Microsoft hosts controversial DeepSeek R1 AI model on Azure

AWS has also begun hosting the Chinese model

Microsoft has announced it will now host DeepSeek R1, the AI model at the centre of a recent controversy involving OpenAI, on its Azure cloud platform.

The move, revealed on Wednesday, follows OpenAI’s accusations that the Chinese company behind DeepSeek R1 violated its terms of service by allegedly using OpenAI’s model outputs to train its own system. Despite this, Microsoft, OpenAI’s largest investor, has chosen to add R1 to its Azure AI Foundry, where it offers access to over 1,800 AI models.

DeepSeek R1 has generated significant discussion in the AI sector due to its reportedly high-level reasoning capabilities, which are said to rival OpenAI’s o1 model, while being trained at a fraction of the cost.

Microsoft corporate vice president Asha Sharma said in a press release: “R1 offers a powerful, cost-efficient model that allows more users to harness state-of-the-art AI capabilities with minimal infrastructure investment.”

Azure provides developers with cloud-based computing power and access to AI models. While pricing comparisons for R1 and o1 on Azure have not been disclosed, DeepSeek lists R1’s API usage at $2.19 per million output tokens, in contrast to OpenAI’s o1, which costs $60 per million tokens, a significant price advantage.

A divisive AI partnership

While Microsoft’s decision aligns with its broader strategy of expanding its AI offerings, it also lends credibility to an AI model that has been at the centre of controversy. OpenAI has accused DeepSeek of employing a method known as “distillation”, whereby AI-generated outputs are used to train competing models, an approach that violates OpenAI’s terms of service.

Concerns over DeepSeek’s data sources surfaced after its predecessor, DeepSeek V3, was reported to occasionally refer to itself as ChatGPT, raising suspicions that OpenAI-generated content was used in its fine-tuning process. This follows similar accusations against Elon Musk’s xAI, which was alleged to have used OpenAI’s outputs to train its Grok AI model in late 2023.

Microsoft has also investigated DeepSeek’s data practices. According to Bloomberg, its security researchers discovered that DeepSeek may have extracted substantial training data via OpenAI’s API in late 2024.

Despite these disputes, OpenAI CEO Sam Altman struck a conciliatory tone earlier this week, acknowledging DeepSeek as a formidable competitor.

AWS also hosting DeepSeek

AWS has also enabled the DeepSeek R1 foundation model to be used within its flagship AI platform Amazon Bedrock and Amazon SageMaker.

“We’ve always been focused on making it easy to get started with emerging and popular models right away, and we’re giving customers a lot of ways to test out DeepSeek AI,” said AWS CEO Matt Garman in a LinkedIn post.