AI will impact every enterprise, whether large or small. Agentic AI will reshape enterprise operations by both improving the efficiency of existing processes and enabling breakthrough innovation in core business areas.
At the same time, AI will place unprecedented demands on enterprise infrastructure whether on premises or in the public cloud.
CIOs must thoroughly understand their options for refactoring existing servers and preparing both on-premises data centres and cloud environments for the impending AI revolution.
The impact of AI on data centre infrastructure
Agentic AI will have a profound impact on existing on-premises data centres. First, new power-hungry infrastructure will need to be deployed and secondly there will be additional strain on already deployed applications as agents multiply the amount compute and storage.
Kumaran Siva, Corporate Vice President, Strategic Business Development at AMD, explains: “Deploying AI at scale on premises requires reworking how the data centre is architected. For example, power is the essential commodity; new GPU servers can be 6 kilowatts per server or more – at a rack level this becomes a significant challenge. But that is not all, regular servers, networking, security and storage will all need to evolve”.
The data centre software platform will also need to evolve to support AI infrastructure. The management infrastructure needs to evolve to comprehend the security and scale that comes with agentic AI.
Siva explains: “Between increasing IT costs and concerns around data control, where and how AI operates will be one of the defining decisions of the next few years.”
Industry innovations offer a way forward
As enterprises’ digital requirements evolve, so too do the capabilities of the processors that support them.
5th Gen AMD EPYC™ processors, for instance, have been estimated to enable up to 7 to 1 server consolidation in data centres on certain workloads compared to legacy servers, allowing organisations to replace older, less efficient servers with fewer, more powerful ones, helping to reduce rack space, power consumption, and associated costs.1
These new EPYC CPU powered servers deliver the power needed for the accelerated GPU servers required by agentic AI. These same EPYC powered server types can also be deployed to efficiently address greater power needs to process the additional load from AI agents.
On the AI infrastructure side, open-source platforms such as Kubernetes and open models such as LLAMA are becoming popular in the enterprise.
Innovation is also coming from more unexpected places. In response to limitations on GPU availability in China, for instance, DeepSeek has shown that software optimisation can enable inference on state-of-the-art AI models to be run with less compute than was once thought.
According to Siva, this type of open model innovation creates opportunity for enterprises to reduce power consumption and lower cost by consuming on-premise GPU deployments.
How AMD supports infrastructure modernisation
In this environment, enterprises need to look for manufacturers that are squarely focused on innovation, and how best they can support the requirements of enterprises.
“Enterprises need a mix of innovation and ability to execute,” Siva explains. “It’s also important that they maintain flexibility in their supply chain to take advantage of the best available technology to solve their business problems regardless of the source.”
For IT leaders seeking AI gains, technology that yields faster inference is a key benefit.
“Take quantisation, as an example,” says Siva. “We’ve moved to lower-precision formats that dramatically improve efficiency without compromising performance. Memory is also key. With our on-package, high-bandwidth memory, AMD GPUs offer leading-edge memory capacity per device. That means more of the model can be kept in memory—enabling faster inference and more efficient training.”
Yet not all businesses have perhaps understood the importance of having the power to drive innovation. Only a quarter are (27%) investing in high performance compute and cloud infrastructure to drive AI according to recent research.2
For CIOs currently considering how best to modernise their infrastructure to prepare for AI, Siva recommends first looking at consolidation, and then ensuring that they embrace an open ecosystem for their AI infrastructure.
“The smartest first step is consolidating your existing server fleet—cutting power, modernising infrastructure, and creating headroom for what’s next. AMD lets you do that. And unlike closed ecosystems, we believe in flexibility, offering leadership silicon, open platforms, and freedom of vendor choice. We carry this philosophy into our AMD Instinct™ GPU offering, where we leverage open-source technologies to allow the latest open models like Deepseek V3 and LLAMA 4 to run efficiently on open-source inference engines like vLLM.”
1 For additional details, see https://www.amd.com/en/legal/claims/epyc.html#q=SP9xx5TCO-002A.