AMD’s MK1 Buy Signals a Push Into AI Reasoning

AMD has acquired MK1, an AI software startup built by former Neuralink engineers. The acquisition is designed to add more depth to its growing in-house AI software bench. The MK1 team will join AMD’s Artificial Intelligence Group, where their work will help speed up and streamline inference and reasoning workloads running on AMD’s Instinct GPUs.

Anush Elangovan, AMD’s corporate vice president of software development, says the deal is a step toward improving the company’s end-to-end AI capabilities. In AMD’s announcement, he basically said that MK1’s software is built to play especially well with Instinct’s memory setup, which is how it’s able to deliver “accurate, cost-effective, and fully traceable reasoning at scale.”

He added, “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.”

Where MK1 fits in: GPU speed at scale

MK1 was founded by Neuralink co-founder Paul Merolla and former team lead Thong Wei Koh, both of whom worked on decoding neural signals. In short, this is a team used to pushing the limits of high-speed computation.

Instead of focusing on model training, MK1 is built for what comes after: the part where AI systems interpret data and make decisions in real time. The company says its software already processes more than a trillion tokens a day across large-scale deployments.

This inference-and-reasoning layer is where things are getting competitive now, especially as companies move toward more automated, agent-style workflows. Yes, GPU muscle matters, but running that reasoning fast, consistently, and at scale is just as important. That’s the space MK1 has been focused on.

Acquisition part of AMD’s larger strategy 

This deal is part of AMD’s broader push to expand its AI software stack and compete more directly with Nvidia. Over the past year, AMD has swept up startups in inference optimization, GPU-agnostic software, and networking, and it recently signed a multiyear compute partnership with OpenAI. The pattern is pretty darn clear… build the hardware, and make sure the software is there to take full advantage of it.

AMD CEO Lisa Su has been vocal about where this is all headed. As reported by Reuters, she told analysts, “Data center is the largest growth opportunity out there, and one that AMD is very, very well positioned for.” AMD expects the data center chip market to grow to $1 trillion by 2030, with AI at the helm. 

Why this matters for MSPs and service providers

Inference and reasoning are parts of AI that appear in everyday business use, not just in research labs. If AMD continues to build out its software stack, MSPs and solution providers may finally have real alternatives to the current Nvidia-centric setups.

In practice, that means more room to choose what fits. Better pricing leverage, fewer lock-in headaches, and the ability to design AI deployments around what clients actually need (not just what one vendor’s ecosystem dictates).

AMD’s latest deal with MK1 builds on the same momentum as its $4.9 billion acquisition of ZT Systems earlier this year. While ZT strengthens AMD’s hardware and data center backbone, MK1 adds the reasoning and inference layer on top, together signaling a more complete, end-to-end AI stack that brings AMD closer to Nvidia’s territory.