6 juli 2026 · 13:33
UN AI Governance Talks Geneva: EU AI Act Implications
The United Nations opened its first Global Dialogue on AI Governance in Geneva today, with a scientific panel warning that AI capabilities are advancing faster than policy can keep up, and explicitly not ruling out catastrophic harm. For European professionals, the stakes are immediate: Brussels sees Geneva as a platform to export the EU AI Act's risk-based framework as a de facto global standard, and the panel's report hands European regulators the scientific backing they have been seeking. This episode also covers Samsung's imminent profit record on AI memory chips, Micron's massive new Japanese factory, Tesla's internal AI cost controls, and two practical open-source tools built for European users.
Beluister deze aflevering:
Transcript
Samantha: Welcome to The State of Tech, The European Edition, Monday July six, 2026. I'm Samantha Lawrence.
Bob: And I'm Bob Russell. Today: the United Nations opens its first global dialogue on AI governance in Geneva, Samsung is poised to overtake Nvidia as the world's most profitable tech firm, Micron breaks ground on a massive memory chip factory in Japan, Tesla puts a weekly spending cap on how much AI its own engineers can burn through, and then two things you can actually use today: the open-source photo library LibrePhotos and the offline transit app Transportr. Let's start in Geneva.
United Nations opens first global AI governance dialogue as scientists warn capabilities are outrunning oversight.
Samantha: The United Nations kicked off its inaugural Global Dialogue on AI Governance in Geneva today, and the framing is striking. Governments, companies, academics and civil society groups are sitting down together for two days, and they are starting from a warning. A preliminary report from the Independent International Scientific Panel on AI says the technology is advancing faster than science or policy can keep up with, and it explicitly does not rule out catastrophic harm.
Bob: That is unusually blunt language from a UN body. The dialogue itself is meant to build the plumbing for international cooperation, so shared standards on cybersecurity, on the digital divide, and on making sure the benefits of AI are not just concentrated in a handful of American and Chinese companies. It follows years of fragmented national rules, from the European Union's AI Act to voluntary commitments in Washington and Beijing, none of which really talk to each other.
Samantha: And European policymakers are among the most active voices in the room. Brussels has been trying to export the AI Act's risk-based approach as a de facto global standard, in the same way the privacy rules known as GDPR spread outward a decade ago. Whether that works at the UN level is another question, because both Washington and Beijing have very different ideas about what governance should even mean.
Bob: The tension is obvious. The scientific panel is warning about existential risk, while national delegations are also protecting their own AI industries. For European users and companies the bottom line is this: even if Geneva does not produce a binding treaty this week, it starts to lock in the language and the benchmarks that future rules will be measured against. Expect the outcome document to feed straight back into how the AI Act gets enforced from next year. And notably, the panel's report gives European regulators exactly the kind of scientific backing they have been asking for to justify a stricter hand.
Samsung set to overtake Nvidia as the world's most profitable tech company on AI memory demand.
Samantha: Over to Seoul, where Samsung Electronics is on the verge of a symbolic milestone. Industry forecasts suggest the company's second-quarter operating profit, due out tomorrow, could hit around 85 trillion won, roughly 55 and a half billion dollars. If that number holds, Samsung passes Nvidia as the world's most profitable tech company by quarterly operating profit. And the driver is not phones or televisions, it is memory chips.
Bob: Specifically the high-bandwidth memory that sits next to AI processors inside data centres. Every time a company like Meta, Microsoft or a European cloud provider orders more AI servers, they are also ordering a lot of these specialised memory chips. Samsung, along with SK Hynix and Micron, sits in an incredibly comfortable spot right now, because supply is tight and demand keeps rising.
Samantha: The catch is that this is the definition of a cyclical business. Memory has boomed and busted many times before. What is different this time is that the buyers are not consumer electronics makers but hyperscale AI infrastructure builders, and their spending plans stretch years out. For European tech companies the picture is mixed. On one hand, more competition between Samsung, SK Hynix and Micron should keep supply flowing. On the other, prices for the highest-end memory have roughly doubled over the past year, and that gets baked into the cost of every AI server a European bank, telecom or research lab buys.
Bob: It also underlines how much of the AI value chain still sits in East Asia. Europe has design capability and some leading-edge equipment through the Dutch firm ASML, but on the memory side there is essentially no European alternative to the three big players. That is exactly the kind of dependency the proposed Cloud and AI Development Act in Brussels is trying to address, though the numbers involved there are still a fraction of what Samsung alone is set to earn in a single quarter.
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.
Micron breaks ground on a nine billion dollar memory factory in Hiroshima with heavy Japanese state support.
Bob: Staying with memory chips, but this one is a non-AI-shaped story in an important way: it is about industrial policy and physical construction. Micron Technology has broken ground on a one and a half trillion yen expansion of its factory in Hiroshima, Japan. That is around nine point three billion dollars, dedicated to advanced memory chips including the high-bandwidth type we just talked about. First shipments are expected in the summer of 2028.
Samantha: And Tokyo is putting serious money behind it. The Japanese government is contributing up to 500 billion yen, so roughly a third of the total. That is a very deliberate bet by Japan to claw back a leading position in semiconductors, after decades of losing ground to South Korea and Taiwan. Hiroshima becomes a strategic node, not just for Micron but for Japan's whole industrial strategy.
