Edge AI for Factories: Australia's Vetra Explained

Samantha: Welcome to The State of Tech, The European Edition, Monday May eighteen, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: Australia unveils a new kind of AI infrastructure that runs right next to the robots, the United Nations warns that AI power is dangerously concentrated, Washington commits fifty-four billion dollars to military AI, the Vatican sets up its own AI commission, Nvidia releases free tools to help quantum computers actually work, and a humanoid robot is being prepped to climb Mount Everest. Let's get into it.

Australia's CSIRO launches Vetra, a new edge AI system that lets robots learn in real time without the cloud.

Samantha: We start in Australia, but with implications that land squarely in European factories and warehouses. CSIRO, Australia's national science agency, has unveiled a new edge AI infrastructure called Vetra. The idea is simple but significant. Instead of sending data from robots and sensors all the way to a distant data centre and waiting for an answer, Vetra puts the heavy AI processing right next to the machines themselves. That means robots can learn and adapt in real time, on the factory floor, in a mine, or out in a field.

Bob: And that real-time piece is the whole point. When a robot arm needs to react to a moving object, or an autonomous vehicle has to decide in a split second, a round trip to the cloud is too slow. Vetra cuts that delay, and it also keeps sensitive operational data local, which matters for security and for energy use. CSIRO is positioning this as a building block other countries and companies can adopt.

Samantha: For European manufacturers, this lands at exactly the right moment. There's a big push across Germany, Italy, and the Netherlands to bring robotics deeper into production lines, partly to offset labour shortages. And there's growing nervousness about shipping factory data through American or Chinese cloud platforms.

Bob: That's the European hook. Edge AI fits neatly with the EU's data sovereignty agenda. If a Volkswagen plant or a Dutch greenhouse operator can run advanced AI on site, they sidestep a lot of regulatory headaches under the AI Act and GDPR. And it lowers the electricity bill, because you're not constantly feeding a data centre somewhere else.

Samantha: There's also a competitive angle. European robotics firms like ABB and KUKA have been investing heavily in smarter machines. A robust open edge AI stack gives them an alternative to building everything in-house or leaning on the hyperscalers.

Bob: The honest caveat is that edge AI is still maturing. Running serious models on local hardware is expensive, and you need engineers who can maintain it. So this won't replace cloud AI overnight. But it does mark a shift. The centre of gravity in AI is starting to move outward, away from giant data centres, toward the places where the actual work happens. And for Europe, which has the industry but not the hyperscalers, that's genuinely good news.

UN digital envoy warns AI power is concentrated in a handful of zip codes and threatens climate goals.

Bob: Next up, a sharp warning from the United Nations. Amandeep Singh Gill, the UN's special envoy for digital and emerging technologies, said at a conference in Seattle that AI wealth and power are now concentrated in just a few geographic zip codes. His words. Low-income countries are essentially locked out of shaping how this technology develops. And on top of that, he flagged AI's exploding energy demand as a serious threat to global climate targets.

Samantha: The UN's response is to set up a new scientific panel on AI governance, modelled on the climate panel, the IPCC. The idea is to bring in voices from across the world, not just from Silicon Valley and a few labs in California and China.

Bob: Gill's framing is striking. He's not talking about countries, he's talking about postcodes. Most frontier AI work happens in a stretch from San Francisco to Seattle, with outposts in Beijing, Shenzhen, and a few European cities. Everywhere else is on the receiving end of decisions made by a handful of companies.

Samantha: For Europe, this cuts two ways. On one hand, Brussels has been the most aggressive regulator in the world with the AI Act. On the other hand, when you look at where the actual frontier models are built, Europe is not in that top tier either. So European policymakers are caught between wanting to set the rules and not having the home-grown champions to back it up.

Bob: The climate point is also one Europe takes seriously. Ireland and the Netherlands are already pushing back on new data centre permits because of grid pressure. If AI energy demand keeps climbing the way it has, that tension only gets worse.

