Amazon’s Bedrock Bet: Six to Eight AI Models Could Handle Most Workloads

Oct 31, 2026 337 views
Mix of similar doors

The strategic value in generative AI may sit less with whoever builds the top model and more with whoever makes it practical to switch between contenders. Amazon CEO Andy Jassy expects six to eight models to become capable enough for at least 95% of workloads within two to three years, making model selection an ongoing engineering choice.

Jassy laid out that forecast on Amazon’s Q3 2026 earnings call, responding to Goldman Sachs analyst Eric Sheridan’s question about the opportunity for Amazon Bedrock. He described a field of models that are “about as good as each other, or at least good enough for 95%+ of the workloads.”

Model choice will follow the task

Jassy’s expectation draws on a familiar pattern in technology: early leaders don’t always keep their lead. He cited databases and analytics, as well as TensorFlow, which once looked poised to become the standard machine learning framework before PyTorch gained ground.

His broader point is that no single model will excel at everything. One may handle reasoning or code especially well, while another offers stronger visual capabilities, faster responses, or a lower cost per token. Jassy put it plainly: “there is not one tool to rule the world.”

That spread in capabilities gives companies a reason to mix models, sometimes within one application. “Most customers are going to use multiple models, and many will use multiple models in the same generative AI application,” he said. For developers, that points to routing work according to performance, speed, and cost instead of treating one provider as the default for every task.

Bedrock makes access the product

Bedrock is positioned for that kind of model mix. Its single API provides access to more than 100 foundation models from providers including Anthropic, OpenAI, Meta, Mistral, DeepSeek, Moonshot AI, and xAI. Teams can call Claude, GPT, Llama, or a specialized smaller model without building a separate integration for each one.

Amazon recently added OpenAI models to the service, alongside the closed models already available. Jassy also said Bedrock now offers managed versions of open models such as Moonshot’s Kimi and Zhipu’s GLM. He sees demand for Western open models as another area the service can serve.

The business logic is clear: Amazon doesn’t have to produce the leading model if developers rely on Bedrock to reach whichever models suit their needs. Google’s Vertex AI and Microsoft’s Azure AI Foundry are pursuing a similar role, offering multi-provider catalogs and a common access layer.

Portability still takes engineering work

A shared API can reduce integration effort, but changing models still calls for testing. Context limits, tool support, and other features vary by provider, and applications may depend on vendor-specific agent tooling. Bedrock’s Converse API gives developers a common interface, but it can’t erase those differences.

Jassy said customers want to “swap them back and forth as they leapfrog each other, as different constraints of their applications require them.” That flexibility will depend on how well teams design and test their applications for model changes, particularly when a workflow relies on one provider’s agent harness.

Data handling is another factor for enterprise buyers. AWS says Bedrock does not give model providers access to customer prompts or responses, and that customer data isn’t used to train the underlying models. Jassy cited these protections as one reason organizations choose the service.

Usage and infrastructure are growing together

Jassy called Bedrock’s opportunity “gigantic” and said it could become “as large as EC2,” one of AWS’s biggest businesses. He said Fortune 500 companies are spending close to seven times more on Bedrock year over year, while token volume has risen nearly 17 times. By his account, it’s the fastest-growing multibillion-dollar product in AWS history.

Those figures arrive as AWS reported 39.3% quarterly growth, reaching $46 billion in revenue and a $184 billion annualized run rate. Jassy also said most Bedrock traffic runs on Amazon’s Trainium chips, which he says helps the company keep prices down while protecting margins.

If Jassy’s forecast holds, developers will spend more time matching models to jobs and less time asking which single provider will dominate. Bedrock’s test will be whether its common access layer makes those choices manageable as models and application needs keep changing.

Source: Amanda Caswell · thenewstack.io

Comments

Sign in to comment.
No comments yet. Be the first to comment.

Related Articles

Amazon CEO predicts 6 to 8 AI models will be ‘good enough...