Why I quit my career in politics to analyse AI. The story might surprise you.
I was a union official, then an adviser to two different members of parliament, and until recently a State Organiser for Victorian Labor working on the 2026 Victorian state election.
Now I analyse the biggest build Australia has attempted in a generation: AI infrastructure.
Uni wasn't for me. I left VU without a degree after attempting to study Law. Then I pivoted to Communications at RMIT and dropped out, again. I found the university sector in Australia bitterly disappointing. Most of the classes were taught by people without real industry experience, and what experience they did have was years out of date.
Pair that with the overwhelming amount of disengaged international students and endless moralising from the universities, I realised I wasn't getting an education. Instead, I was being strung along for someone else's profit. I made the choice to accept an offer to work full time in politics. What I learned there was more insightful than anything I could have learned in class.
My four years as a union official taught me the things the data centre industry is learning the hard way: you don't get social licence by announcing it. You earn it slowly, one conversation at a time, with people who can say no and ask questions.
As a political staffer I saw how policy really gets made. It was all about building relationships between industry, community and government.
I also felt first hand the frustrations of a political system that was slow to adapt, in many ways woefully out of touch with broader society, and in a constant struggling with the perverse incentive of self-interest and remaining in power. This is not a shot at political parties, but at our broader system. I met many incredibly dedicated and good-hearted people in my time in politics. The system, more often than not, let them down.
As a State Organiser for Victorian Labor I ran a portfolio of electorates. I worked with candidates, volunteers and communities, had to consider the broader electorate's attitudes on every issue, and pored over endless polls and focus group outcomes.
A campaign is about building trust across the populous and a belief that, as a government, you can achieve your goals and that those goals are right and just. It requires defining and agreeing upon the licence to operate, and developing a binding social contract. In many ways, this trust and our broader social contract are under more pressure now than we have seen and felt in modern history.
Enter the AI technology explosion and the data centre investment boom that followed: a bonfire set to ignite the dry kindling of our polarised world and stagnating economies.
I wanted to write about and analyse AI and digital infrastructure because it's where many of the previously unconnected fault lines of our society have seemingly connected. The economy, politics and increasingly dissatisfied communities are all staring at this oncoming freight train of investment, disruption and uncertainty. This is a reckoning for our society, and it has really only just begun.
Once the first stage of this multi-trillion dollar build-out is completed, the scale of compute, and the impact the increased capacity of frontier AI models will have will be profound. These are subjects most people cannot conceptualise as they try to cling on to their dwindling economic circumstances and a sense of normality that is becoming harder and harder to maintain.
And too many of the people talking about it have never knocked on a door in the suburb they want to build in. I have.
Three things I believe.
One: the industry's problem isn't NIMBYs. It tried to skip the community part. This week the US voted 44–1 in a comprehensive and democratic message to the industry. You still live in a democracy, no matter how good your idea is.
Australia doesn't have to make the same mistake, but it is at serious risk of losing out on this investment if the right deal isn't struck. You only need to look at NEXTDC's West Footscray campus, where paused grid works took about A$140 million out of next year's guidance, to see what an upset community can do to progress and profit.
I will be closely watching the Office of AI, and I hope to continue my conversations with some of the big players in this industry and share them with you, the public.
Two: the people and communities who fight data centres aren't always anti-progress. It is not simply a left-wing reactionary position, or a right-wing conspiracy backlash. Many of the questions these communities ask are questions I used to ask. Who pays? Who benefits? Who was asked? Do I really win from this arrangement?
Three: Australia can and should benefit greatly from this industry, both by hosting the models and by using them. We know that national rules are coming, but we need a plan to deal with the grid demand, possible skill shortages and community backlash that could stand in the way of the economic boost this country desperately needs. That is the gap I work in.
I will be interviewing the biggest players in the AI and digital infrastructure space on https://t.co/tYZDtpmbKw, and I hope to have as many conversations about our new world as I can. Follow along if you want the inside scoop and a new perspective on AI, politics, infrastructure and the economy.
Views my own.
@matt_barrie@DrewPavlou Doesn’t this reduce government intervention by lowering red tape regulation on super funds? Not trying to be snarky but this seems to be a small step towards fixing the capital allocation of super funds.
Right now funds are really restricted from investing in growth industries.
It’s a Bigger deal for anthropic in their fight against openAI than other services. They are building the a platform you don’t have to leave. This I think is in part to fight against model swapping. It’s very easy to use another companies model, if Claude desktop has docs, meetings etc it creates more resistance.
Next stop will be to replace slack. They have the models to build this stuff in an afternoon.
I have a major result to share.
I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China.
This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model.
Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves.
I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems.
I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used.
It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20.
These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs.
I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at https://t.co/YOtzND0Jqk.
There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
@TheXato1@Evinc0@DrewPavlou I don’t work for Labor - it’s pretty all spelled out in my bio post that I don’t. I can’t spell you can’t read. We can help each other.
@BarbaraPocock I think this is inevitable for big companies. When times are tougher, AI gets seen as a way to take serious cost out. Qantas isn't the first and won't be the last.
@Evinc0@DrewPavlou Dario Amodei is the most influential AI leader in the world, and he is in support of strict regulations on safety for AI because he understands the potential risk of allignment.
Are you suggesting you know better?