NAVFIELD, Mr Dan, Director, Nexrupta Holdings Pty Ltd SENTI, Mr Dale, Director, Nexrupta Holdings Pty Ltd TAYLOR-PRICE, Mr Rupert, Chief Executive Officer, Vault Systems Pty Ltd [10:02] CHAIR: Welcome. I understand that information on parliamentary privilege and the protection of witnesses has been provided to you. Do you have anything to say about the capacity in which you are joining us today? Mr Senti : I am the founder and director of Nexrupta as well as one of the owners of Gulanga. Mr Navfield : I am one of the directors of Nexrupta, as well as a director of a business called Hide and Seek Digital. Mr Taylor-Price : I am the CEO and founder of Vault Cloud. CHAIR: I now invite you to provide us with a short opening statement, after which we'll get straight into questions. Over to you. Mr Senti : I'd like to thank the chair, the deputy chair and distinguished members of parliament that join us today on Gadigal country. I thank the committee for the opportunity to speak this morning. I'd first like to acknowledge the Gadigal people, pay respects to their elders present and emerging, and thank them for the opportunity to have this important yarn on their country. Nexrupta is a joint venture between Gulanga and Hide and Seek to provide opportunities for Australian companies to develop new capabilities and compete on a global scale. We work alongside companies like ID Exchange that create a global consent platform that, with the right levels of investment, would be a world leader in consent management. Today I'm here in my capacity as director of Nexrupta and general manager of Gulanga, an Indigenous business that delivers ICT to the public and private sectors, working alongside both Australian and multinational technology providers and creating opportunities for Indigenous Australians within the technology sector. As a technologist with 32 years experience in the technology sector, I believe I offer the committee a very unique perspective, having worked directly within all layers of the AI technology stack: data centres, infrastructure, cloud providers—you saw that with Vault—operating systems, model development—with Trellis Data, who are not here today—and the application layer, not to mention running multiple small businesses, helping startups, raising capital in Australia and now vibe coding, which I learnt on the weekend, and selling to government and the private sector. We are working with and alongside multinational businesses in Australia and extracting value from their value chain. I'm here today to provide a direct perspective on the challenges of small businesses in relation to AI and the levers I believe government can make to create better opportunities for Australians in the market. I believe Australia has a significant opportunity right now to build a sovereign AI market that is built on a fabric of global trust, which we have as a country. With the right investment and economic levers, we have the ability to create a strong export opportunity. The US companies that have come before us in this hearing have suggested Australia has the right conditions to create a strong AI industry. While I believe the investment for these companies is a necessity, we shouldn't disregard the opportunity to create strong sovereign resilience by establishing our own domestic market. As such, we are here today from our respective companies and small businesses of Australia advocating for AI sovereignty, as we believe there is a significant opportunity for Australia to take advantage of the next industrial revolution that's available to us with the development and adoption of AI. I thank the committee for its time today and am happy to answer any questions. Mr Navfield : I'm here today representing Nexrupta and Hide and Seek Digital, a business that consults into Australian government on IT, as well as working with small businesses and companies in Australia. We're working with a number of businesses to generate new technology across some important categories around trust, around consent, around identity and around the buzzword of 2026, which I think is 'context'. We are here today to help with the more geeky side of this conversation, so I am happy to ask and answer any questions on that side of things. Mr Taylor-Price : Thank you for the opportunity to be here on behalf of Vault Cloud. Vault's core purpose is to secure data and enable intelligence. For over a decade, we have built and operated secure, national hyperscale sovereign cloud infrastructure in Australia to support government, critical infrastructure and highly regulated industries. Our perspective comes from the practical realities of developing and running high-assurance cloud and artificial intelligence infrastructure where data security, operational resilience and jurisdictional control are paramount. As Australia considers its strategic approach to artificial intelligence, our central message to the committee is that Australia must treat AI infrastructure as a core strategic capability. Australia's ability to benefit from AI depends on the entire supply chain, from land and power through to cloud providers, model builders, system integrators and applications. On our current path, Australia will capture lower-tier supply chain benefits, such as data centre real estate and neocloud compute. However, we are actively transferring our national know-how and long-term economic value to multinational platforms, including foreign frontier AI labs. This transfer of value off Australia's balance sheet is occurring in exchange for cheap tokens. We have seen this economic extraction pattern before: domestic taxi services replaced by multinational rideshare; domestic retail yielding to global online platforms; traditional media superseded by foreign social networks; and domestic payment systems giving way to mobile payment gateways. In each case, consumers gained short-term convenience, but long-term value left Australia, reducing our national self-reliance. Australia possesses vast natural resources and technical talent. Why then are we not generating sovereign, cost-competitive AI tokens locally? I will put forward three reasons. First, procurement psychology—Australian business and government retain an ingrained bias towards buying multinational tech over domestic capability. I have spoken on this at other Senate inquiries. Second, taxation asymmetry—Australian companies face full domestic tax obligations, whereas foreign multinationals leverage trade protections and offshore structures. When a local business loses a deal over a one per cent higher cost, Australia loses the entire tax yield and sovereign capability. Third, capital cost handicap—Australian tech providers face capital costs at least 15 per cent higher than foreign competitors, compounded by bank serviceability models that penalise local tax structures. In other words, if I go to the bank, they