ajao is a news writer covering artificial intelligence software and systems.