Amazon’s Bedrock success prompts a surprising question: Do you need a top model to win in AI?

Amazon executives Matt Garman, Andy Jassy, and Werner Vogels.

So far, the generative AI race has been about who can build the most powerful models.

Amazon missed out on this. But it might not matter.

While OpenAI, Google and others joust over wonky model benchmarks, Amazon’s all-important cloud business is trying to be a neutral platform for AI developers not tied to any single model.

At the core of this approach is Bedrock, a service that gives access to multiple third-party large language models, such as Anthropic’s Claude, Meta’s Llama, and Cohere offerings.

Bedrock’s success is important as Amazon believes AI models are not in a winner-take-all market. It’s a different approach than OpenAI, which focuses on its own proprietary models. Microsoft has also closely partnered with OpenAI and relies heavily on the startup’s GPT models. Google has crafted its AI strategy mostly around its in-house Gemini offering, although it offers other models, too.

And just over a year into its launch, Bedrock is proving to be a solid new business, according to investors and internal data obtained by B-17.

Matt Murphy, a partner at Menlo Ventures, told B-17 it would be a “fool’s errand” for AWS to try to develop a model as powerful as Anthropic’s Claude at this point. Instead, it’s smart that AWS is closely partnering with Anthropic and using Bedrock, which gives access to Claude and other underlying infrastructure services to grow its AI business, he added.

“Stick with your strength, find a great partner, and execute to get the most market share. Distributing and featuring the best model on the market, Claude, is a competitive weapon for them,” Murphy said.

This is part of a new stage in the generative AI boom that began when ChatGPT took the world by storm in late 2022. Almost two years later, there are so many powerful AI models out there, it’s no longer that special. Spending another $200 million or more for six months to create yet another one of these models might not pay off as much now.

An Amazon S-team goal

AWS’s spokesperson Patrick Neighorn told B-17 that “tens of thousands” of customers are using Bedrock, including Intuit, Toyota, and the New York Stock Exchange.

Amazon’s “strategy of providing a choice of leading foundation models and a broad set of capabilities is resonating with customers,” he added. “Amazon Bedrock is one of the fastest growing AWS services in the last decade.”

Internal Amazon data obtained by B-17 show Bedrock is reaching company expectations.

The global sales team at AWS, which has thousands of employees, hit over 97% of its sales goal as of July for Bedrock, according to the internal data.

In two regions — Europe, Middle East, and Africa (EMEA) and Asia Pacific and Japan (APJ) — Bedrock exceeded sales targets. In North America, Bedrock met 83.5% of its sales goal.

It’s unclear how this translates to dollar figures, but the goals are closely monitored by the S-team, Amazon’s most senior group of decision-makers, including CEO Andy Jassy.

For Amazon, Bedrock’s success could help mitigate views of the company falling behind in AI. Its in-house AI models, such as Titan, are less powerful than those of its rivals, while other initiatives, like its new AI chips and AI chatbot, are facing early challenges. Amazon previously said the company is on pace to generate multibillion-dollar revenues this year from AI products.

‘Important but not critical’

AWS CEO Matt Garman appears to be fine not having a fancy, home-grown AI model. In a recent No Priors podcast interview, Garman said Amazon believes there will be “a lot of models for a lot of different purposes” and developers will want a “diversity of option,” ranging from big and small ones to open-weight models.

“It’ll be important but not critical,” Garman said when asked about the importance of offering a top-performing first-party AI model. “We want there to be the best set of options and we want them all to run at AWS.”

Matt McIlwain, managing director at Madrona Venture Group, told B-17 that most, if not all services at the “intelligent/agentic application level” will use multiple AI models. It’s why AWS rivals, like Microsoft, have launched similar offerings to Bedrock, he added. Google also has a competing service called Vertex AI.

AGI team

This doesn’t mean Amazon has given up on building its own powerful AI model. Last year, Amazon created a new AGI team to work on the company’s “most ambitious” large language models. The company also increased its spending on high-profile AI startups over the past year, investing $4 billion in Claude-maker Anthropic, while backing Hugging Face, Scale AI, and Skild AI.

Brent Thill, an analyst at investment firm Jefferies, wrote in a recent note that Amazon stands to benefit from a market that is increasingly demanding multiple models, given AWS’s cloud market share and “model-agnostic approach.”

He cited data from Amazon showing that only 3% of enterprises use just one language model provider, while 34% are using two, 41% have three, and 22% are leveraging four.

“Despite investors viewing AWS as being behind on AI innovation due to not having its own frontier model, AWS has the broadest set of AI capabilities among cloud providers,” Thill wrote.