will factor in the tax I have to pay then reduce the amount they will lend me and then increase the cost of capital in doing so, which ingrains a systemic disadvantage. I am pro regulation and specifically pro AI regulation. However, our major trading partners have made it clear that they will not allow regulation to slow their domestic AI capability. If Australian regulation creates disproportionate friction, it will accelerate the movement of Australian IP offshore, out of the reach of Australian jurisdiction entirely. Australia has extraordinary natural advantages in AI and no fundamental systemic disadvantages. Whether AI becomes the driver for national prosperity or a source of economic and security vulnerability depends entirely on government regulation and procurement choices. To secure Australia's AI future, we urge the committee to recommend enforceable sovereign procurement targets across government AI and cloud spending, tax and capital equity policies that level the playing field for domestic tech infrastructure and regulatory guardrails that protect security without driving local innovation offshore. I look forward to your questions. CHAIR: Thanks very much for those opening remarks. Something that I'm particularly keen to get input on is this sense of what sovereign capability means and what AI sovereignty means. I am really keen to get your views on what this means for Australia. What do we, as part of the AI stack, need to control, rather than simply access, for sovereignty to be meaningful? Mr Taylor-Price : There are different value propositions at different layers of the stack. Data centres have obviously been where there's been a massive economic boom for Australia. You can't displace that. The power is connected. The water is connected. It's not something that can be extracted or moved away. There's almost natural value that will occur there, regardless of what happens. I think there are a lot of questions around social licence and things like that and how you make that more successful. As you go further up the stack, it's where the intellectual property sits. You can simply look at it as—what's the value on the balance sheet of these businesses? When a company that owns an asset, like a farm, decreases in value because the know-how in how to use that farm transfers into another offshore organisation, it's a very clear transfer of value. The more of that value stack that Australia participates in, the more capability it has. I don't think that it necessarily needs to be Australian capability for Australia. We also want to participate where we're exporting that capability and participate in the global market as well. Mr Senti : I don't think I have anything to add to that. There was the Qantas test that, I think, DFAT released a number of years ago that talked about sovereignty—IP ownership, control of the technology et cetera. I think referencing, going back and having a look, at that would definitely be worthwhile to understand how we map sovereignty to these sorts of things. Like Rupert said, all elements of the stack—data centres, I think, are pretty well ticked. It's the next layer in—who owns the infrastructure, who maintains that infrastructure and so on. CHAIR: We've got these big hyperscalers. We heard from them yesterday. They are indicating a keen interest to come here and bring significant compute to Australia. What, in your view, is the opportunity for us to leverage off of that keenness to invest? How do we make sure we really benefit from that compute? We are being told that that is how we get a say in our own AI future. That's how we keep up with frontier access and that's how we keep up with our defence capabilities and national security. I'm particularly interested, given your experience, in how we make sure that translates to an Australian AI ecosystem. Mr Senti : From a view of those large multinationals—we need them, without a doubt. I think the investment they make is critical to be able to build out this infrastructure and scale it and so on. But we can't disregard the fact that Australian companies can do the same—can build equivalent models and can build things here, domestically, as well, to give us that sovereign resiliency. So I think we can leverage them where we need to, but, at the same time, we do need to focus on creating our own infrastructure and our own capabilities as well. At the same time, that infrastructure is going to be deprecated over time, and some of those providers will offload that infrastructure, which we then can use for other models, so there's a way to recycle some of that investment over time. That's a five-year sort of journey. I think there are opportunities on both sides. Mr Navfield : I think the multinationals are looking at a vertical solution—as in, taking each part of the stack and taking that business—so I think we have to test where we can have sovereign capability in the different layers. I think the test is not all of it should go to the multinationals. If they get the hyperscaling compute, maybe they don't get the model and the different parts. We need to consider where sovereign can provide that solution and ensure that the complete vertical is not completely consumed by multinationals. CHAIR: If we don't—if they decide not to pursue Australia as a place to train—what does that mean? What do we potentially lose out of that? What does that potentially mean that we don't have to consider? Obviously, there are a range of things—not obstacles but challenges—that we need to make some pretty tough policy decisions around, but what does that mean from your end, do you think, if we do end up missing out? Mr Senti : Start investing—we really have to start investing. We need to at least be in the game, and if we're not in the game then we're going to miss out in a number of areas. I think Australia as a society is really good with technology. A lot of the large multinationals come and test their technology in Australia before delivering it out to the rest of the world because we're sort of isolated but we're also great innovators and we adopt technology probably better than a lot of countries. I think investment is critical in all aspects of this, but investment is multifaceted. It's investment in Australian sovereign companies from a government and corporate perspective, in buying our technology, but also making strategic investments in the underlying infrastructure. Mr Taylor-Price : I'm pro foreign investment into this space. I think it's a good thing for Australia. I'll split the discussion into training and inferencing. On the training side, it doesn't really matter where that training occurs, to a large degree. When that training is performed in Australia, apart from any argument around social licence and natural resources that it's consuming, it's ultimately delivering economic value to Australia. It doesn't really