Research Finder
Find by Keyword
AWS hosts DeepSeek R1 - What does this mean for AI Adoption?
DeepSeek R1 models now available on AWS Bedrock and SageMaker. Explore cost-effective, customizable AI solutions for generative AI development.
Key Highlights:
- DeepSeek R1 and distilled models now available on AWS - in certain regions
- Cost effective generative AI development.
- Flexible deployment options via Bedrock and SageMaker.
- Deepseek enhanced reasoning capabilities via unique training.
- Secure and scalable generative AI solutions.
The News:
DeepSeek R1 and distilled models are now accessible on Amazon Bedrock and Amazon SageMaker. This allows developers to leverage DeepSeek's advanced AI capabilities for building and scaling generative AI applications. The models aim to deliver improved reasoning and cost efficiency. Find out more by checking out the article.
Analyst Take:
AWS ran an analyst only DeepSeek demonstration of the R1 and distilled models running on AWS infrastructure and the announcements that DeepSeek’s model is part of the AWS ecosystem is on message for the Hyperscaler.
The tenor of the analyst only briefing aligned with the messaging from AWS re:Invent last year, where Amazon CEO Andy Jassy discussed insights from deploying almost 1,000 generative AI applications within Amazon, highlighting three significant observations. He noted that at scale, the cost of computing becomes critical as people seek better price performance, and that building a high-quality generative AI application is challenging. Additionally, he observed that giving builders the freedom to choose led to a diversity of models, reinforcing the lesson that no single tool will dominate. Jassy emphasized that Amazon's wide range of models allows customers to select those that best fit their specific needs, with AWS continuously updating its model offerings to keep pace with technological advancements and customer demands.
The key takeaway from Jassy’s keynote and the analyst only briefing was choice matters and AWS is embracing giving its customers choice when it comes to models.
The availability of DeepSeek R1 models on AWS represents another development in the generative AI landscape and is a validation of AWS’ strategy for choice. AWS's commitment to providing a diverse range of models caters to the evolving needs of its customers. This move reinforces the idea that a single, universal AI model is unlikely, despite what OpenAI will tell you. Instead, a spectrum of specialized models will likely dominate the future of AI for specific use cases. The focus on cost performance is a key factor. As generative AI applications scale, the cost of compute becomes a significant consideration. DeepSeek's focus on cost effectiveness is a smart strategy. It aims to democratize access to advanced AI capabilities.
What was Announced:
DeepSeek R1 and its distilled variants are now available on AWS in certain regions. DeepSeek R1 models are designed to offer enhanced reasoning capabilities. This is achieved through innovative training techniques such as reinforcement learning. The DeepSeek R1 family includes models with parameter sizes ranging from 1.5 billion to 671 billion.
In the analyst only call AWS confirmed that to run the full 671bn parameter model requires a P5e.48xlarge instance which is an eight H200 GPU system with 192 vCPUS with 2TB of memory. That instance type is running at $90-95 per hour (region dependent). These instances are not available in every region, so please be aware of that.
AWS’ approach allows developers to select the model that best suits their specific needs and aligns best with their budget is going to provide maximum optionality and is on message for AWS. The models are architected to be deployed on both Amazon Bedrock and Amazon SageMaker. This provides flexibility for developers with varying levels of expertise. For those seeking quick integration, Amazon Bedrock offers a streamlined approach through APIs. For organizations requiring advanced customization and control, Amazon SageMaker offers access to the underlying infrastructure.11 DeepSeek R1 can be deployed on AWS Trainium and AWS Inferentia. This aims to optimize cost effectiveness for deploying the distilled models. The models are designed to integrate with Amazon Bedrock Guardrails. This adds a layer of security for generative AI applications.
Looking Ahead
Based on what we are observing, the integration of DeepSeek R1 models into the AWS ecosystem is yet another validation point for AWS' strategy around making advanced AI more accessible and giving customers choice. The key trend that we are going to be looking out for is how developers leverage these models to build innovative applications. Does DeepSeek get traction after the hype dissipates? Too early to say, but a key point to track going forward.
Based on HyperFRAME’s analysis of the market, our perspective is that the focus on cost effectiveness will be a major driver of adoption. Another point to note is that more models equals more inference and that is good for GPU vendors.
Going forward we are going to be closely monitoring how DeepSeek R1 models perform in real world applications. When you look at the market as a whole, the announcement of running DeepSeek on AWS highlights the increasing importance of specialized AI models and optionality. The competitive landscape in generative AI is dynamic, and this move positions AWS with DeepSeek to be on trend, and proves that the company can move fast if a new model comes along in the future, even if mass adoption never happens for DeepSeek.
Steven Dickens | CEO HyperFRAME Research
Regarded as a luminary at the intersection of technology and business transformation, Steven Dickens is the CEO and Principal Analyst at HyperFRAME Research.
Ranked consistently among the Top 10 Analysts by AR Insights and a contributor to Forbes, Steven's expert perspectives are sought after by tier one media outlets such as The Wall Street Journal and CNBC, and he is a regular on TV networks including the Schwab Network and Bloomberg.