Abstract Bullets:
- DeepSeek-R1 Distill Llama and Qwen fashions grew to become out there in Amazon Bedrock Market and Amazon SageMaker JumpStart in February 2025.
- DeepSeek-R1 grew to become out there as a totally managed, usually out there, serverless mannequin in Amazon Bedrock on March 10, 2025.
Amazon Internet Companies (AWS) simply grew to become the primary cloud computing supplier to supply DeepSeek-R1 as a totally managed, serverless massive language mannequin (LLM) basically availability on Amazon Bedrock, a service that gives entry to third-party basis fashions. The publicly out there DeepSeek-R1 sequence has taken the generative AI (GenAI) market by storm, with its superior reasoning capabilities and diminished computing prices, made potential because of an progressive structure based mostly on combination of specialists applied sciences.
This can be a logical transfer by AWS, which has been providing entry to basis fashions by different suppliers since early 2023, when Amazon Bedrock was launched. Again then, Amazon began out internet hosting fashions by a roster of third-party corporations, together with startups similar to AI21 Labs, Anthropic, and Stability AI, on its cloud platform. It additionally launched its personal proprietary expertise, the Amazon Titan basis fashions, which was additionally made out there via Amazon Bedrock. The Amazon Titan household has by no means been a selected energy when in comparison with basis fashions by rivals similar to Google Cloud Gemini. For that reason, Amazon selected to deal with the managed service which, being built-in with the remainder of the AWS cloud platform, had an enormous addressable market. Since then, the corporate has diversified into one other proprietary model, the Amazon Nova household.
Though Amazon clients needn’t fear about technical setup or upkeep as a result of it’s totally managed by AWS, with expertise by DeepSeek, the query of safety and compliance may probably be extra problematic. In keeping with Amazon, customers can relaxation assured that the inbuilt safety features, together with knowledge encryption and strict entry controls, are ample. They embrace instruments to take care of knowledge privateness and regulatory compliance. Regardless, Amazon Bedrock clients are really useful – when utilizing any AI system on the platform – to robotically arrange safeguards similar to Amazon Bedrock Guardrails, which can also stop hallucinations.
Lots of AWS’ opponents made DeepSeek-R1 out there on their platforms weeks in the past; Google Cloud, Nvidia, and Microsoft amongst them. By no means one to overlook a revenue-generating alternative, Microsoft went a step additional by turning into the primary firm to implement a per-token pricing construction for DeepSeek-R1, whereas Google Cloud customers pay just for the computing sources they eat, quite than the quantity of textual content the mannequin generates. The latter is a extra widespread pricing construction in the case of open-source AI. By incorporating DeepSeek’s fashions, Microsoft is following a method to diversify its AI choices and cut back over-reliance on OpenAI’s ChatGPT, specializing in a variety of inner and third-party AI fashions to strengthen the Microsoft Copilot suite. The Microsoft Azure AI Foundry at the moment boasts an array of 1,800 fashions, and DeepSeek-R1 can also be out there on GitHub Fashions free expertise.
Nevertheless, the inclusion of DeepSeek-R1 in Amazon Bedrock as a fully-managed service is critical as a result of it’s providing customers the chance to leverage the expertise with built-in safety, observability, monitoring, and customizable guardrails. Prospects have AWS’ infrastructure and are ready to make use of the mannequin on a serverless pay-per-token foundation, serving to them scale from experimentation to manufacturing. Amazon revealed that 1000’s of shoppers have deployed the DeepSeek-R1 mannequin utilizing Amazon Bedrock’s customized mannequin import characteristic for the reason that system was first hosted on AWS in late January 2025.