matter that much from a consumer perspective because those models then get moved and distributed around the world. From an inferencing perspective, I think Australia is better off, when it is using global systems, if those systems are housed within Australia. With inferencing, your own personal information and sensitive information is getting consumed and used by those models. Having that under regulatory control from Australia, I think, is a positive thing, but that doesn't stop the transfer of know-how and IP. Whether you're consuming from a server in Australia or from a server offshore for inferencing, you're still transferring your know-how, and they're training on your information as that goes offshore. So there's still a transfer of economic value from Australian businesses into the balance sheets of AI companies. When it's a foreign company onshore, I still think that's better than it being a foreign company offshore. But, to really retain the economic value, you need that Australian supply chain through the whole stack. Mr Navfield : To give a slightly contrarian view on what Australia misses out on, I think we end up at an N minus one for frontier models. We have the ability to deliver models that are almost as good as the frontier. If we didn't have the investment from the hyperscalers and the overseas inference, we may just have a slightly behind model that is perfectly satisfactory for most of Australia. I think international investment is incredibly important to give us the frontier, but we're not missing out on a huge amount if we don't get it. CHAIR: Deputy Chair. Senator DUNIAM: Good answers on sovereignty—tick. I suppose a key focus of this committee, as we try and find a way from where we are today to where I think, as a country, we want to be is, as part of that, making sure we get the best deal or outcome for Australia. Part of that is what's in it for us. It's not just about allowing foreign operators to come and plonk themselves here and do what they want and get away with it. In the Vault submission, echoed by what Nexrupta said around some of the things that need to be dealt with there, I do want to go to the procurement targets. Obviously, a sign of confidence in Australian enterprise and innovation is when a government puts some skin in the game, so can I get from all of you some more detail on how you would see the procurement thing being set up in a successful way here. Are there any jurisdictions elsewhere around the world that we could look to as a rip and read? Mr Taylor-Price : There are two issues. The issue I said before is around the nature of government procurement—how it's administered, the rules and the outcomes—and that's not specific to AI. You can see that broadly across technology procurement. It's something that's been discussed in parliament many times. The other chilling reality is if you just look at the scale of what the Australian government is doing in AI. For instance, for GovAI, in our forecasts we would look at the total addressable market of that as probably about 200 GPUs. If I try and sell 200 GPUs globally, most people will tell me that's too small a package to actually sell. They'll say that we don't buy in small chunks like that. Even 1,000 units is a small procurement. The scale of what Australian businesses and Australian government are doing in AI is relatively small to the size of our economy here, so we're behind the eight ball in terms of delivery. Then we have those procurement biases put on top, and those two things compound to be quite significant. For our organisation, even though we're an entirely Australian focused business, we have to fill essentially the volume of our pipeline by supplying global supply chain. That's a sad thing, but that's the reality that we're in today. So it's not just around having more equitable procurement processes in government and in businesses; it's also around getting Australia to have the scale of consumption as well. Mr Senti : From my perspective, in our submission, we put forward an AI economic flywheel, which talked about some government targets: 15 per cent, first year; 25 per cent; and 30 to 35 per cent in year 3. I actually modelled that on the Indigenous Procurement Policy, which is now in its 11th year. I think it's about a $12 billion industry. The learning from that is, 'Hey, why can't we do other things in a similar way?' The percentage I came up with was reflective of some unsolicited proposals that I put forward to government in the past. I put a multidomain defence security capability on cloud several years ago, and then lately, to the Department of Finance, I put forward an AI automation and orchestration platform. Within both of those unsolicited proposals, in the defence one I was able to achieve at least 35 per cent sovereignty, and with the AI orchestration one there was about 60 per cent sovereignty in that. So, if I can put forward two proposals like that and achieve those sorts of numbers, I can't see why 15 per cent wouldn't be achievable if we incorporate something that we're working on at the moment, which is Dukes. We're working with Trellis Data, who's not with us today, to have an Indigenous LLM called Dukes, which is named after my partner, who passed away a couple of years ago. The benefit of that in that unsolicited proposal model is that we can increase sovereignty up to about 70 per cent. The way we do that is by partnering with Workato, who you'll hear from later today. They've agreed and put forward and signed an MOU to say that they're willing to include Dukes in the way that they deliver their agentic models. So we now have the ability to slot in an Australian sovereign provider into a global company and stand on the shoulders of giants. That opportunity, when you cascade it out, is pretty exciting, because not only do we do that in Australia. If it works effectively in Australia, then they've got 25,000 customers globally. We could potentially create an export market in and around that. The wrap-up of that and the exciting part for me is that I'm not actually just doing that. A portion of what Workato spends on Dukes is put into an Indigenous fund, and that's going to go towards building a platform that creates AI agents to train and provide tutors for Indigenous kids—so map it on the Australian curricula, where Indigenous kids will then get access to these tutors to help them actually learn. The inspiration actually came from my kids. I've got two daughters, 16 and 13, and they've learned how to vibe code and do stuff. They've built their own platforms to teach them how to do various things at school. There are complex things that they learn at school, and, if I can create a bot to help me learn and a tutor to do that, then I can actually start to get better grades and so on. The result of that is that they're getting a lot better grades. They're pretty well straight-A