Samantha: A UN scientific panel won't have enforcement power, but it can shift the conversation. The IPCC didn't stop climate change, but it made denial much harder. A similar body for AI could at least force governments to confront the inequality and the energy footprint with shared data.

Bob: The risk is that it becomes another talking shop while the companies keep building. Whether this panel has real teeth depends entirely on who sits on it and whether the big AI players cooperate.

Samantha: Quick interruption. If you listen to The State of Tech regularly, hit that like button and subscribe, that way you'll never miss an episode. Okay, moving on.

The United States earmarks fifty-four billion dollars for military AI and autonomous weapons.

Bob: Washington is preparing to spend roughly fifty-four billion dollars on autonomous weapons and military AI. That's the headline from the AI Expo in Washington this past week. The Pentagon already has active contracts with OpenAI, Google, Microsoft, Nvidia, Amazon, Oracle, and SpaceX for everything from data analysis to AI-driven targeting.

Samantha: Fifty-four billion is not a small number, even by Pentagon standards. And the framing is explicit. This is about the race with China. The US sees AI as the decisive military technology of the next decade, the way nuclear weapons defined the last century.

Bob: What's notable is which companies are on that contractor list. These are the same firms that build the chatbots and cloud services Europeans use every day. OpenAI, just a few years ago, had a clear policy against military work. That's gone. The line between consumer AI and defence AI has basically disappeared.

Samantha: For Europe, this puts NATO members under real pressure. If the US is spending fifty-four billion on autonomous weapons, allies are expected to keep up, at least proportionally. France, Germany, and the UK already have national AI defence programmes, but nothing close to that scale.

Bob: There's also a values clash. The European Parliament has repeatedly pushed for meaningful human control over lethal autonomous systems. The US approach is far more permissive. So European defence ministries are stuck trying to match American capability while their own legislators want stricter rules.

Samantha: And there's the supply chain question. If a European army wants this kind of AI, they're likely buying it from American firms, which means dependency on US export rules and software updates.

Bob: The deeper worry is normalisation. When the biggest AI labs in the world are deeply embedded in weapons programmes, that shapes what gets built and what gets prioritised. Safety research, alignment, transparency, all of that competes with military deadlines. And military deadlines tend to win.

The Vatican sets up a new commission on artificial intelligence under Pope Leo XIV.

Samantha: From the Pentagon to the Vatican. Pope Leo XIV has approved a new Vatican commission dedicated entirely to artificial intelligence. The commission will coordinate the Holy See's response to AI, looking at human dignity, economic effects, and how the Church itself uses the technology internally.

Bob: And this is a prelude to something bigger. Pope Leo's first encyclical, his first major teaching document, is expected to focus heavily on AI. He's framing it as the defining moral question of this era. That's a significant statement from an institution with a billion-plus members.

Samantha: The Vatican has been more engaged with tech than people often realise. There was already the Rome Call for AI Ethics a few years back, signed by IBM, Microsoft, and others. This new commission formalises that work and gives it real institutional weight.

Bob: For Europe specifically, the Catholic Church still has cultural and political influence in countries like Italy, Poland, Ireland, and Spain. When the Vatican speaks on ethics, it shapes public debate, even among non-believers. And it can give cover to politicians who want to push for stricter AI rules.

Samantha: There's also the workplace angle. The Pope has been vocal about AI's impact on jobs, particularly lower-wage work. Expect the encyclical to push hard on dignity at work, fair wages, and the responsibility of companies that automate.

Bob: What's interesting is the contrast with what we just discussed. On one side, fifty-four billion dollars for autonomous weapons. On the other, a global religious institution framing AI as a moral question. Both are trying to set the terms for how this technology develops, just from very different starting points.

Samantha: And both will land on European desks. EU regulators, national governments, and corporate ethics boards all have to navigate between the commercial pressure to deploy AI fast and the ethical pressure to slow down and think. The Vatican adds weight to the slow-down camp.

