Australia's AI gamble: Inside Albanese's plan to bring big tech to heel

[Music] The Australian Financial Review Big news out of the World AI Conference in Shanghai. Kini K3, this is the new model created by the Chinese startup Wunshort AI. Every week, there is big news when it comes to developments in AI. Things are moving fast, and governments and businesses are playing catch-up. The company says it all performs all rivals except for anthropics, fatal five, and open AI's GPT 5.6. Corporate leaders like Palantir Technology CEO Alex Karp are starting to speak out about the unbridled power of the AI giants. Something is going completely wrong. The basic view among enterprises in this country is I'm going to get no value, and they're going to get my IP. And governments are cracking down on the development of massive data centers to power the AI models. That is why they all be signing the nation's first ever statewide moratorium on hyper-scale data centers. In Australia, the federal government finally laid out its plan for AI regulation. This is about Australia shaping the future rather than letting the future shape us. Now it's about getting the balance right. This is the biggest investment boom since the mining boom, and yes, Elbow is right that we do need it to last, but we can't afford to miss this boom. And there is a danger that we raise the hurdles too high for the AI companies to come and invest here. Welcome to the Fin. I'm Lisa Murray. This week, Sean to clear columnist James Thompson, and political correspondent Ron Meisen. On why CEOs are losing patience with the AI giants, what's driving the government's policy shift, and whether Anthony Albanese is overplaying his hand. It's Thursday, July 23. Hi Ron, hi James, thanks for coming back on the podcast. Thanks for having us Lisa. Good to be back Lisa. We're going to get to Anthony Albanese speech and what it means for Australia's approach to AI. But it wasn't just our government that appeared to receive the AI wake up call last week. Some powerful voices in business also raised the alarm. James, you wrote about this in one of your columns. What are CEOs getting so worked up about? Well, I think there's a few things that different CEOs are getting worked up about. One of the big problems for CEOs is that it's hard to see where all the work they're doing in AI is really paying off. A lot of individuals are feeling more productive thanks to AI. They can organize their life better and organize what they do at work a lot better. But we're not seeing a whole lot of benefit across companies and certainly not across sectors yet. Now, perhaps that's coming. Maybe it isn't. At the moment, what companies are getting relatively minor productivity benefits, but increasing AI bills. The big AI model makers, OpenAI and Anthropic, have really jacked up the prices of the AI tokens that run their models. Yeah. And we're starting to see companies having to come to terms with significantly higher bills than they were at the start of this AI revolution when, look, essentially, AI was basically free. So that's one lot of tension in the AI sector. The other is the tension between the model builders, Anthropic and OpenAI and the big tech companies like Microsoft, Amazon, Alphabet and Palantir, which is another big tech company that does a lot of work in the defense space. What they are basically warning is that companies risk paying these AI businesses twice. First, the companies are paying for high token costs. But then there's this suggestion, particularly from Sachin Adela, who wrote a little essay on this, is that they're paying a second time because companies are having to essentially hand over their intellectual property to the AI model makers. They get the best bang for buck out of their AI usage. The quote from Sachin Adela was, "In consuming intelligence, you're creating intelligence and what you create should be long to you." Hmm. The guy who runs Palantir, a guy called Alex Carp, has pointed out, "I'm going to get no value and they're going to get my IP." The big risk is that you take an AI model like Claude and all of a sudden, Anthropic comes out with Claude for Lawyers, Claude for Scientists, Claude for Accountants. And what companies are essentially seeing is that the AI models they're using are suddenly becoming competitors. This is the voice of American business that has been channeled through me. Carp went on this pretty amazing 20-minute rant about this on CNBC a few weeks ago. And I'm telling you, it is absolutely a problem for this country. This is an issue that Australian companies are thinking about. Who owns all the AI data? And most importantly, who owns the right to get the biggest value out of the AI data? And so, Ron, as companies are starting to question their engagement with the big AI giants, Anthony Albanese finally laid out his vision for the government policy on AI. People were waiting for his speech last week. How was it received? I think it was received fairly well across the board. It's not just companies that have been grappling with these questions, every aspect of government, every government minister, every department, every department head. They are all grappling with these exact same questions. Everyone expects that they will be protected at the same time as everyone wants to take advantage of these new AI models. It is attention