students now as a result of technology helping them learn. If we can deliver that into a model and create it as a by-product of commercial transactions, then you can create a pretty interesting sovereign capability around this. Senator DUNIAM: Mr Senti, you've just explained a solution to one of the problems that was referenced earlier. Mr Senti : One-hundred per cent. Senator DUNIAM: The other one is the scale. You talked about 200 GPUs. Is that just down to government being at an early stage of trying to grasp what this might or could do for government? Is that the problem? Mr Taylor-Price : Yes. I probably don't have a full explanation as to why. We've got a business in the US and a business in the UK as well that are sister companies under the same brand. What we're seeing there is government is buying absolutely large quantities on large programs. There are numbers—certainly in the US obviously—that absolutely dwarf what we're doing here, but, even in the UK, there's appetite from the UK government to roll out large-scale programs in AI. We've just seen a stronger appetite for investment in that space. Why is it so conservative here? I'm not sure, but we've certainly comparatively seen Australia's investment be lower. Mr Senti : From my perspective, what I've been seeing is this trust barrier. If you have a look at the way government's using AI now, it's very much chat connecting to some data, but it's not necessarily connecting into the corporate systems of government as such. You can't really do that until you put a layer in there. Like I said, you'll hear from Workato later. They've got an orchestration layer that sits in the middle that then connects all of those corporate systems in. You can expose it to your AI layers, and then you can actually break what we are seeing as the trust barrier, and then, if you focus your AI efforts on internal systems as opposed to external facing stuff, you can actually start to deliver some really inherent capability. We've been doing work with Fair Work Australia, who's getting absolutely inundated now from requests, because everyone that gets sacked or loses their job goes straight to AI and asks, 'Hey, do I have a case?' and then, all of a sudden, they're getting inundated. They've started to create systems internally where their IT staff, as an example, can now get agents to do things that, traditionally, IT people would have to do. As an example, we're all using videoconferencing today. How do you make sure that those things are up before a session like this? They've got an agent doing that on a very, very quick interval every day. And then, if something goes down, they can act upon it, which is a really good use case of AI, right? It doesn't need a large language model to do it at a frontier level. It can be a pretty mediocre model to do that sort of stuff. Mr Navfield : Just to offer a viewpoint, I think Australian government has had its learnings come from Microsoft around AI. Every government department was provided Copilot at a discount or free—whatever the incentive was. It was very early in the AI journey. The chat experience the government have had with that interface, which is a very early-stage version of AI, has taught them that they can't trust it, and so they shouldn't look at work models that involve AI at this stage. We're quite behind the ball. In the UK, as we've said, there's massive demand for new innovative systems using AI, whereas we presume we already have it and this is how good it is, which is not the case. Mr Taylor-Price : I'd also add to that. Yes, there's a propensity, I think, for Australian government to consume services rather than build services. The ability to just pay for more licences for something is quite an easy process for them to run through the procurement process. To actually design, architect and understand something and invest into the creation of capability—I think the Australian government finds that much harder to do than other comparative governments. Senator DUNIAM: You, at some level, have interacted with government agencies. What's their willingness to talk to Australian vendors or service providers generally? Secondly, to that point around willingness or otherwise, when it comes to delving into the world of AI, embracing it and taking it, are you seeing a mindset or is it just—to your point, Mr Senti—all a trust thing? What's the experience within government agency? Mr Senti : Part and parcel for us, but, if you look at the way government are operating today, I think they're just getting inundated by vendors. It's pretty hard to understand what's available to them and what's real, what's not. It's a journey. AI is a journey. Even for myself, it's a personal journey. I learnt to vibe code on the weekend, and it was like a light bulb went off in my head because there's so much capability I can create. Senator DUNIAM: Does anyone here vibe code? I don't. Senator Sheldon does! Mr Senti : It's learning that sort of stuff and then being able to apply that. I think the way to actually overcome it is to put challenges out there to industry and make it sovereign challenges. The department of industry did this a few years ago. It was an innovation challenge. They went out to sovereign companies only, and they said: 'Hey, we've got these business problems in these agencies. Can you put in a response?' I'll talk on behalf of Trellis here because I was working with Trellis at the time. They put forward a response to department of agriculture on one of their challenges. One of the things they were saying was: 'Hey, we're getting a whole heap of containers coming through. We need to be able to check if those containers have insects and other foreign things coming in.' They used camera technology to identify those insects and stuff like that. They won that challenge and subsequently got more funding to then go out and build it. I think if government was to take on more of those things, I think we'd have a very, very quick and rapid capability to respond. I'll give you a great example. On the weekend, preparing for this, I downloaded 245 of the submissions to get an understanding of what was being talked about. It took me more time to download them than what it did to actually create an agent to tell me anything I wanted to know about anything. If I wanted to make an assessment on the deputy chair, I could ask, 'Hey, what sort of questions does he ask?' Senator DUNIAM: It would not take long. Two words—and they're not good! Mr Senti : The technology is there. It's understanding the art of the possible. When you start to play with this technology and get exposed to it, your learning curve goes through the roof. Senator DUNIAM: Thank you. CHAIR: Ms Chaney. Ms CHANEY: Thanks very much. I've got some questions about how the market is likely to evolve and where Australia could really have some advantage. My understanding