Bedrock bets on a market where multiple models coexist
Andy Jassy anticipates 6 to 8 AI models “good enough” for most uses within two to three years. Amazon pushes Bedrock as a single access layer, at a time when companies want to compare, switch, and keep control over their costs.
Amazon expects to see 6 to 8 AI models emerge within two to three years that are “about as good as each other, or at least good enough for 95%+ of the workloads,” said Andy Jassy during the group’s quarterly earnings presentation on October 31, 2026. The Amazon CEO sees less a race for a single winner than a market of choice, where companies will use multiple models depending on the task, cost, or speed. This is precisely the position Amazon wants to occupy with Bedrock, its unified access layer to models from multiple providers. And the message is clear: in AI, monogamous loyalty is not really in vogue.
Le pari d’Amazon : moins un champion qu’un catalogue
Jassy presented the market as a succession of specialists rather than a lasting domination by a single player. “there is not one tool to rule the world,” he said, citing different use cases for reasoning, code, visuals, speed, or cost per token. Amazon sees a direct business opportunity: if customers switch from one model to another, they need a neutral layer to do so without rewiring everything each time they change providers.
Bedrock fulfills this role with a common API, the Converse API, which allows calling models from Anthropic, OpenAI, Meta, Mistral, DeepSeek, Moonshot AI, or xAI from a single interface. The service now hosts more than 100 foundation models, according to Amazon, and also includes managed versions of open models. In this context, Amazon’s interest is not only technical: the platform captures orchestration demand, even if the chosen model changes tomorrow.
Changer de modèle reste plus compliqué qu’un simple clic
The promise of a single layer does not erase differences between models. Context length, tool capabilities, supported functions: all still vary from one provider to another. A team that has built its workflow around a given agent or model does not switch without testing. Jassy acknowledged this when talking about customers who want to “swap them back and forth as they leapfrog each other, as different constraints of their applications require them.”
Security also enters the equation. Amazon states that Bedrock does not give model providers access to customers’ prompts or responses, and that this data is not used to train the underlying models. For companies, this separation matters as much as raw performance. For cloud providers, it also serves as a very serious commercial argument, one that prevents data from taking a ride elsewhere.
Une bataille de plateformes plus que de modèles
Amazon is not playing this card alone. Google with Vertex AI and Microsoft with Azure AI Foundry also advocate unified access to multiple providers, in a marketplace logic rather than a single model. The real issue then becomes where one chooses, measures, and orchestrates models, not just who trains them.
Jassy also emphasized the potential market size. He considers the opportunity “gigantic” and thinks it could become “as large as EC2.” Amazon already claims signs of traction: Fortune 500 companies spend, according to the group, nearly seven times more on Bedrock year over year, token volume has increased nearly 17-fold, and the service would be the fastest-growing multibillion-dollar product in AWS history. On the AWS side, the segment grew 39.3% quarterly to $46 billion, with an annualized pace of $184 billion. Bedrock already mostly runs on Amazon’s Trainium chips, which helps the group keep prices tight while protecting its margins.
Key points
- 6 to 8 “good enough” models targeted within two to three years.
- Bedrock aggregates more than 100 foundation models.
- On October 31, 2026, Andy Jassy described a multi-model market.
- Fortune 500 spends nearly 7 times more on Bedrock year over year.
- Token volume increased nearly 17-fold.
By the numbers
- 6 to 8 models — Andy Jassy’s estimate for around 2028, during Amazon’s earnings conference.
- More than 100 models — catalog accessible via Bedrock, according to Amazon.
- 39.3% — AWS quarterly growth to $46 billion.
- $184 billion — AWS annualized pace, announced by Amazon.
- Nearly 17x — increase in token volume on Bedrock, according to the group.