Maincode today announced a $30 million investment to build Australia’s most advanced AI factory, a new state-of-the-art facility in Melbourne set to open in January 2026.
Known internally as MC-2, the new factory marks a major leap forward in Maincode’s mission to make Australian-Made AI a global force. It will expand Maincode’s capability to design, build, and operate advanced, task-specific AI models that are engineered to handle the kind of structured, rule-based, and context-sensitive workloads that traditional general-purpose LLMs struggle to deliver.
A New Chapter for Australian-Made AI
Following the success of MC-1, which produced Matilda, Maincode’s first fully built and trained large language model, the company is now doubling down on what it does best: building precise, high-performance models that solve real operational problems for customers.
“MC-2 is more than an expansion, it’s a statement of belief, ” said Dave Lemphers, Co-Founder and CEO of Maincode. “We’re not chasing size for its own sake. We’re building advanced, focused models that actually work, models built with the precision, efficiency, and practicality that define Australian Engineering.”
Beyond Hardware: Accelerating Through the Token Layer
At the heart of MC-2 is next-generation compute and storage infrastructure, powered by advanced GPUs and platforms from AMD. The factory combines AMD Instinct™ MI355 accelerators, AMD EPYC™ 9575F server CPUs, and AMD ROCm™ software to deliver high-performance training and efficient inference within an integrated environment designed for reliability, scalability, and fast time to value.
The token layer is where intelligence is shaped, through deep software systems that govern how models are trained, adapted, and applied to specific problem spaces. By focusing on the token layer, Maincode moves beyond the one-size-fits-all limitations of generic LLMs. It allows the company to build purpose-built models from the ground up, unlocking real capability without the endless trial and error of prompt engineering.
“Anyone who’s spent time trying to prompt-engineer a general-purpose model to follow complex rules or logic knows the limits,” said Lemphers. “That’s where we come in. MC-2 lets us build and operate specialized models from the ground up, custom-built to handle the exact decision-making and reasoning tasks our customers need, without the guesswork.”
“At AMD, we have a bold vision to enable AI everywhere, bringing open, high-performance computing to developers, start-ups, and enterprises around the world,” said Peter Chambers, Senior Director, OEM Sales, APJ and Country Manager, Australia, AMD. “With our latest generations of GPUs, CPUs, and software stacks, MC-2 is designed to deliver frontier AI performance with exceptional energy efficiency. We’re proud to support Maincode in driving locally led innovation with global reach and to help advance Australia’s AI Ecosystem.”
From Frustration to Factory
Many of Maincode’s partners began their AI journey with off-the-shelf LLMs, experimenting with prompts, testing APIs, and running into walls when precision and control mattered most. MC-2 exists for what comes next. It gives those companies a place to build their own purpose-designed AI models, developed, trained, and managed by Maincode within an environment optimized for fast iteration, reliability, and measurable value.
“We built our first factory because we needed to,” Lemphers said. “We wanted AI that could think and reason in the ways businesses actually work. MC-2 is the next evolution, a place where we can build those models faster, smarter, and at scale.”
From Metal to Model
MC-2 brings together compute, data, research, and deployment into one integrated environment, a foundation for continuous AI creation and improvement. It will power the next generation of Matilda models, customer-specific systems, and applied AI research programs, all managed end-to-end by Maincode.
This new foundation also accelerates the next frontier of applied intelligence: agentic AI. By building models from scratch for purpose, rather than repurposing general chat systems, Maincode enables AI that can reason, plan, and take action within defined boundaries, the kind of adaptive, goal-directed systems that will transform how work gets done across industries.
Building the Future of Applied AI
Maincode’s $30 million commitment represents one of the largest private AI infrastructure investments in Australia’s history, a statement of long-term belief that the future of AI lies in applied intelligence, not artificial generality.
“What we’re building with MC-2 is capability, ” said Lemphers. “We’re using the best technology in the world and applying it with Australian inventiveness. We’re not a chatbot company, we’re an AI company that builds entirely new models from scratch, systems designed to handle the work that general-purpose models can’t, no matter how much prompt engineering you try. MC-2 isn’t about scale, it’s about precision. It’s where the next generation of useful, usable AI gets made.”
Key Facts
- Investment: $30 million
- Facility: MC-2 – Melbourne AI Factory
- Opening: January 2026
- Technology: AMD Instinct™ MI355 GPUs, AMD EPYC™ 9575F server CPUs, and AMD ROCm™ software providing high-performance training and inference capability
- Focus: Building advanced, task-specific AI models through deep investment in the token layer
- Flagship Model: Matilda, Australia’s first fully built and trained large language model
- Outcome: Expands Maincode’s capability to design, build, and operate customer-specific AI systems, turning frustration with generic LLMs into high-performance, purpose-built solutions