is, right now, we're using frontier models to do a lot of things, but, when they are fully costed, there's likely to be more of a desegregation of the market with lower cost models used to do lower complexity tasks and frontier models used to do the cutting-edge stuff. Is that how you see the market developing? Then, in which types of models or, alternatively, parts of the stack do you see Australia having some sort of advantage or being able to be competitive, especially if we start to move into cost competition? Added to that also is in which parts of the economy or government do we actually need the frontier models as opposed to versions that we can build ourselves? Mr Senti : As a de-technical person, I'm going to hand over to Dan on this, because he'll probably understand how to answer this better. Mr Navfield : To give an opinion, we're already seeing the evolution of the market of LLMs where certain LLMs are starting to provide specialist capability. I see the future being an expansion of that where there's going to be a lot of providers of LLMs, not just the frontier models, and they will likely provide unique capability in a certain area. Australia has some amazing data around certain areas—geographical, minerals, biotechnology et cetera, and so there's a real opportunity for us to be the LLM provider or the inference provider around this specialist area. It's something we're really good at, and that's an opportunity to export that information. Where I see the future being is that you will have a platter of inference available to you, and you will select the right model for the right job. We're already seeing a change. OpenRouter data from this year sees that 75 per cent of inference for development was going to frontier models just six months ago. That has changed the other way now to 75 per cent of the inference going to open models now for software development. We're already seeing a shift to different models, not just the big frontier ones, to do certain tasks. The opportunity for Australia is to focus in on choosing the right models for the right job—where we need to consume them at the right financial position—and then to offer up specialist capability back to the world that we can export. Mr Senti : If you look at what we're doing with Duke's, as an example, with Workato, it's a 40-billion parameter model. When we put it against a US model, it outperformed. It provided a better response. It was more Australianised, and it was more nuanced to the questions that were asking and the capability of the agent that we were trying to deliver. We can certainly compete, in my opinion, in certain nuanced activities. The frontier models, in my opinion, are really, really good for deep research and things that you really want to dig into quite deeply. You don't need to have all models for everything. Ms CHANEY: While we're talking about building models, is it data sets that are likely to be a source of potential competitive advantage for Australia, or are there other sources of competitive advantage in LLM development that Australia has that means we could specialise? Mr Senti : Interestingly enough, I spent some time with Nvidia probably about four or five years ago, when I was working with one of the data centre providers, and they made the claim they were doing a lot of work with CSIRO. Apparently, we have the largest data set of cotton in the world. We know more about cotton than any other country. One of the things that we were talking about through some of the conversations we had with government was what if you could use that data set and a model to grow different coloured cotton. As opposed to just white cotton and then dye it, could we grow different coloured cotton which, potentially, could be environmentally better, or could we use less water to grow cotton? There are industry— Ms CHANEY: The data sets are the advantage there. Mr Senti : Yes. It's a matter of identifying those industry verticals where those data sets in Australia are quite prevalent. Ms CHANEY: That's in LLM development. What about across the stack? Are there sources—obviously, data centres are a bit of a source of advantage with land and power. Are there other parts of the stack where Australia has some potential competitive advantage where, as the market evolves, we could emerge across the stack? Mr Taylor-Price : I'd say we have a trust and regulatory advantage. If you think about inferencing where you do have a lot of sensitive information going in, our experience—if you think about Asia, Australia is not in the population centre, so, from a latency perspective, in terms of let's say sort of two billion people in the world, we don't have a particularly strong advantage compared with other parts of Asia. However, from a regulatory and trust perspective, we're essentially at the top of the pile, so people's willingness to put to put an inferencing system in Australia to serve Asia is higher than someone's willingness to put an inferencing system somewhere else in Asia to serve Australia, for example, so there's a natural advantage for inferencing models in neo-clouds in that space as well. In actual fact, Trellis' biggest success right now is in Asia. Malaysia is taking on—Indonesia has got opportunities in there. They're very, very excited about the Australian technology. To Rupert's point, that trust—as Australians, we have global trust, compared to the likes of China and the US. The US has more of a diminishing trust. There's not a lot of trust in China, but, in Australia, as a global citizen, we are well trusted. Ms CHANEY: I understand the training model's latency is less of an issue. For inference models, it likely to become more or less of an issue in inferencing, or is that likely to just be constant? Mr Taylor-Price : It depends on the use case. I'll try and give a short answer rather than a long one. There are many items where you can do, for instance, batch processing of testing of systems where even five minutes latency—it could almost be on another planet. That's okay even in the inferencing case. There are other things where you're trying to have real-time interactions with real-world scenarios where even milliseconds count. There's a really wide spectrum of that, and so there will always be cases where inferencing needs to go to the edge, and then there are other cases where you want the efficiency of centralised inferencing as well, so you're going to see the market evolve in a number of ways, I think, to meet the various needs of consumers. Mr Navfield : I think it's really important to mention that most of the use cases of Australia and, indeed, the world at the moment can be served by an N-minus-one model, as in they do not require the frontier model and they do not require frontier latency. Mr Taylor-Price : Probably 90 per cent, I would say, in our experience. Ms CHANEY: That leads to my last question. What