The combination of MK1’s software and AMD’s Instinct GPUs will enable the delivery of ‘accurate, cost-effective and fully traceable reasoning at scale,’ says an AMD executive, referring to the process at the heart of agentic AI workloads.
AMD said Monday it has bought an AI software startup founded by two veterans of Elon Musk’s Neuralink firm as part of an ongoing acquisition spree to build out its AI capabilities against Nvidia.
Anush Elangovan, AMD’s corporate vice president of software development, wrote in a blog post that the chip designer completed its acquisition of MK1, a Mountain, View, Calif.-based provider of inference and enterprise AI software, to “advance AI performance and efficiency across the stack.”
[Related: Exclusive: Intel Is Losing A Data Center AI Executive To AMD]
Elangovan, who joined AMD in 2023 through a previous AI acquisition, said the MK1 team will join AMD’s Artificial Intelligence Group, “where their technology and expertise will play a key role in advancing our high-speed inference and enterprise AI stack.”
MK1 was started by Neuralink co-founder Paul Merolla, who co-led chip design and the development of algorithms that decode brain activity, along with Thong Wei Koh, a former team lead at the Elon Musk-owned firm who focused on brain neural signal processing.
Other team members include former engineers at Neuralink, Meta, Tesla and Apple.
According to Elangovan, MK1 has been focused on “high-speed inference and reasoning-based AI technologies optimized for large-scale deployments” running on AMD hardware, currently serving more than 1 trillion tokens a day.
Elangovan said MK1’s Flywheel and comprehension engines are designed to take advantage of the memory architecture of AMD’s Instinct GPUs—which have gained momentum in the AI infrastructure market with the company sealing a deal with OpenAI to deploy 6 gigawatts of Instinct-based infrastructure as part of a strategic partnership.
The combination of MK1’s software and AMD’s Instinct GPUs will enable the delivery of “accurate, cost-effective and fully traceable reasoning at scale,” added Elangovan, referring to the process at the heart of agentic AI workloads.
“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,” he wrote.
The Santa Clara, Calif.-based company announced the MK1 acquisition after disclosing last week that it paid $36 million on acquisitions outside its blockbuster $4.9 billion ZT Systems deal earlier this year to boost its AI and data center capabilities.
Those earlier acquisitions consisted of silicon photonics startup Enosemi, which AMD scooped up to “support and develop a variety of photonics and co-packaged optics solutions across next-gen AI systems;” compiler startup Brium, which the company is using to provide “highly optimized AI solutions;” and the technical employees of AI chip startup Untether AI.
The disclosure was made as AMD reported a “sharp” jump in sales for its CPUs across the PC and server segments as well as its Instinct data center GPUs, which allowed the company to deliver record revenue of $9.2 billion for the third quarter.
As CRN previously reported, AMD has leaned heavily into acquisitions over the past few years to flesh out its GPU, system and software capabilities in its growing rivalry with Nvidia in the AI infrastructure market.
The company has used its largest acquisition this year, ZT Systems, to develop rack-scale AI solutions based on its Instinct GPUs, and the move has played a key role in helping AMD win over big customers, including OpenAI.
AMD sold the manufacturing unit of ZT Systems to U.S. electronics services giant Sanmina for $3 billion in October, keeping the business’ design and customer enablement teams.