Bob: Whether tech CEOs actually listen is another matter. But the moral framing is now firmly in play, and that changes the conversation.

Nvidia releases free open-source AI tools called Ising to help quantum computers actually work.

Samantha: To close out, two things you can actually try or look up today. First, Nvidia has just released a family of open-source AI models called Ising. And this one is genuinely useful if you're in research, education, or even just curious about quantum computing.

Bob: The problem Ising tackles is one of the most stubborn in quantum. Quantum bits, qubits, are incredibly fragile. They need constant calibration, and they make a lot of errors. Traditionally, tuning a quantum processor takes days of painstaking work by specialists. Nvidia's Ising calibration model automates that and brings it down to hours.

Samantha: And the error correction piece is the other half. Nvidia says their decoding models are up to two and a half times faster and three times more accurate than the standard methods. For anyone working on quantum, that's a meaningful jump.

Bob: The reason this counts as something you can try today is that it's open source. The models are freely available. If you're a student, a researcher at a European university, or working at one of the quantum startups in France, Germany, or the Netherlands, you can download these tools and start experimenting immediately.

Samantha: Europe has a strong quantum scene. Pasqal in France, IQM in Finland, Delft in the Netherlands. All of them are wrestling with exactly these calibration and error problems. Free tools from Nvidia that genuinely accelerate the work is a real gift.

Bob: The catch, and there's always one, is that you do need Nvidia GPUs to run these models efficiently. So it's not entirely free. But for institutions that already have that hardware, this is a serious step forward.

Samantha: Even if you're not building quantum systems yourself, it's worth knowing this exists. Quantum is one of those technologies that sits just over the horizon, and tools like Ising are part of what brings it closer to actually being useful.

Bob: Check Nvidia's developer site if you want to take a look. The documentation is reasonably accessible.

A humanoid robot called PEMBA is being prepared to climb Mount Everest, prompting Nepal to draft new laws.

Bob: And finally, something a bit more out there. A Chinese-built, US-backed humanoid robot is being prepared to climb Mount Everest. The project is called PEMBA, using a robot from Unitree, with software help from Google DeepMind on the locomotion side.

Samantha: No bipedal robot has ever attempted Everest. We're talking mixed rock, ice, moraine, thin air, extreme cold. It's a brutal test bed for any walking machine, and that's exactly the point. If you can teach a robot to handle Everest, you can teach it to handle pretty much anything outdoors.

Bob: DeepMind's contribution is reinforcement learning for unstable terrain. Essentially, the robot learns by trying, falling, and adjusting, millions of times in simulation, before it ever sets foot on a glacier. The Everest attempt is the real-world graduation exam.

Samantha: What's just as interesting is the policy reaction. Nepal is now openly discussing whether it needs new laws for non-human expeditions. Who's responsible if the robot gets stuck and needs rescuing? Who pays the permit fees? Does a robot count as a climber?

Bob: Those questions sound silly until you realise they'll matter for European national parks too. Imagine autonomous drones or robots in the Alps, in the Scottish Highlands, in Scandinavian wilderness. The rules we write now for Everest will set precedents.

Samantha: And for the research community, the data coming out of this expedition will be valuable. How robots handle cold, altitude, unpredictable surfaces. European robotics labs working on search and rescue, agriculture, or disaster response will be watching closely.

Bob: There's also the optics. A Chinese robot, American funding, British AI lab, climbing a Nepalese mountain. That's the global tech industry in one image.

Samantha: Today we covered: Australia's new edge AI infrastructure Vetra, the UN's warning on AI concentration and energy, the fifty-four billion dollar US push into military AI, the Vatican's new AI commission, Nvidia's open-source quantum tools, and a humanoid robot's Everest attempt.

Bob: Want to know more or react? Visit stateoftech.eu or email us at info@doorzetters.net.

Bob: State of Tech, the tech world in 15 minutes.