Bob: For European industry watchers this is a useful reference point. Japan is spending roughly a third of the factory cost in public money to secure a single facility. Compare that to the European Chips Act, which set aside about 43 billion euros for the entire continent through 2030. The scale of public support that Asian and American governments are willing to put behind a single fab makes Europe's numbers look modest.
Samantha: What changes for European users is subtle but real. More memory capacity coming online in 2028 should help stabilise prices and reduce the risk of the shortages we saw during the pandemic. European automakers, industrial firms, and cloud providers all buy Micron memory. A diversified geographic footprint, so factories in the United States, Japan and Taiwan, rather than one country, also reduces the exposure if geopolitical tensions flare in East Asia. That matters for anyone building products that need a predictable chip supply three years out.
Tesla caps employee AI spending at 200 dollars a week as token bills spiral out of control.
Bob: This next one is small in dollar terms but revealing. Tesla has reportedly introduced an internal policy that caps how much each employee can spend on AI tools at 200 dollars a week. The trigger, according to the report, is that some Tesla software engineers were burning through thousands of dollars a week in AI tokens. So the company at the leading edge of AI in cars is now telling its own staff to slow down on AI subscriptions.
Samantha: This is the awkward reality of the current AI boom that does not get talked about much. Coding assistants and agent tools charge by usage, so-called tokens, and a single engineer running long automated tasks can rack up bills that would have seemed absurd two years ago. The pricing models were built assuming light, occasional use. What is happening in practice, especially with agent tools that keep running in the background, is much heavier consumption.
Bob: And this lands on every finance department in Europe eventually. Companies rolled out AI tools last year with generous free tiers and predictable per-seat pricing. Now the bill is arriving, and it is variable, opaque, and sometimes shocking. Chief financial officers are starting to ask the same questions Tesla is: who is using what, how much, and is any of it producing measurable value.
Samantha: The bottom line is that AI is quietly becoming a serious operating cost line, not just a research budget. For European firms subject to strict cost controls, especially in the public sector, that raises awkward choices. Do you cap usage per person like Tesla just did, do you route staff to cheaper open-source models, or do you accept the bill because the productivity gains justify it? None of those answers is obvious yet, and the honest truth is most companies do not yet have the internal reporting to know which employees are getting value and which are just experimenting.
LibrePhotos is the free open-source photo library that finds faces, places and objects on your own machine.
Bob: Let's close with two things you can actually try today. First up, for anyone drowning in decades of photos on a laptop or an old hard drive: LibrePhotos. It is a free, open-source photo management tool that you install on your own computer or a home server. It scans your library and automatically organises photos by face, by location, by object, and by date, so you can search things like "dog on beach 2019" and it just finds them.
Samantha: The appeal is straightforward. Users report they get the kind of smart search that Google Photos and Apple Photos offer, but without uploading a single image to a cloud service. Everything runs locally. For people in Europe who are increasingly wary of handing family photos to a large American platform, that is a meaningful difference. And it is free, with no subscription tier.
Bob: Setup is the one honest caveat. You need to be comfortable installing something a bit more technical than a normal app, essentially running a small piece of server software on your machine. There are step-by-step guides on the project's site, and for someone who is reasonably confident with computers it is an afternoon's work. Once it is running, though, it just sits there quietly indexing.
Samantha: A useful comparison: it does the same job as the paid tier of some cloud photo services, so a few euros a month adds up over years. LibrePhotos does not charge, and the code is open, meaning security researchers can audit what it does with your images. For families with tens of thousands of photos going back to the early digital camera era, it turns a chaotic folder into something searchable. It works on Windows, Mac and Linux, and there is a phone app to browse your library from anywhere on your home network.
Bob: If you have been meaning to sort out that photo archive for years, this is the weekend project. LibrePhotos, one word, free, open source, runs on your own hardware.
Transportr is the free open-source app that plans public transport journeys across European cities without tracking you.
Samantha: Second find, and this one is genuinely simple. Transportr. It is a free, open-source public transport app for Android that works across dozens of European cities and regions. You type in where you are and where you want to go, and it gives you routes using trains, trams, buses and metros, pulled from official transit data.
Bob: What makes it different from the big-name journey planners is what it does not do. No account required, no advertising, no tracking, no data sold on. The app talks directly to the open data feeds that transit agencies publish, so it is not routing your queries through a commercial middleman. That matters if you use public transport several times a day and would rather not have every trip logged somewhere.
Samantha: Coverage is broad across Europe. It supports networks in Germany, Austria, Switzerland, the Netherlands, Belgium, France, the Nordic countries, parts of Italy and Spain, and a growing list of others. Users note it is particularly handy when travelling, because you can switch between city networks without downloading a separate app for each country. It works offline for schedules once you have looked something up, which is useful in metro stations.
Bob: The interface is deliberately minimal. It is not trying to sell you tickets, or restaurant recommendations, or scooter rides. It just plans your journey and shows you the connections. For anyone who has been frustrated by cluttered transit apps full of pop-ups, that restraint is the point. It is available free from the F-Droid app store and from Google Play.
Samantha: A small, useful, distinctly European tool. Transportr, spelled T-R-A-N-S-P-O-R-T-R, free, open source, on Android.
Samantha: Today we covered: the UN opening its first global dialogue on AI governance in Geneva, Samsung set to overtake Nvidia in quarterly profit on AI memory demand, Micron breaking ground on a huge new memory factory in Japan, Tesla capping how much its own engineers can spend on AI tools, and two things to try yourself: the self-hosted photo library LibrePhotos and the open-source European transit app Transportr.
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.