Gemini hits 1 billion users: EU digital sovereignty at stake

Samantha: Welcome to The State of Tech, The European Edition, Saturday August fifteen, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: Gemini races past one billion users, OpenAI loosens the leash on a cyber-focused GPT, SpaceX floats the idea of data centres in orbit, a new consortium tries to set enterprise AI standards, plus two things to try yourself, the offline map app Maps dot me and the free writing tool Ellipsus. Let's get started.

Google's Gemini becomes the fastest AI app ever to reach one billion global users.

Samantha: Google says its Gemini assistant has crossed one billion users worldwide, making it the fastest-growing product the company has ever launched. That puts Gemini in the same league as Chrome, Gmail and YouTube, and it happened despite Gemini arriving four months after ChatGPT. As a standalone app it only landed in 2024, so this is a very steep curve.

Bob: And the framing matters. Google is claiming it reached the billion mark faster than ChatGPT did, which flips a story that for two years has been about OpenAI setting the pace. Part of that is distribution: Gemini is baked into Android, into Google Search, into Workspace. When your assistant lives on billions of phones by default, adoption looks very different from a standalone download.

Samantha: The number to watch, though, is not users but active users. A billion people opening an app once is not the same as a billion people using it weekly. Google did not break that out. What is clear is that AI assistants are becoming background infrastructure, the way maps and email did before them. And the competitive picture is tightening: OpenAI, Google, Anthropic and Meta all now claim hundreds of millions of regular users each.

Bob: For Europe, this cements a familiar dependency. The most-used AI assistants on the continent are almost all American. European alternatives like Mistral's Le Chat exist, but they operate at a very different scale. Regulators in Brussels have been pushing the AI Act and digital sovereignty language for years, and a headline like this is exactly the kind of thing that keeps that debate alive. If a billion people are already inside Gemini, the question of where their prompts and data are processed becomes less abstract by the week.

OpenAI releases a less restricted version of GPT-5.6 aimed at professional cyber defenders.

Samantha: OpenAI has quietly released GPT-5.6 Cyber, a version of its flagship model with fewer guardrails around cybersecurity content. It is meant for vetted defenders, not the general public. According to internal benchmarks the company shared, the model completes ninety-five percent of tasks involving exploit-chain development and privilege escalation. The previous generation managed just over fifty-seven percent, and the standard consumer version with full safeguards sits at one and a half percent.

Bob: That is a striking jump, and it is also a genuinely uncomfortable one. OpenAI's argument is that defenders need parity with attackers, and that criminal groups already use unrestricted models bought or built outside the mainstream. If your security team is testing whether a bank's network can be breached, a model that refuses to discuss exploits is not useful. So the company is essentially saying: we will give this power to people we can verify, and hope the vetting holds.

Samantha: The obvious risk is leakage. History with restricted AI models is not encouraging: weights get exfiltrated, credentials get shared, contractors move jobs. And a model that can autonomously chain together exploits is a very different category of tool than a chatbot that explains a concept. Security researchers have been warning for months that offensive AI capability is scaling faster than the defensive side can absorb it.

Bob: For European buyers, this collides directly with the AI Act's high-risk provisions and with national cybersecurity rules. A model tuned for offensive testing sits close to dual-use territory, and export-control lawyers will be reading OpenAI's terms carefully. Expect European security agencies, from Germany's BSI to France's ANSSI, to want a very clear answer on who exactly counts as a vetted defender, and how that vetting is audited.

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.

SpaceX proposes putting AI data centres in orbit to escape Earth's power and cooling limits.

Bob: SpaceX has filed a proposal with the US Federal Communications Commission suggesting that the answer to the AI power crunch might be, literally, off-planet. The pitch: put data centres in orbit, where solar power is effectively unlimited and space itself provides cooling. On Earth, the constraints piling up are electricity, water, grid capacity and land near substations. US data-centre power demand is projected to roughly double by 2027, and something has to give.

Samantha: The engineering questions here are enormous. Getting a modern AI training cluster into orbit means launching thousands of tonnes of hardware, keeping it aligned with ground stations, and moving vast amounts of data up and down without lag killing the workload. Cooling in space also is not as simple as it sounds: without air or water, you have to radiate heat away, and that limits how densely you can pack chips. This is a proposal, not a roadmap.

Bob: Still, the fact that a serious FCC filing exists tells you how tight the terrestrial squeeze is becoming. Hyperscalers are already signing deals for nuclear power, buying entire power plants, and queueing for grid connections that will not arrive until the 2030s. Orbital compute is a moonshot, but it is being taken seriously enough to end up on regulators' desks.

Samantha: For Europe, the immediate relevance is less about orbital ambitions and more about the underlying problem. Ireland, the Netherlands and Germany are all wrestling with data-centre moratoria and grid queues. If American operators keep pushing exotic solutions, European policymakers face the same fork: either build far more clean power quickly, or watch AI workloads migrate to wherever the electrons are, whether that is Texas, the Gulf, or, one day, low Earth orbit.

A new consortium of fifty security leaders launches to set real standards for enterprise AI.

Bob: A new group called the AI Trust and Security Consortium has launched, with fifty senior security and technology leaders as founding members. Their goal is practical: co-author reference architectures, control frameworks and board-level governance models for enterprise AI. In other words, moving past high-level principles like "AI should be safe and fair" toward concrete checklists that a chief information security officer can actually apply on Monday morning.

