What MK1 does

MK1 is a California-based artificial intelligence company that built high-speed inference and reasoning technologies optimized specifically for the memory architecture of AMD Instinct GPUs. The company's product suite consists of two core components: Flywheel, a high-speed inference engine that was processing over 1 trillion tokens per day across large-scale deployments, and Comprehension Engines, which deliver accurate, cost-effective, and fully traceable reasoning at scale. Together these products address enterprise pain points around inference latency, computational cost, and the absence of auditable reasoning in large language model outputs, targeting large enterprises that require high-throughput AI for automated business processes.

MK1 operated with a small team (1-10 employees) and was headquartered in Menlo Park/Mountain View, California. Its go-to-market was narrowly focused on enterprise AI inference and complex business process automation; pricing was not publicly disclosed. Prior to its acquisition, the company had received early-stage capital from Interface Fund (February 2023) and Pacific Alliance Ventures (June 2024). Following AMD's completion of the acquisition in November 2025, MK1 became a wholly-owned subsidiary whose team and technology were absorbed into AMD's Artificial Intelligence Group. The technology is now distributed through AMD's enterprise sales channels and partner ecosystem, positioning it within AMD's broader strategy to compete with NVIDIA in AI infrastructure by offering hardware-software co-optimized inference performance.

Name

MK1

Legal name

MK1

Website

https://mk1.ai

Company type

Private

Founded year

2022

Operating status

Acquired

Headcount range

1–10 employees

Short description

MK1 built high-speed AI inference and reasoning technology optimized for AMD Instinct GPUs, serving over 1 trillion tokens daily. The company was acquired by AMD in November 2025 and integrated into its Artificial Intelligence Group.

Ownership category

akta.pro rank

Product category

AI Inference Software

NAICS

Software Publishers (5132), Custom Computer Programming Services (541511), Computer Systems Design Services (541512)

SIC

Services-Prepackaged Software (7372), Services-Computer Programming, Data Processing, Etc. (7370)

akta.pro primary industry

AI Compiler, Runtime & Kernel Optimization Software (CUDA/ROCm/XLA, graph compilers) (HDAAAAAI)

akta.pro secondary industries

AI Accelerators (GPUs/TPUs/NPUs/ASICs) (HDAAAAAA), GPU-Accelerated & AI Training/Inference Servers (HDACABAG)

Keywords

• AI inference software
• GPU-optimized AI
• AI reasoning engines
• Enterprise AI platform
• High-speed inference

Headquarters

HQ city

Menlo Park

HQ country

United States

HQ region

North America

Markets served

GTM type

B2B

Offering type

Software

Cost components

Technology or R&D, Personnel, Operations

Core offering

MK1 builds high-speed AI inference and reasoning software optimized for AMD Instinct GPU memory architecture. Its platform, centered on Flywheel inference technology and Comprehension Engines, delivers accurate, cost-effective, and fully traceable reasoning at scale, serving over 1 trillion tokens daily in large-scale enterprise deployments.

Product overview

MK1 offers a unified AI inference platform consisting of Flywheel technology and Comprehension Engines. Both products work together as integrated components optimized specifically for AMD Instinct GPU memory architecture, providing scalable reasoning capabilities for large enterprise AI deployments. The Flywheel serves as the primary inference engine processing over 1 trillion tokens daily, while Comprehension Engines enable accurate and traceable reasoning. Following AMD's acquisition, these technologies were integrated into AMD's Artificial Intelligence Group to enhance the company's enterprise AI software stack.

Differentiator

Problem solved

Functional benefit

Products and services

• Flywheel High-speed AI inference technology optimized for AMD Instinct GPUs, designed to process over 1 trillion tokens daily for large-scale enterprise AI deployments requiring fast, cost-effective inference.
• Comprehension Engines Purpose-built AI engines that leverage the memory architecture of AMD Instinct GPUs to deliver accurate, cost-effective, and fully traceable reasoning at scale for enterprise AI applications.

Segments1 record

Ideal customer profiles1 record

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

AI capability3 records

Feature2 records

Partnerships

One partnership is on record.

• AMD (Advanced Micro Devices)coreStrategic or Co-development PartnerAMD acquired MK1 in November 2025. The MK1 team joined AMD's Artificial Intelligence Group, bringing high-speed inference and reasoning technologies optimized for AMD Instinct GPUs. MK1's Flywheel technology serves over 1 trillion tokens daily and is now integrated into AMD's enterprise AI software stack to enhance AI performance and capabilities.

Scale indicators1 record

Recent moves4 records

Expansion highlights4 records

Direct peers

• Replicate: Replicate runs a cloud for running and deploying open-source ML models, with a focus on fast, scalable inference APIs. It is comparable to MK1 because both target developer-friendly, high-throughput inference as a service for production AI applications.
• Anyscale: Anyscale, the company behind the open-source Ray and vLLM projects, offers a managed inference and compute platform for LLMs. It is directly comparable to MK1 as an enterprise-facing inference optimization layer serving large-scale token workloads.
• DeepInfra: DeepInfra operates a serverless inference API for open-source LLMs optimized for low-latency, low-cost serving. It is a direct peer to MK1's Flywheel because both monetize fast, cost-efficient LLM inference for enterprise developers, though DeepInfra is hardware-agnostic.
• Fireworks AI: Fireworks AI provides a high-performance inference cloud and runtime optimized for open-source and custom LLMs. Like MK1, it competes on inference speed, cost-per-token, and developer ergonomics for enterprise AI deployments, making it a direct comparison for MK1's inference stack.
• Modular AI: Modular builds the Mojo language and the Modular Inference Engine, focused on high-performance AI inference across hardware backends. It is a direct peer because it competes on inference runtime performance and developer-facing optimization, though it is hardware-agnostic unlike MK1.
• Together AI: Together AI runs an inference and training cloud purpose-built around fast LLM serving and custom model optimization. It is a direct peer to MK1 because both target high-throughput, low-cost enterprise inference at scale, though Together is hardware-agnostic whereas MK1 is AMD-specific.

Emerging players

• Cerebras Systems: Cerebras builds wafer-scale AI accelerators and an associated inference cloud. It is comparable to MK1 as an alternative integrated-stack challenger in enterprise inference, though its hardware strategy is fundamentally different from AMD Instinct-based deployments.
• SambaNova Systems: SambaNova provides an integrated hardware-software AI platform (RDU chips plus a full-stack inference and training system) targeted at enterprise AI. It is comparable to MK1 because it pursues a tightly co-designed stack to win enterprise inference workloads, though at much larger scale and with proprietary silicon.
• Groq: Groq offers a custom LPU inference cloud and is best known for low-latency, high-throughput LLM serving. It is comparable to MK1 because both sell 'fast inference at scale' as their core value proposition, although Groq pairs its software with proprietary silicon whereas MK1's stack is AMD-only.

Broad incumbents

• OctoAI (acquired by NVIDIA): OctoAI provided an enterprise AI inference and fine-tuning platform before being acquired by NVIDIA. It is comparable to MK1 because both sold inference optimization for production LLM workloads, though OctoAI's stack was NVIDIA-centric and is now part of NVIDIA's enterprise AI offerings.

Market position

Strengths4 records

Weaknesses4 records

Competitive moat4 records

Key risks5 records

Key highlights5 records

Customer concentration

Revenue estimate

Valuation estimate

Number of profiles

Funding overview

Funding rounds2 records

Investors2 records

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about MK1