at the heart of everything here. But it was an important speech in the sense that it was the first time that Anthony Albanese really himself went out and articulated a vision for AI for this country. This is our time to decide what AI looks like here in Australia. At the same time, it was very much a political speech in that he recognised that there is a lot of angst and anxiety and anger in some quarters about what we're seeing in AI and obviously in the AI infrastructure as well around data centres. To make this technology work for us, for workers and communities. And so there is absolutely an element of trying to ensure that he's covering off on the politics of this. Four-hour economy and our environment. Then we saw that about six months ago, the government outlined a set of data centre principles. These were optional. Basically, the government said if you comply with these, if you bring as much electricity to the grid as you're consuming, if you comply with water standards, if you adopt good employment standards and bring in apprentices, we'll prioritise your approvals processes. And if you don't hit those, you'll get pushed to the back of the queue. But what Anthony Albanese outlined in that speech last week was more strict. It was our requirement now that these companies comply with strict water standards that they bring as much energy to the grid. And the other thing he covered off on in that speech was a clear message about what the government intends to do on copyright. Australian writers, musicians, artists and journalists must retain ownership and control of their work. You know, making very clear that rights holders be protected, but also that the current regime is not necessarily up to scratch. Our laws will spell that out plain as day. But there's still six to 12 months until we see it entered into the parliament, and there's still a lot of details to come about what will actually be comprised in that legislation. A year ago, the government was taking a very hands-off approach to AI, so why the change? The government doesn't call it a U-turn. They like to call it an evolution in policy thinking, but it's, you know, it's pretty clearly a pivot in what their position was. I think there's a few elements to why. The first is Anthony Albanese is emboldened by the social media age ban. 'Cantige was seeking us out to learn about the approach that we were taking.' He clearly was very happy with how that has gone down around the world. Other governments are now adopting it. 'And it is proof of what a strike can do when we back ourselves.' There is a national security element to this. We saw earlier this year, the US government's stoppy export of anthropics, mythos and fable models. Their most powerful artificial intelligence models specifically around coding and cyber security. And that was a wake-up call as well, I think, because if you were relying on those models to achieve particular outcomes, if it was running something, it would have stopped. 'If we are always dependent on someone else, somewhere else, we will always be vulnerable.' And I think that was a wake-up call as well, to say, well, if we don't have the ability to control these models here in Australia, then we're really exposing ourselves. How did you think the speech was received, JT? Well, I think it was interesting that within, you know, less than two hours after the speech, both the National Secretary of the Australian Council of Trade Union, Sally McManus, and the Chief Executive of the Commonwealth Bank, Matt Common. Both came out and said, 'Good speech. Great to see the Prime Minister taking control of that.' I think that's what a lot of people, including myself, have been saying that, you know, there's a lot of risk in the AI world where is the government on these issues. And now we've got the Prime Minister intervening. I think that's has to be welcomed. It's good that the Office of AI is, you know, essentially down the corridor from the Prime Minister's own office. I guess the one sort of niggling concern I've got, and I think this is the concern of the business community, is the question of whether Albanese is coming to this, as Ron said, from a place of of politics and a place of managing risks, and he's a politician in first and foremost. So, Albanese can smell the risks building around data centers and he can smell the risks building around copyright. So, sort of come back to first principles. Why did he come out and make that speech now? Is it because he's a huge believer in the power of AI and what it's gonna do for the economy? Or is it because he sees points to be scored and some risks to him in the labor party if AI is left unconstrained on issues like data center built out and politics? I think it's probably the latter. So, let's get into some of that detail around what they are proposing around data centers. Ron, in terms of what Albanese is saying about some of the rules he wants to legislate, what will that mean for the investment boom underway? Well, anything that has been built already won't be affected by this. Now, this is the data centers that run your Apple iCloud account. Your Netflix account that sync your messages between devices. Very few of these data centers will be affected by these new laws. It's very clear from what the government announced that they are targeting those big hyperscalers, the