Samantha: The gap they are trying to fill is real. Most companies deploying AI right now are improvising. They are stitching together vendor promises, internal policies and whatever the legal team can approximate from GDPR and the AI Act. What has been missing is the kind of shared playbook that exists for cloud security or payment systems, where you have well-known frameworks that everyone builds against.

Bob: The consortium is peer-governed and independent, which matters because a lot of existing AI safety work is either academic or funded by the model providers themselves. A body led by buyers rather than sellers has a different set of incentives. Whether it produces something genuinely useful or just another set of principles will depend on how prescriptive it is willing to be.

Samantha: For European organisations, the timing is convenient. Parts of the AI Act come into force in stages through 2026 and 2027, and compliance teams are hunting for concrete guidance. If AITSC's reference architectures line up with the Act's high-risk categories, they could become a de facto implementation standard on the continent. If they diverge, European firms will end up maintaining two rulebooks: one for regulators and one for their US-led consortium partners.

The world's first cyclic quantum heat engine could make future quantum computers far simpler.

Bob: Researchers at Aalto University in Finland have built what they describe as the world's first cyclic quantum heat engine inside a superconducting circuit. In plain language: a tiny engine that turns heat near absolute zero into useful work, at the quantum scale. It is a physics milestone, but the practical angle is what makes it interesting.

Samantha: Today's quantum computers are surrounded by a thick jungle of microwave cables running from room temperature down into the coldest part of the fridge. Each cable carries control signals to the qubits, and each one also carries noise and heat that make the whole system harder to scale. If future versions of this engine can operate autonomously inside the quantum computer itself, some of that cabling could disappear.

Bob: That would matter for anyone trying to build a machine with thousands or millions of qubits, which is roughly what you need before quantum computing starts breaking encryption or designing new drugs at scale. Cabling and cooling are two of the biggest bottlenecks right now, and they are not glamorous problems, but they determine whether the field ever leaves the laboratory.

Samantha: This is a European result, and it fits a pattern. Finland, the Netherlands and Germany have been quietly building serious quantum ecosystems, from IQM in Espoo to QuTech in Delft. If the underlying components get smaller and simpler, Europe's position in that supply chain becomes more valuable, not less. It is early-stage science, but it is exactly the kind of work that could shape which continent hosts the first genuinely useful quantum machines.

Bob: To close out, two things you can actually check out today, and neither of them is a chatbot.

The offline map app Maps dot me gives European travellers full navigation without a data connection.

Samantha: First up, a map app called Maps dot me. It is a free navigation tool built on OpenStreetMap data, and its main trick is that everything works offline. You download the countries or regions you need over WiFi, and after that the app runs without any mobile data, without tracking your location back to a server, and without ads getting in the way.

Bob: It is popular with travellers precisely because it does not need a signal. Hikers use it in the Alps, cyclists rely on it across rural France and Portugal, and city visitors like it for walking directions where roaming would otherwise cost money. Reviewers highlight that points of interest, like small cafes or drinking fountains, tend to be surprisingly detailed because the underlying map data is community-maintained.

Samantha: For anyone travelling around Europe this summer weekend, it is worth installing before you leave the house. Available on iOS and Android, free to download, and the whole of a mid-sized European country typically fits in a few hundred megabytes. Once the maps are on your phone, you can put it in airplane mode and still get turn-by-turn walking, cycling and driving directions.

Bob: A small warning: because it uses OpenStreetMap, remote areas can be hit or miss depending on how active local mappers are. Users generally recommend cross-checking a planned hike with an official trail map before setting off. But as a permanent offline backup on your phone, it is one of those apps that pays for itself the first time you lose signal in a mountain valley.

The free European writing tool Ellipsus lets teams draft and edit long documents together without lock-in.

Samantha: And the second one is a writing tool called Ellipsus. It is a free browser-based app aimed at writers who work on long projects, novels, scripts, research pieces, essays, and want proper version history and collaboration without handing everything to a big US platform.

Bob: The pitch is basically: what Google Docs is for office work, Ellipsus tries to be for long-form writing. You can branch a chapter, try a rewrite, compare versions side by side, and share a draft with a co-author or editor. Users highlight that the tracked changes model is closer to how novelists actually work, where you might rewrite the same scene five times before settling.

Samantha: It is built by a small European team, works entirely in the browser, and does not require an account for basic reading of shared documents. That makes it handy if you are collaborating with someone who does not want to sign up for yet another service. There is no AI writing assistant baked in by default, which for some users is precisely the appeal.

Bob: If you are the kind of person who has a long-form project sitting in a messy folder of Word files, it is worth spending a Saturday afternoon importing one into Ellipsus and seeing whether the branching model clicks. Free to try at ellipsus dot com, works on any modern browser, and the export options mean you are not locked in if you decide later it is not for you.

Samantha: Today we covered: Gemini's billion-user milestone, OpenAI's cyber-tuned GPT for defenders, SpaceX's orbital data-centre proposal, the new AI Trust and Security Consortium, Finland's quantum heat engine breakthrough, and two things to try yourself: the offline navigation app Maps dot me, and the collaborative writing tool Ellipsus.

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.