AMD sees no slowdown in AI demand as businesses look for bigger productivity gains

AMD expects demand for AI computing to remain strong as businesses continue to adopt AI for productivity gains, while the company expands its presence across chips, software, networking and complete computing solutions.

Mark Papermaster, CTO & EVP - Tech & Engineering at AMD, said the company is still seeing strong demand for AI infrastructure, with businesses at an early stage of adopting the technology. He said companies may become more selective about how they use AI to manage costs, but this does not point to a slowdown in overall AI adoption.

He added, “AI is speeding up everything in the industry. We are actually having to use AI to accelerate our product development. We will continue to grow, and we'll continue to adopt AI at a very fast pace.”

Agentic AI could change the computing mix

Papermaster sees agentic AI as an important shift in the technology cycle. Unlike applications that use AI for individual tasks, agentic AI can link multiple tasks and apply reasoning across an entire workflow.

“So, you are taking workflows, not point tasks, but entire workflows, and you are bringing a smart and an efficient productivity to it with agentic workflows. It's absolutely game-changing. So that's where people are getting the big productivity gains. So it is an inflection point - the adoption of AI and honestly, for us, it's also an inflection point in the technology needed.”

This is also changing the balance between CPUs and GPUs. As AI moves from model training towards inference and real-world applications, businesses need computing across CPUs, GPUs and networking. Papermaster said the CPU-to-GPU ratio in some AI workloads could move closer to one-to-one from the earlier model of one CPU for every eight GPUs.

The company also expects AI workloads to become more distributed. More demanding applications could continue to run on cloud-based frontier models, while established workloads may shift to on-premise data centers or AI-enabled PCs to control costs.

Efficiency remains a key technology focus

AMD expects improvements in computing efficiency to remain a major driver of innovation. Papermaster said silicon-level efficiency could improve by around 30% per generation, while broader optimisation across software and hardware could deliver much larger gains.

He highlighted three technology trends for the coming years: greater computing density through chiplet technology, full-stack optimisation using AI, and more specialised computing systems designed for specific workloads.

India investment ahead of schedule

India remains an important part of AMD’s global innovation and talent ecosystem. The company had committed $400 million to expand its India operations over five years, but Papermaster said the investment is already ahead of schedule.

AMD has invested $245 million in the first two years and added nearly 4,000 roles, exceeding its original commitment of 3,000 hires.

The company is also expanding its data center footprint in India. AMD is working with NASA on a data center and has an agreement with TCS HyperVault, with the latter expected to start ramping up in 2027.

Papermaster also expects AMD to deepen partnerships in India as AI adoption grows.

From chipmaker to solution provider

AMD's strategy is increasingly moving beyond individual processors and accelerators towards complete computing solutions.

The company plans to roll out AI rack-level integration towards the end of this year, alongside the software and other capabilities needed to support these systems.

“We have transformed from being a computer chip provider to a solution provider,” Papermaster said.

For AMD, the focus ahead is therefore not just on supplying more computing power, but on improving efficiency, supporting inference workloads and integrating the different components needed to run AI at scale.