are the areas where we do need frontier models? I imagine national security is one of them. What are the other ones? Mr Taylor-Price : Again, I don't think it's quite as simple as sectors. There are bits of workflow that you need. If you think about the medical system, sometimes you need a GP, sometimes you need a specialist, and sometimes you need the best person in the world that knows exactly the thing that you've got. It's the same in models. I don't think you can just take a sector and say, 'This sector needs frontier models.' You've got to break up the workloads within the sector. I'd say all sectors benefit from a doctor or a PhD-level person in some areas, but the majority of the work—even if you look at an organisation, the majority of work that an organisation needs is not the absolute smartest brain at all time. It's very inefficient, in fact, to get the smartest brain in the organisation to manage the diary and things like that. So it is a nuanced answer unfortunately; it's not a simple answer. Mr Navfield : I'll give you a very direct answer: curing cancer. Really complex, world-changing problems are where frontier models are best suited. Deep research—the very strong development of ideas—is where they do their best work. As we said, you don't need them for most work that happens every day. It is important to have them available. They absolutely make up part of the offering that we need to have available to Australia, but they make up a smaller portion than we are led to believe. The important thing at the moment is to look at that bigger portion and say, 'Well, who can provide that bigger portion of everyday inference for doing the everyday tasks?' Again, having a super brain run your diary is not sensible. Having your super brain run complex payments in a space around entitlements might be important. And that will slowly change over time. So having availability— Ms CHANEY: For the 90 per cent that don't need a frontier model, can Australia be cost competitive? I'm trying to imagine being a worker, and you sit there and—in 10 years time—you've got five different models that you might choose from to do your daily tasks. You only use your frontier model for the really hard stuff. Is it realistic to think that some of those models will be Australian, if it's driven by cost? Mr Senti : I can actually answer this definitively. The Dukes model that we're working on with Workato and Trellis—comparable models—were one sixth of the price. Ms CHANEY: Why? How? Mr Senti : Infrastructure. We're taking out multiple layers of the way a business is run globally. There are multiple facets in that. The models aren't as big. There's not as much training that goes into them. But they're as capable in some areas, and they're capable for the things that we need to do with them. And, when you look at workflow and AI agents, most of that stuff is pretty 'workflow-y', so it's pretty easy to achieve what we need with these models. Ms CHANEY: And why wouldn't we get costed out of that as it's commoditised and Asia becomes more competitive? Mr Senti : Good point. I think scale is always going to create a competitive landscape in that respect. But, if government gets behind these things and we can start to deliver on a domestic scale, I think we can remain cost competitive. Mr Navfield : And then it's moving into that specialist provider category, right? It remains cost competitive because we have the data or the information to provide that specialist capability. Ms CHANEY: Thank you. CHAIR: Ms Mascarenhas. Ms MASCARENHAS: It's interesting because I know that we come from technical backgrounds, and that's often the space that we like to spend time in. What we're trying to do in Australia is explain to the public whether there is a social licence for data centres and AI, and you guys have referred to that. I want to understand what that means to your organisations and what you've embedded internally to create the social licence to operate here in Australia. Mr Navfield : In terms of social licence, the starting point is working with other boutique Australian businesses to provide our services. Living in the shadow of big four consulting and big four service providers that have been winning a lot of government work over the years has really banded together the boutique providers to use Australian products, to build Australian, and to provide them as Australian businesses. Mr Taylor-Price : Yes, we just closed something called 'Australian first right of refusal', which is essentially where we went to Australian industry and government and said, 'We'll let you essentially skip the queue to get access to capacity early.' I think, ultimately, the thing that's needed is that Australians benefit from AI. That's the fundamental thing that's needed. There are lots of things we can do to reduce harm, such as environmental factors and things like that. We've been net zero since July 2024 through the use of renewable energy and things like that. There are some things around reducing harm, but, ultimately, AI needs to economically improve the lives of Australians. And, if you think about what I said in my opening remarks about the transfer of value away, that's going to create job instability, self-reliance instability, national security instability. They're the things we need to fix. We need to make it very clear that, when Australia is coming to a town near an Australian citizen, it is a good thing for them, it is a good thing for the nation and it's a good thing for the economy. I think Australians are smart enough to see that, at the moment, that's not the trajectory that's happening and that's not what's happening. I think they have valid concerns around that. Ms MASCARENHAS: From a physical footprint perspective, at what stage does consultation start with the community? I heard what we're doing for the business community, but for locals that are physically within, let's say, the proximity of one of your facilities, what does consultation look like? Mr Taylor-Price : We operate inside data centres. We don't build data centres. We're reliant on the data centre providers to do that, so I should probably leave it to them to answer that. Ms MASCARENHAS: Perhaps, then, from when you decide on which data centre you go to, what are the procurement processes you have in place to ensure that you're going to, let's say, a data centre that has community buy-in versus a data centre that the community doesn't want? Mr Taylor-Price : It's a bit of a gold rush in data centres and AI at the moment, so there are a lot of players coming into the market that are very focused on the financial returns rather than creating a sustainable long-term business. We've been doing this as a business for 14 years, so we know who all the players are. We're really focusing our procurement towards those players that have honoured social licence in the