Amazon introduces custom AI capabilities in race against cloud rivals

Summary
The company’s cloud-computing unit hasn’t made as big a splash as its rivals in AI, betting instead that businesses want to use a variety of AI models.Amazon.com will let companies use their own generative artificial-intelligence models inside its AI app development platform, part of a package of moves the cloud-computing giant is launching to keep up with its rivals in AI.
Amazon Web Services said “tens of thousands" of businesses are using Bedrock, its AI app development platform. Giving companies the ability to add DIY models to Bedrock makes it easier for enterprise developers and data scientists to work together, said Swami Sivasubramanian, vice president of AI and data at AWS.
The Seattle-based company also released two new AI models on Tuesday, its Titan image-generator model, which can create images from text, and its Titan text-embeddings V2 model, which is designed for things like Q&A chatbots and making personalized recommendations.
As companies test and use generative AI for various tasks, many are building their own AI by customizing a vendor’s model, or tailoring an open-source model with their own data. Generative AI is expected to drive global information-technology spending to $5.06 trillion this year, an 8% increase from 2023, according to market research and advisory firm Gartner. This adds to the urgency among cloud providers, software and device makers to offer AI-based products and services to enterprise clients.
Amazon has fallen behind tech rivals in the AI race, though it has been trying to boost its standing with new offerings at AWS and its retail operations, The Wall Street Journal previously reported. AWS doesn’t have a defining AI partnership such as Microsoft’s with OpenAI, or Copilot, Microsoft’s generative AI assistant for its business software. In November AWS introduced Amazon Q, an AI chatbot for companies and developers, and it offers the Titan models, but they are not as well-known as Google’s Gemini chatbot and models.
Like its rivals, AWS offers companies using its platform a sort of one-stop shop for AI, cloud-computing, and data and software services. The company has focused on helping developers build generative AI applications—a harder task than building software apps—and giving customers the choice between many models, Sivasubramanian said.
So far, AWS has positioned itself as a neutral provider of AI technology, making a wide variety of AI models—from its own to proprietary models from Anthropic and open-source models like Meta Platforms’ new Llama 3—available through Bedrock. Its model evaluation tool, made fully available on Tuesday, will cut down the time businesses would have spent testing and analyzing different models, Sivasubramanian said.
Microsoft and Google also let customers use AI models from other companies, and open-source models from Meta and Paris-based startup Mistral AI.
AWS’s AI strategy could work while businesses are still deciding which vendors and technologies to use, and which actually provide a return on investment, said Steven Dickens, vice president and practice leader for cloud at Futurum Group. “They’re not as invested in their own models as Microsoft and Google are," he said. “AWS is more of a home for every model."
Three months after Bedrock became generally available last September, a majority of AWS customers were using more than one model to build AI applications, the company said.
“No one model will rule them all, and now you can see practically every cloud provider has pivoted their strategy towards the same," Sivasubramanian said. “Customers do not want to get locked into a single model because this space is so early."
Sheldon Monteiro, chief product officer of Publicis Sapient, said the consulting firm encourages customers to keep their options open when choosing AI vendors. While businesses may already have a contract or committed spend with certain cloud platforms, they should still “plug and play different components from providers," Monteiro said.
Still, businesses prefer to stick with AI services from the cloud provider they already work with. It is easier to apply generative AI in cloud platforms where corporate data is already stored, some chief information officers say. The trend suits AWS, the world’s largest cloud provider, because when businesses use AI services, they spend more on cloud.
Kamran Ziaee, Verizon Communications’ senior vice president and CIO for technology strategy and global infrastructure, said the telecommunications company is using some of Google’s generative AI services as it moves more company data onto Google’s cloud platform.
“The foundation of a good AI platform is really dependent on very good data," Ziaee said. “That will be the foundation of everything that we’re going to build on, from an analytics perspective, insights, AI engines and machines."
Some CIOs also don’t want to work with too many AI providers, as doing so adds to an already heavy vendor management load. Brad Lightcap, OpenAI’s chief operating officer, previously told the Journal that the company increasingly hears businesses “really want to go with one partner."
“We don’t want to have too many [large-language models]," Verizon’s Ziaee said, adding that just one or two are enough for all of its needs. Verizon will use a greater number of small AI models for specific tasks, he added, while limiting its use of the largest AI models, such as OpenAI’s GPT-4 and Google’s Gemini.
In the flurry of AI news from Microsoft and Google in the past year, Amazon’s generative AI approach hasn’t been quite as flashy or well-defined, said Jim Hare, Gartner’s lead Amazon analyst. That can make it hard for businesses who aren’t as familiar with generative AI to understand Amazon’s value.
“They’re not viewed as always being an innovator because they are heavily promoting partner solutions," Hare said. “There’s a risk that you can be marginalized because it’s becoming easier to move data out of AWS into other platforms."
Write to Belle Lin at belle.lin@wsj.com