Transcription of The Stock Network Interview with Scx.Ai Holdings (ASX:SCX), Co-Founder and CEO David Keane
David Keane: Hello, my name is David Key, the CEO and co-founder of SCX.AI Holdings. We are bringing faster, cost-effective and fully sovereign AI to all Australians. We believe that AI is changing the way every Australian works and lives their lives, but we need to provide access to the high-speed, cost-efficient and fully sovereign AI infrastructure to power the next generation of AI applications and agents that every Australian will need.
Before we talk more about the business, I want to run through an important notice and disclaimer as an ASX listed company for your review. Great, having said that, let’s go back and talk about Australia’s AI challenge. We know that Australia faces critical challenges in AI implementation.
We know that all Australians today are already embracing AI, but millions of Australians every day are sending their most sensitive information overseas just to use AI. As regulatory pressure increases and as ordinary Australians continue to feel the need to understand what happens to their information and to be able to afford to use AI to build these new AI applications and agents that they need in their businesses, we need to think in a new way about how the technology can be redeployed. So if you think about those security vulnerabilities and data sovereignty concerns that many Australians have today, you know, a lot of organisations are simply using overseas hyperscalers or overseas AI providers to be able to build their AI applications.
And we believe that’s causing significant challenges for the deployment of AI. But at the same time, the amount of power and water that’s required to create these next generation AI data centres really can impact the ability for Australians to be able to use AI successfully in their everyday lives. And those requirements impact cost and also they impact the speed at which we can deploy AI infrastructure.
And lastly, building AI is expensive. The cost of the AI chips and when we’re deploying international capacity, we have to factor that in as well. So these are real challenges that have been holding back Australia’s deployment and adoption of AI at scale.
To understand how SCX solves that challenge, we like to start by breaking the AI world into two distinct camps. First of all, there’s AI training. AI training is where we take a whole lot of information and we start to build these things called AI models.
And the process of training is a very particular process with a very particular way of working. You get a whole lot of information, you put it into an algorithm, you put it onto a very large compute cluster and you sit there and run a training load when no one’s using it, it’s actually just running these algorithms to create these AI models. And that can take weeks or months to run.
When we do that, we create the model weights, which are the numbers that go inside the deep neural networks that power AI models. But once we’ve trained an AI model, we have to use it. And the industry has chosen the word inference to mean using AI capacity.
Now, inference is a very different kind of AI workload. It’s where we use those trained models in those applications and agents. And those can be dynamic.
They can be different models used at different times to run an application or an agent. You can have many users using the same system concurrently and you require low latency and high speed. And an inference workload is measuring tokens.
So we believe that inference is the core production market in AI now and into the future. Looking at some of the industry analyst numbers, you can see here how the dark green bar in that middle chart for AI inference is growing very, very rapidly. We are seeing every Australian start to become an AI inferencer as they use AI applications every day, whether it’s a chat bot that they’re talking to or an application for banking, an application in healthcare, even an application in their private lives.
And so AI inference is where the AI workloads are going. And we believe that by focusing on AI inference, we can add even more value to our business. SCX is using some purpose-built inference infrastructure to address that growing market.
We use application-specific integrated circuits or ASICs inside specialized AI hardware machines that can increase the amount of tokens per watt we generate. And they can do that delivering significantly faster performance speeds. And whilst achieving those goals, they also run in air cooling-based environments.
They do not use water to cool these chips. And we’re working with this partner called Sanbonova Systems that makes these chips. And this enables SCX to have very strong unit economics and enables us to produce the capacity in AI, these tokens more efficiently than others in the market.
It also means that we work inside existing buildings. We don’t need new data centers and new construction to power our AI factories. We just run an existing established facilities today.
All this translates to those leading unit economics and to the infrastructure efficiencies that we think are necessary to build a successful company into the future. But together with those new specialized chips, SCX also has an own your model strategy. And this means that our customers are able to run leading open weights models and even fine tune their own, create their own AI model and run it directly on the SCX infrastructure.
And they can maintain access to those models indefinitely. They’re not gonna be forced to upgrade and move when a vendor changes their mind on the delivery of a new model. And this concept of no forced upgrades or API dependencies gives our customers additional safety, security and control over how their AI workloads are performing.
We call this the no lock-in, no platform risk model and enables our open standards and interoperability to drive the model choice for our customers. But at the same time as supporting the open weights models, SCX also has created our own Australian sovereign models. These are models where we’ve taken the open weights foundations and we’ve fine-tuned or post-trained or localized those models to work well in the Australian context.
And this is incredibly important for people building AI applications in Australia as it enables them to ensure that their applications are perfectly designed to work inside the Australian regulatory and judicial system as well as understanding the nuances of our culture. We have two models today, SCX Coder which is our high-performance sovereign coding assistant designed for reasoning and helping to build agentic applications and SCX Magpie, which is our fine-tuned Australian model with local Australian context reasoning. And these models together with our open weights strategy give SCX the flexibility to deliver the models customers need with specialized models built for Australians.
So overall, SCX delivers on what we call the three pillars of AI inference. We provide energy efficient inference using our ASIC based AI infrastructure designed with power and cooling efficiencies in mind. We have an inference optimized system designed for high throughput and low latency AI inference which we believe is important for the future of AI applications and agents.
And it’s designed with privacy and sovereignty in mind Australian based infrastructure supporting data residency, security and regulatory compliance which enables SCX to build a solution that gives our customers confidence in how they use AI. So we believe SCX is well-placed to create the future of AI capacity here in Australia. We are already serving a number of Australian organizations who need access to that high-performance, cost-efficient and fully sovereign AI infrastructure.
And we’re looking forward to talking to many more organizations about how SCX is delivering the future of AI for all Australians.
Ends