AMD is set to acquire MK1, a software startup focused on high-speed AI inference and reasoning, in a bid by the chip designer to expand its enterprise AI strategy.

The acquisition, detailed in a blog post by Anush Elangovan, AMD’s corporate VP of software development. will see AMD gain access to MK1’s Flywheel engine. Currently serving more than one trillion tokens per day, MK1’s offering is directly tailored to the memory architecture of AMD’s Instinct GPUs.

This, and MK1’s comprehension engine, deliver what Elangovan described as “accurate, cost-effective and fully traceable reasoning at scale.”

“Together, we’ll accelerate the next generation of enterprise AI, enabling customers to automate complex business processes and unlock new opportunities in high-value applications,” Elangovan said.

MK1’s team will also join AMD's Artificial Intelligence Group as part of the agreement.

With MK1’s team on board, AMD said it is set to deliver enhanced capabilities in enterprise AI workflows, enabling automation of complex business processes and opening new opportunities in high-value applications.

The move is also hoped to improve the accessibility of compute tools, with Elangovan saying the deal will help to unlock “AI for everyone.”

Broader strategic context: AMD gets aggressive

AMD’s move comes amidst a growing expansionist strategy aimed at strengthening its AI software and system-integration capabilities.

Earlier this year, the company announced a multi-gigawatt compute deal with OpenAI to deploy its Instinct GPUs to power its infrastructure stack.

In its largest deal this year, AMD acquired ZT Systems for $4.9 billion, a move that accelerated development of rack-scale systems for cloud and enterprise AI deployments.

AMD later sold ZT Systems’ manufacturing arm while retaining design and integration capabilities.

The MK1 acquisition is positioned to further strengthen AMD’s enterprise inference-AI offering by fusing its compute hardware leadership with differentiated reasoning-software capability.

While the acquisition is relatively small compared to AMD’s other deals, the software piece is strategically important.

Inference workloads represent one of the fastest-growing segments in AI infrastructure, with findings from S&P showing 30% of organizations are already using inference and operational workloads, with this share expected to overtake training and fine-tuning in the coming years.

Inference and reasoning make up the decision-making core of AI systems. Once a model has been trained, the inference engine puts that knowledge to work, applying logic and reasoning methods to interpret data and form autonomous conclusions.

As enterprises scale AI deployments, combining hardware, software, and integration may increasingly determine who wins in the enterprise AI arms race.