big AI labs, Microsoft, Google, OpenAI and Thropic, and making sure that those massive data centers that are going to be training AI models that are going to be doing inference processing that they are covered by these new laws. There is still some ambiguity around who will be covered and those details will need to be worked out, but there is going to be a clear distinction in who will be covered. The same with the water efficiency requirements. There's a lot of anxiety about data centers taking water away. There was some data from Sydney water suggesting that in coming years, water consumption for data centers could go from 2% to 25% of total consumption, which is a really massive uptick. And there's a lot of angst about that. But the other elements of this as well is the ability to actually build the energy production that these data centers will need in time. You know, I was only talking to a major data center operator a bit before this podcast was recorded. And one of the concerns is that you can get a data center approved and built much faster than you can get the relevant associated energy approved and built. And so there is a disconnect there and there's some concern about how that will flow out in the coming years. And all of this is responding to the community angst about data centers isn't it JT? And obviously there's been a lot of cases overseas where there's been a huge backlash. Oh, absolutely. I mean, America is grounds year-o for this. Public health is not for sale. We've seen this almost guerrilla activist campaign start to build in some American communities. That is why they all be citing the nation's first ever statewide authoritarian on hyper scale data centers. Last week we had the governor of New York State announced they're going to put a moratorium on new hyperscale data center developers. Texas is at a crossroads. Texas has done the same thing in rural areas. We are here to push back against the wealthiest corporations in the world, capitalizing on our resources and ignoring the will of the people. So both sort of democratic and Republican red and blue states are rolling out these moratoriums and pauses. And this is the place in the world where AI is really driving the economy. So you can feel that angst starting to build in Australia we've seen protests in Melbourne, in Sydney, in Lankov, even in rural New South Wales. So I think wherever these data centers go, they are going to be chased by this community angst. Some of it well informed, some of it not so well informed, but it's there. And clearly, Albert Easey has seen fit to respond to it. And he is responding to it, which might affect the pipeline of projects. What is that pipeline in Australia? Well, it's hard to pin down Lisa, but I think it's massive whatever way you cut it. UBS says that the value of data centers under construction in Australia is already 21.6 billion. They reckon it could double towards a peak in the next 18 months that would be 2% of gross domestic product. OK. That's just one estimate. Westpac's got another estimate. They reckon the investment pipeline for data centers will eventually exceed $155 billion. Throw in the money we're chucking at the energy transition, which is something like $200 billion. As Ron said, the energy transition in the data center build out go hand in hand. That's an awful lot of money. And we are seeing the big guys come here. Microsoft's promise to invest $25 billion here. The figure's pretty rubbery, but there's some big commitment. Open AI. It's got a local office. They've got a $7 billion plan for Australian AI. Meta platforms that they're basically underwriting the growth of the local data center hopeful firmness, just got huge plans and is apparently heading towards an ASX listing in the tens of billions of dollars. And then we've got an anthropic, which keeps sending out these smoke signals that it wants to invest tens of billions of dollars here too. Yeah. And anthropic is promising to commit those billions of dollars, but it's contingent on where Australia lands on copyright. Ron, can you explain why that is such a big issue for the AI giants when they're looking at Australia and how the urban easy government might address it? Yeah, sure. So obviously when an AI model has to learn, it has to be trained. It has to be trained on some kind of data. Now, historically, a lot of these models have been built on data that has been illegally obtained or has been obtained using very lax rules in the United States. For example, a lot of models were built on illegally obtained books, internet information is just scraped from the internet. Now, in the US, some of that is fine. Some of it is still working through the courts. The AI giants are facing dozens and dozens of lawsuits now that will be tested in the coming years. In Australia, it's far more strict. Basically, it comes down to, if you want to train a model here, you've got to pay for all the content that you're using in that model. Now, in some senses, the AI giants, the tech giants, would prefer a text and data mining exemption, which would basically mean that you don't have to pay. But the government ruled that out. It's not going to happen. The question is how that system works. At the moment, you can do deals with large rights holders. We've seen deals with immediate companies. You can do deals with organizations that