Amazon: Winning the AI Race Without the Best Model
N.R. Finch
CEO Andy Jassy told investors AWS can win in AI without owning the top model — backed by AWS revenue up 37% year-on-year. His bet: don't build the smartest model; make every model run on your platform.
Why does Amazon think it can win without the strongest model?
Jassy's core thesis: the AI industry will not converge on a single dominant model. Different models will lead on different tasks at different times.
This means → enterprise customers will need access to multiple models at once. Whoever provides that access layer controls the gateway.
In plain terms = Amazon is not competing on "whose model is smartest" — it is competing on "whose shelf holds the most models."
What business is Bedrock actually building?
Bedrock — Amazon's foundation-model access platform — already hosts models from Anthropic, OpenAI, and other major AI vendors, positioning itself as the industry's largest inference engine.
Jassy said last year that Bedrock could eventually rival AWS's flagship compute service EC2. This means → internally, Amazon treats "selling model calls" as a business on par with "selling raw compute."
In plain terms = AWS used to make money renting out servers. Now it wants a second revenue line — renting out access to AI models.
Its own models are struggling — what does Amazon say?
According to *Business Insider*, Amazon is scaling back its existing Nova model line and redirecting resources toward a new frontier-model effort.
Jassy did not sidestep this. He argued that owning a competitive in-house model delivers stronger cost control and faster customer response — no need to depend on outside vendors.
A live example: voice assistant Alexa is already shifting away from Anthropic models toward Amazon's own AI to cut costs.
What is Amazon's endgame?
Jassy predicted "at least six comparably capable top-tier models" will emerge within the next few years. "They'll all be on Bedrock, and one of them will be ours," he said.
This means → Amazon's ultimate goal is platform first, in-house model as one option on the shelf — not the sole bet.
This reflects a fundamentally different playbook from Google and OpenAI: Amazon is not wagering on who builds the best model — it is wagering on who builds the most indispensable platform.
Content is for reference only, not financial advice.