Amazon Web Services said enterprises and developers can take DeepSeek's R1 model for a spin on Amazon Bedrock via its Custom Model Import feature. IBM also said it will add DeepSeek R1 models to watsonx.ai via its Custom Foundations Model feature and Microsoft Azure made a similar move.
DeepSeek, a Chinese AI startup that has torched US AI stock valuations such as Nvidia, has released models that can perform as well as pricier foundation models for a fraction of the cost.
That price compression has spurred a flurry of opinions about how DeepSeek may affect the broader market. Price compression is highly likely.
DeepSeek: What CxOs and enterprises need to know
For AWS, which started with a LLM agnostic strategy, adding something like DeepSeek to Amazon Bedrock isn't a concern. In a community article, AWS said Bedrock's custom import feature can be used to leverage DeepSeek. AWS is also holding a webinar on deploying DeepSeek models on Bedrock.
Key items in the walkthrough include:
- The Custom Model Import feature allows you to use externally fine-tuned models on Bedrock's infrastructure.
- Your DeepSeek R1 model should be based on supported architecture, such as Llama 2, Llama 3, Llama 3.1, Llama 3.2, or Llama 3.3.
- Prepare your model files in the Hugging Face format and store them in Amazon S3.
Holger Mueller, Constellation Research analyst, said:
"AWS wastes no time to keep its 'Switzerland' status when it comes to being home for all LLMs - large and small - as it supports DeepSeek in AWS Cloud. With CISOs probably concerned about any enterprise access - there is likely interest in the AI / Data Science community."
More:
- GenAI prices to tank: Here’s why
- Chinese startup DeepSeek drops a bombshell on the AI world
- Ray Wang on the AI war, the freakout, and big tech valuations and disruption.
IBM followed the AWS with a similar custom import approach. IBM said it will add DeepSeek R1 models to watsonx.ai via its Custom Foundations Model feature. The feature is similar to what AWS has in Bedrock, but IBM said DeepSeek R1 models can be based on Llama or Qwen architectures. Qwen models are created by Alibaba.
The watsonx.ai workflow is similar to the custom import on Bedrock.
Developers need to prepare DeepSeek files and bring the model into IBM Cloud Object Storage. From there, the model needs a config.json file as well as be in a safetensors format before deployment.
Microsoft said DeepSeek R1 is available in the Azure AI Foundry catalog and GitHub. In a blog post, Microsoft emphasized that DeepSeek models were put through its paces.
"DeepSeek R1 has undergone rigorous red teaming and safety evaluations, including automated assessments of model behavior and extensive security reviews to mitigate potential risks. With Azure AI Content Safety, built-in content filtering is available by default, with opt-out options for flexibility. Additionally, the Safety Evaluation System allows customers to efficiently test their applications before deployment. These safeguards help Azure AI Foundry provide a secure, compliant, and responsible environment for enterprises to confidently deploy AI solutions."
