Jassy: AI Market Will Settle on 6 to 8 Competitive Models, Not One Winner

Amazon CEO Andy Jassy outlined his vision for the artificial intelligence landscape during the company's third-quarter 2026 earnings call, telling investors that the competition among frontier models will yield several strong contenders rather than consolidate around a single leader. The key challenge for enterprises, Jassy suggested, will shift from picking a winner to selecting among winners.

Responding to a question from Goldman Sachs analyst Eric Sheridan about market opportunities for Amazon Bedrock, Jassy projected that within two to three years, "6- 8 models that are about as good as each other, or at least good enough for 95%+ of the workloads" will dominate the market.

No single AI winner

Jassy drew on two decades of technology industry history to support his thesis. He cited databases and analytics platforms as examples of markets where early leaders failed to maintain dominance, and pointed to machine learning frameworks—specifically how TensorFlow once appeared destined to become the standard before PyTorch emerged as the preferred choice.

The same pattern will likely repeat with generative AI models, according to Jassy. Different models will excel in different domains: some will demonstrate superior reasoning or coding capabilities, others will outperform on visual tasks, latency, or cost efficiency. This specialization creates strong incentives for organizations to deploy multiple models rather than standardize on a single platform. As Jassy stated, "there is not one tool to rule the world."

Jassy predicted that "most customers are going to use multiple models, and many will use multiple models in the same generative AI application." For development teams, this means the ability to select tools based on specific requirements rather than locking into a single vendor's ecosystem.

One API, many models

This vision of model diversity describes Bedrock's fundamental architecture. The service provides access to more than 100 foundation models from vendors including Anthropic, OpenAI, Meta, Mistral, DeepSeek, Moonshot AI, and xAI through a single API interface. Teams can invoke Claude, GPT, Llama, or specialized smaller models without building separate integration pipelines for each.

Amazon recently integrated OpenAI's models into Bedrock alongside proprietary models already available on the platform. The company has also added managed versions of open-source models such as Moonshot's Kimi and Zhipu's GLM, reflecting what Jassy characterized as sustained demand for Western open models that Bedrock intends to continue satisfying.

Switching models isn't free

Bedrock's Converse API provides a standardized interface for working with models across different providers, reducing the need to rewrite integrations when switching between them. However, the process remains constrained by practical limitations. Models differ in their context windows, available tools, and supported features, requiring developers to validate how applications perform with each option.

Complexity increases when teams have constructed workflows around a particular vendor's agent framework. Jassy acknowledged that customers want the ability to "swap them back and forth as they leapfrog each other, as different constraints of their applications require them."

Data privacy considerations also influence these decisions. AWS maintains that Bedrock prevents model providers from accessing customer prompts or responses, and that this information does not feed into model training. Jassy highlighted these protections as a factor driving enterprise adoption of the service.

Why Amazon wins anyway

The underlying strategy assumes Amazon need not build the best model to capture value, provided it controls the marketplace where developers choose among them. Amazon is not alone in pursuing this position; Google's Vertex AI and Microsoft's Azure AI Foundry offer comparable multi-vendor catalogs with unified access layers.

A market where six to eight models continuously trade leadership and teams regularly rotate between them creates ideal conditions for a neutral access layer to monetize switching. Jassy made little attempt to obscure Bedrock as the intended beneficiary, describing the opportunity as "gigantic" and suggesting the service could eventually rival EC2, which remains one of AWS's largest revenue generators.

Early performance metrics support that ambition. Fortune 500 companies are now spending nearly seven times more on Bedrock compared to the prior year, token volume has grown almost 17 times, and Jassy characterized Bedrock as the fastest-growing multibillion-dollar product in AWS history.

This momentum arrives as AWS itself accelerates, with the segment expanding 39.3% in the quarter to $46 billion and reaching a $184 billion annualized run rate. Jassy noted that the majority of Bedrock traffic already runs on Amazon's proprietary Trainium chips, allowing the company to maintain competitive pricing while preserving profit margins.