represent musicians, organizations that represent authors. But what we know from documents obtained under Friedman Information, when anthropic visited Australia earlier this year and met with Treasurer Jim Charmers, is it the big barrier for them is what they call the long tail of rights holders. Now, that's the thousands of individuals that might have a substack page or website a couple of books that have been out of publication for a while. And how that information can be absorbed into their training models in a way where they're not going to be breaking the law. I think the government is really having to walk a line here. They don't want to catch the eye of the creative sector and the media companies. So 80% of what they say in this space is, will protect you, it'll be OK. And that's always been the case. The existing system kind of does that. What will be interesting to know is how they augment the copyright system to cover that so-called long tail. And how they make it easier for AI companies to obtain that data. Anthropon easy, still unclear. The Attorney General is working on a solution. But he did put down some, I guess, markers. One of those is that if you don't want your content used in AI training, you won't have to have it. So that's an interesting one because it means it either has to be an opt-in model or an opt-out model created. And that gets really complicated. You can see that Albo is taking a really hard line on the copyright issue. He's really put a stake in the ground here. And this is one where he says he's not going to budge. But I think the big risk in doing this is that Albanese has given the impression that Australia's got a really strong hand here. That the AI world needs us more than Australia needs the AI world. I'm just not sure that that's correct. This is the biggest investment boom since the mining boom. And yes, Albo is right that we do need it to last. We can't just have a burst of CapEx and that's it. But by the same token, we can't afford to miss this boom. [MUSIC PLAYING] We're talking about the Australian government's pivot in the way it deals with AI policy. And what that means for businesses, the economy, consumers and voters. James, do you think the biggest risk is that Anthony Albanese overplays his hand? Because lots of other countries are also competing for this investment. Albanese does seem to be coming into this thinking. We've got a really strong economic hand. But it's hard to know what that's really based on. I mean, the economy is slowing. It has to slow so we could get inflation under control. Productivity growth has been terrible for 10 years. And I'm not sure we're in a great position to say, OK, this is an investment boom we can afford to miss. We're happy if the money goes elsewhere as long as our copyright's protected. And we've managed the data center built out in the right way. I think it's right that Albanese wants to capture sustainable benefits from the AI boom. But I just worry Australia isn't a bit of a beggar's can't be chosen as positioned here. And this is not a boat. we can afford to miss. And there's a lot of rhetoric here and not necessarily much money behind it, at least from a government perspective. Yeah. You know, just days before Anthony Albanese gave his speech, "We will reinforce Canadian sovereignty, so Canadians can make their own choices on how AI is built, governed, and used." We saw the Canadian Prime Minister Mark Carnegie in Canada outline a really big AI agenda. We need a strategy that not only addresses these challenges, but also builds on ways in which Canadian businesses are using AI. Made it clear that the Canadian government had a few priorities. One was putting money behind AI, but the other was also trying to get Canadians to embrace it. And a lot of countries are trying to embrace AI because they're aware that this is a way to really reverse what has been a productivity slowdown, not just in Australia, but around the world. And the countries have a similar size to Australia, South Korea, for example, has put almost a trillion dollars behind making itself an AI leader. It wants to be a top five AI country in the world. And when you think about how that compares to Australia, I mean, we're not putting anywhere near that amount in there. But at the same time, I think the speech did give some clarity on where we want to head. I mean, Air Trunk Chief Executive Rob and Kudor said that the standards on data centers would provide greater clarity. He made the point that private capital is available to invest, but the supporting infrastructure has to keep pace. Otherwise, this isn't going to work. And Ron, as you mentioned, one of the criticisms of the Albanese government is that it hasn't announced any additional spending on Australia developing its own AI. There's a concern that Australia might be too reliant on the US AI giants rather than developing its own technology. On last Friday, we saw the release of another high-powered AI model from China. How much a national security concerns driving the government's AI pivot? It is absolutely one of the main concerns. I wouldn't say it is the only concern, but it is very much a serious concern. And as we touched on earlier, you only have to look at when the Trump administration imposed a block on the export of Fable 5 and Mythos and Thropics, most powerful AI models. It's a