past over a long period of time. To be honest, it's very obvious when you work with a data centre provider that they're just trying to get something stood up as quickly as possible, with as little care as possible, to try and make as much money as possible. It doesn't take very long for us, as a buyer of these services, to work out the difference between the serious, mature, long-term players that are doing it the right way and the people that are trying to make a quick buck. Ms MASCARENHAS: When you were referring to net zero emissions, was that for your physical offices, or does that also include the compute that's happening in it? Mr Taylor-Price : That's everything. We'd be Australia's—and probably one of the world's—leading provider on that. We have zero scope 1 emissions. We're neutral—we don't use carbon credits; it's electrically neutral—on scope 2. We do a lot of work with our supply chain around reducing their carbon emissions as well. That's in part because government's a key customer for us and government's communicated very strongly that that's a big buying factor. It's also very important to me personally. CHAIR: Senator Pocock. Senator DAVID POCOCK: Thanks for your time here today. It's great to see two Canberra based companies here. We hear a lot about sovereign capability and sovereign compute. You've outlined some of the concerns around government procurement and the way we overlook Australian businesses. I'm interested in how you would rate the Commonwealth government's direct investment in building this sovereign capability, both investing in Australian companies and investing in sovereign compute that can be accessed by Australian companies. How does it stack up against the billions we're seeing in Canada, the UK, Singapore, South Korea and all these other countries? Mr Taylor-Price : I feel bad for biting the hand that feeds me! I note that we are a partly Commonwealth owned entity through the National Reconstruction Fund, which has invested in us, and that has been a phenomenal experience with a phenomenal team of people. That has been very positive, but, when you compare it to the scale, as you said, in other countries, it's just not comparable. It's in millions and hundreds of millions when other countries are doing billions and hundreds of billions. It's orders of magnitude different. It's the private sector that's essentially filling in that funding gap at the moment. The government is not a credible investment partner for industry at this point, I would say. I think that that can change, but, if you just look at the numbers, they speak for themselves. Mr Navfield : In our submission we talked about guaranteed demand from government. At the moment there is very light demand to Australian business. The demand has assumed that it must be foreign sourced at the moment, so the demand needs to be recentred around Australia first, to see if there even is a capability, before we start to look at the assumption that we must go overseas. Mr Taylor-Price : Can I add something on the irony of that. I would say my backlog—about 98 per cent of our pipeline is offshore. The foreign companies think that we're able to provide the best service in the world, but the Australian government thinks it has to go to foreign companies to get the best service in the world. There's a direct contradiction in opinion there. Mr Navfield : Yes. Unfortunately, we don't have Trellis Data with us here today. They're building a very competitive model. They would say that they are starting to get some traction as a tier 3-something provider when they should be a tier 1 to tier 2 provider in terms of the interest they gain from government. Mr Senti : The flywheel that we put together, that talks about creating AI targets for government with investment and then creating an industry, is no different to what was achieved with the Indigenous Procurement Policy. We should be able to do the same. I think that's a great example. Rupert's company, Vault, got investment from the Future Fund. Now, if the Australian government was investing heavily in Vault, they would get a better return on their investment and create almost like a self-licking lollipop for themselves. I think it's crazy that we're not doing these things. Across the world it's happening. Why aren't we investing in our own capability? Because we are really good at what we do, I believe. Senator DAVID POCOCK: One of the things you talked about, and we canvassed this quite a bit yesterday, was this tax asymmetry. You have Aussie companies competing in procurement, competing on price, just selling to Australians and paying pretty much full freight company tax, and then you have big US tech companies paying—for some—no tax or 1½ per cent tax or 2½ per cent tax. How should the parliament be thinking about this? Clearly, with the way that you can profit shift and minimise taxable income as a multinational, should we be looking at some form of levy on non-Australian AI or tech companies to try and level this playing field? Mr Taylor-Price : I think the real challenge there is the free trade rules, right? They essentially constrain— Senator DAVID POCOCK: I guess the political perspective is that Donald Trump has basically kicked free trade agreements in the head. It's maybe an opportunity to actually do something as a country. Mr Taylor-Price : I think ultimately it just needs to be a level playing field. I'm not here asking for Australian companies to have an advantage. They should just not be disadvantaged by the very nature of them being Australian. That's my only request. I think, when you look at the taxation thing, you've really got to look deeper than just the tax bit you're taking; it's the whole investment case that you're taking away from when that happens. And I think companies should pay tax. I'm not arguing that we should pay less tax. But we should be taxed fairly and equally, and there should at least be a rule that says that an Australian company should not be disadvantaged by the nature of it being Australian in Australia. That seems like a very basic thing to agree on. Senator DAVID POCOCK: One of the things I hear in this space is that maybe we need to be looking at retained economic benefit. If you start to weigh up what Australian companies are doing—doing all the same things, like paying payroll tax, that multinationals are but then actually keeping IP here, building this sort of flywheel self-licking lollipop as I think you called it—should retained economic benefit be something that we're thinking about in this space? Mr Taylor-Price : In our experience in other countries, when we've gone overseas and expanded into the UK and US, they're absolutely clear on what that is and how to enforce it. There are teams of hundreds and hundreds of people in government whose job is to work out how to do that. They actively encourage us to redomicile into