Amazon has just officially entered the new-generation AI infrastructure race by introducing the product "AI Factories", a model of implementing AI right in the data center of a business or government agency.
This is considered a strategic step to meet the growing demand for data sovereignty, a factor that many large corporations and governments prioritize when applying artificial intelligence.
According to Amazon's description, the operating model is quite simple when customers provide energy and data centers, while Amazon is responsible for integrating AI systems, deploying software, hardware, administration and connecting with other cloud services if customers need it.
Thanks to that, all sensitive data is retained at the business's facility, minimizing the risk of leakage or dependence on third-party infrastructure.
Notably, the name AI Factory is not strange. This is the concept Nvidia uses for its comprehensive AI hardware systems, from Blackwell GPUs to high-speed networking technology.
The AWS AI Factory system is actually a collaboration between Amazon and Nvidia, combining the power of specialized GPUs with the AWS's infrastructure capabilities.
Customers deploying this model can choose between Nvidia's latest Blackwell GPU or Trainium3, a new generation AI chip developed by Amazon.
Along with that is an internal cloud storage, network, security, and database system, while fully supporting platforms such as Amazon Bedrock (opting for AI models) and cloud SageMaker (building - training models).
Not only Amazon, Microsoft is also installing Nvidia's "AI factories" in the global data center system.
In October, the company introduced the AI Factory model to serve OpenAI's workloads. However, unlike AWS, Microsoft did not mention the ability to locate these AI super machines in its own data center.
They mainly emphasized the strategy of building AI Super Factories in Wisconsin and Georgia.
Along with that, Microsoft also develops infrastructure services for data sovereignty, including the Azure Local model, which is hardware managed by Microsoft but installed directly at customer locations. This shows that the company is taking a step back to approach the strict requirements for data in each country.
An interesting paradox is that AI, a technology that is powered by the cloud, is dragging technology giants back to on-site deployment models, similar to around 2009 when businesses built their own data centers.
However, the big difference lies in the fact that AI factories today are not ordinary machines, but expensive super computer clusters, requiring a combination of leading hardware and specialized management services.
Amazon considers this an important piece to strengthen its position in the global AI infrastructure race, especially as customers are increasingly cautious with their data. The competition between AWS, Nvidia and Microsoft is therefore getting hotter than ever.