concern that we had been incorporating into our national security establishment and all of a sudden the US government turned them off. I mean, that's a massive risk to Australia. That's not just if you're incorporating it into the national security establishment. That's if you're incorporating it into critical infrastructure. What if a bank had been relying on it? It got turned off at 6pm on a Friday in Australia. And if you had everyone go home and all of a sudden discover on Saturday morning that all of these transactions hadn't been processed or critical things hadn't been working. That's all part of this. Sovereignty is a really interesting concept. So there are three local Australian companies that are trying to develop large language models. One of them is based here in Canberra called Zana AI. It's been doing work with government for quite a while now. Then we've also got main code in Melbourne, which has been producing Matilda AI model. And another one called Sovereign Australian AI, which is developing its own large language model called Australia. They're not big name brands. You wouldn't probably have heard of them unless that you were using them. And the government is talking up the need to create more of this localized capacity. At the same time, you know, when it looks to use models or AI companies for its Guv AI, which is this new big project within government to adopt artificial intelligence across the government. It didn't look to an Australian company. It looked to open AI and its chat GPT. It looked to anthropoccalcord and it looked to the French mistrial. That's questionable in terms of our, we actually embracing sovereignty over convenience of these good models that already exist. The other way of looking at it is actually by attracting, say, training and inference processing to Australia's shores. We are bringing it within our legal jurisdiction. And so we are able to actually assert more control over it. And so there is a question over will actually how much do we want to invest in local sovereign capability versus simply taking control of some of that overseas capability and bringing it under our purview? I mean, I think the main contest here is probably going to be between China and the US about who can release the biggest and fastest AI model. Up until Friday, we thought that the US was well ahead, but all of a sudden we've got this new Chinese model released by a startup called Moonshot AI. Moonshots Kimi AI model took the tech world by storm just last week. The model is called Kimi K3 AI. It is almost on par with anthropics and open AI's best versions out there. And it says that it's almost as good as anthropic and open AI's best models. It is no longer a given that the US Frontier Labs have a safe lead in AI. Which is an absolute shock for the American AI establishment. I think they thought they had a 12 month gap over their Chinese rivals that the gaps probably more like three or four months. Based on what we know about Kimi K3 so far. You know what we're seeing here from China is that they can develop very good models with much less capital and possibly even inferior technology than the US are using. So you bring that back to the national security issues. I mean China could become an AI powerhouse militarily as well. But everybody's trying to do this. I mean, India, Singapore, South Korea, everybody's got AI ambitions. I just wonder if we end up with the AI models essentially being a commoditized product when nobody's got a really great edge. And so yeah, the government could invest billions and not get an amazing return out of it. Because we're all using cheap models built in China or India or Singapore. Eventually anyway. Ironically, in America, we are seeing the uptake of Chinese models explode. The Chinese model market shares around 30% and has got a bit higher. So I think we'll see as token cost increase and the cost of running these AI models go up. We'll see more Australians turn to the cheapest model. I mean, that's generally what we do in a globally competitive economy. So everything is risky in this. Nobody knows what's going to happen. It's such a Pandora's box of possibilities. Ron, the timing for the government to outline its vision for AI is very interesting. The Labour Party conference starts on Thursday. An AI policy has been a big focus particularly for the unions. How much will anxiety around job losses play into the government's AI policy? I think it's absolutely no coincidence that Anthony Albanese made this speech a week out from Labour's National Conference starting. If he hadn't taken a firm a hand with AI, I think it would have been a far more disruptive issue at conference. We saw the Department of Industry and Workplace Relations put out a report last week on AI job losses. Anthony Albanese cited this in his speech. It showed that in industries and jobs most exposed to AI, there had been a reduction in growth, about two percentage point reduction in the growth of that sector, but not necessarily a significant loss or a backwards step. Now, it's important to note that this data only goes through till February. So it did miss out on some of the large scale job losses that we saw announced earlier this year in the tech sector, particularly with Wise Tech and Adelacian. It also noted that in the later quarters of the analysis, there had been an