their countries—to move IP. There's no-one in Australia trying to retain us here. We're here because we love it, whereas in other countries they're very surgical around the value that they're trying to create in their economy. We just don't seem to have that same behaviour as a nation. I think that that is partly because we have a good ride on exporting minerals and things like that, so we've been able to get away without having that. But it's just such a loss for us to not have maturity in that space. Senator DAVID POCOCK: Finally—I know we're probably over time already—we hear a lot about access to compute. Does just having more US tech companies here, with data centres that they either own or are tenants at, solve that problem? How should we be thinking about Aussie researchers, startups and businesses having access to the best compute available in the world? Mr Taylor-Price : Australian companies are essentially able to make equally as good compute as any other country. There's no magic sauce in any country for organisations like this around neocloud. A lot of the differentiation comes around security and compliance and things like that. The raw product is pretty similar the world over. So—sorry, can you ask the question again? Senator DAVID POCOCK: How do we solve this issue of access to compute? We're kind of being told on the one hand that all we need to do is get the Anthropics and the OpenAIs here, with big data centres, and that that somehow solves the problem. But I fail to see how Aussies get access to that compute if there's a compute shortage. Mr Taylor-Price : I think it's better, as I said before, to have it here than not here. However, it depends on how it's done here. You really want that economic value to be here. One thing that I'd caution on and that I've seen discussion on is to say, 'Hey, why don't we just get these large multinationals to give Australians free compute?' I'd give the analogy that, if you said, in order to get flight paths in Australia, foreign companies need to give free flights to Australians, your domestic airline industry would collapse overnight. So it's very nuanced policy positions that you need. You don't want multinationals subsidising Australia; you want open and equitable market positions where Australian companies can thrive alongside that compute. There is a shortage of data centre capacity globally. That is a real issue. But the amount that's in Australia dwarfs the amount that Australia needs. So I don't know what the exact answer to that is, but we do need to find a way that Australian companies get compute but without distorting the market such that we get unintended consequences. Senator DAVID POCOCK: I might put a few other questions on notice. CHAIR: Senator Dolega. Senator DOLEGA: Thanks for the great evidence today. I thought I was a geek, but hearing you guys today has put me back in my place! I want to go back to some earlier evidence about the data centres. You mentioned being able to tell the difference between who is providing high-quality data centres and who isn't. How do you make that determination? Mr Taylor-Price : Obviously we have many conversations. There are procurement processes and due diligence questions and things like that. There is also a track record. There's also a fairly decent amount of regulation in the space. You've got the Hosting Certification Framework. I do have some issues with that. There's a lack of transparency, where the government may know an issue with a particular data centre and then when we procure from that the issue is not disclosed. I think there are some issues around that. There's also ownership and retained economic benefit. Obviously we prefer to use Australian companies where we can within that supply chain. There's plenty of information in the media about people who have issues with particular data centres and the way they conduct themselves. There's no magic that we've got access to that you can't see as a public consumer. Mr Navfield : And I'd just mention a commitment to security in Australia—things like IRAP. Seeing a provider of a data centre or products or software commit to the security process early in their journey of being in Australia is another signal that shows that they're committed to the right parts of the Australian economy. Mr Senti : It's also the construction of the data centre—having worked with a data centre provider, how they actually do their power, their cooling and the management of all that. Is their water consumption something that they're bringing in every day, or do they have a closed-loop water system? Are they using immersive cooling versus air cooling? There are a whole heap of different things you can— Mr Taylor-Price : The temperatures of the cold loops and things like that. There's a ton of technical detail that makes a better data centre provider. Senator DOLEGA: And, if I may, who, in your opinion, is providing high-quality data centres in Australia at the moment? I ask this as a Tasmanian senator. We've got quite a bit of interest, because, from what I understand, having a cooler climate, it is a really good place for data centres to be built. However, I don't think their work has been done so far for the social licence thing. Who's doing a good job so far? Mr Taylor-Price : If possible, I'd like to decline to answer, because these are my key stakeholders that I work with! Mr Navfield : I'll start with a quip—that I believe Tasmania is better for cooling than space. Mr Senti : I think there's a significant opportunity for Tasmania because of the climate and because of where you're geographically located. I've always said to the data centre providers that I've worked with and been involved in that it's a great destination for data centres. Social licence is going to be a problem. You've got a lot of renewable energy there. At the same time, if you can build closed-loop systems that don't consume water or that minimise water or you have immersive cooling—and for those who don't understand what immersive cooling is, think about putting server infrastructure in baby oil; it's a similar chemical compound, non-conductive—basically the thermodynamics of that cooling is significantly better because it's in a fluid, not in air. Air eats up a lot quicker than a fluid does, so you don't have to cool it as much—those sorts of things. Mr Taylor-Price : I thought of a non-controversial answer. There's a strong correlation between the providers listed on the Hosting Certification Framework and the high-quality providers. Senator DOLEGA: Thank you. I do appreciate that. CHAIR: That brings us to the end of this session. Thank you very much to all of you for giving up your time to give your evidence today. Again, if there's additional information that you've been asked for or are keen to provide, please do so through the secretariat. Proceedings suspended from 10:59 to 11:10 .