increase. It started to increase the amount of disruption that they were seeing. So I think it's a watching brief. The union movement is highly sensitive to it. They also have a watching brief on this. I think the jobs issue remains top of my mind anyway, but it's just so hard to get an estimate of what's actually happening. I mean, as Ron said, the government says, look, nothing much to see here at the moment. UBS asked its analysts last week, how big could the job losses be? They said 10 to 15 percent across ASX listed companies. Tim Tui, their economist at Yarra Capital Management, he thinks you, you know, in a sort of baseline scenario, you could see a productivity booster, may I have 3.4 percent, but unemployment goes to 6 percent under that scenario. So there's puts and takes here who really knows. I was really interested last week. We saw the big US investment banks, JP Morgan, Jamie Dimon, CEO there, he said that in some jobs and some tasks, they've already cut 30 to 40 percent of their workers by deploying AI. Now, he said, don't worry, we've redeployed all those workers into other parts of the bank, so no one's actually lost their job. But this is a bit of a hidden risk here. A company like JP Morgan say, I'm picking a number out of a hat, say it employs 100,000 people today. It might still employ 100,000 people in five years time, but it might be five times bigger because of AI. And so what we're seeing is it's not just the jobs lost, it's the jobs not created, as AI comes in and starts to improve the productivity of individual companies. But again, is this going to happen tomorrow? Is it going to happen in one years time, two years time, three years time, or a decade? Nobody knows, and that's what makes this such a difficult topic for unions and the rest of us to get their heads around. A final question for both of you. We were all waiting for the Australian government to articulate its approach to AI. It's done that now. There's still a lot of detail to come, but we have a better idea of their thinking. Based on what we know, who do you both think are the winners and the losers? I think from the speech, there is something in it for everyone. The rights holders in Australia will be thrilled with it because it's clearly putting a line in the sand that they're going to be safe, and if they don't want their information used in AI, they don't have to. Equally, the AI companies are going to be happy with what they heard, because they're looking for certainty and looking for a model that works for them in being able to actually buy this. data easily in a way that they know they're not going to face lawsuits later on. I think Anthropic is a bit of a winner from this, mainly because they've been looking for certainty around copyright. And while they might not get everything they want, that certainly I think what they'll get at the end of the day. They've also been really talking the same language as the government around needing to protect jobs and needing to comply with Australian values. And certainly we saw earlier this year they signed up to that memorandum of understanding with the government which puts them in good stead of being a partner of choice. And you know, we certainly saw Claude taken up as one of the models in Gavaeai. Yeah, I think we're all winners in that it's good to see the Prime Minister of Australia putting this issue at the top of his list of priorities in a world where there's so much short-termism in politics. Now the devil, of course, is in the detail. I'm probably a little bit different to Ron. I reckon Anthropic might end up being a loser. The Prime Minister's taken a really hard line on rights. I think there's probably still a path to a deal as Ron says, but the path's probably a little bit narrower. It feels like that to me after this speech. Maybe there can be a deal. The question for a company like Anthropic where they do have options as to where they want to put these data centers and what countries they want to use for training. Is there another option where they'll do an easier copyright deal? And we just say, thanks for your interest Australia. We're moving over here with these guys. I do agree with Ron on the data center thing. These guys want certainty. I don't think they would have heard anything out of elbow speech last week that would have particularly scared them. They just want to get on with getting this stuff out of the ground. So look, I think elbows done a pretty good job of trying to get the balance right. This idea that there are risks to this, but we've got to lean into it and we've got to be flexible and we've got to make fast decisions. So that's the right setting. My concern is that in Australia we always default to safety. We always want to manage risk and this is not a time to be timid. We need to lean forward into growth opportunities because the economy is stuck and we need to start going somewhere. Thanks James. Thanks Ron. Thanks Lisa. Thanks for listening to the fin. If you want to read more about how governments and CEOs are responding to the rapid developments in AI, check out the show notes. The fin is produced by Mandy Coolan. Fiona Bafini is head of premium content. Alex Gow composed our theme. For more stories about markets, business and power, subscribe to the financial review at afr.com/subscribe. I'm Lisa Murray. See you next week. Bye.