GPT-5.6 vs Grok 4.5: Frontier AI Race Hits Europe

Samantha: Welcome to The State of Tech, The European Edition, Thursday July nine, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: OpenAI and xAI drop new frontier models on the same day, a US worker shortage threatens the American chip boom, identity security firm Keeper crosses two hundred twenty-five million in recurring revenue, Google launches Gemini for Science to speed up research, and two things you can actually try today: the AI eye screening trend explained through a tool called EyeArt, and a handy travel companion called Rome2Rio for planning trips across Europe. Let's get into it.

OpenAI and xAI launch rival frontier AI models on the same day, raising the stakes for Europe.

Samantha: We start with what's easily the biggest tech story of the day. OpenAI is publicly rolling out its new GPT-5.6 series today, after clearance from the US government for a broader launch. The line-up has three tiers: a flagship model called Sol, a mid-range one called Terra, and a fast, low-cost model named Luna. And on the very same day, Elon Musk's xAI is putting Grok 4.5 out into the wild. Two of the biggest names in AI, going head to head, on the same morning.

Bob: The framing from OpenAI is interesting. They're highlighting big gains in coding, science, and cybersecurity, but they're also openly acknowledging what they call elevated cyber and bio risks. That's not the usual launch language. It suggests these models are capable enough that the company itself feels the need to flag the downsides upfront. The naming shift, dropping the numbered generations for Sol, Terra and Luna, also hints at a strategy: sell it more like a product family, less like an incremental version bump.

Samantha: For European users and businesses, this matters in two ways. First, access. Both models arriving simultaneously means European developers, startups and enterprises get more choice at the frontier, and that usually pushes prices down and quality up. Second, and less comfortable, is the regulatory angle. The EU AI Act is now the reference framework for anything considered high-risk, and models that their own makers admit carry elevated cyber and bio risks will draw close attention from regulators in Brussels and national supervisors.

Bob: There's also a competitive question for Europe. Local AI developers, from France to Germany to the Nordics, keep saying they need breathing room to build sovereign alternatives. Two American frontier models landing on the same day makes that harder. Enterprise buyers tend to standardise on what's available and cheap right now, not what might exist in eighteen months. The counter-argument is that Europe's edge is not in raw model size, but in applications, safety tooling and vertical use cases like healthcare and industry, where these new models can be plugged in rather than replicated. Either way, this is the moment where the frontier just moved again, and everyone else has to react.

A shortage of skilled workers threatens the entire US semiconductor building boom.

Samantha: Next up, and this one is not about AI models but about the physical world underneath them. The United States is staring down a serious problem: it doesn't have enough skilled workers to actually run the chip plants it's building. A new report drawing on analysis from McKinsey, the industry body SEMI and the National Science Foundation projects a shortfall of up to one hundred fifty-seven thousand full-time workers by 2030. That's engineers, technicians, specialists, the people who keep a fab running.

Bob: The states named as flashpoints are the ones you'd expect: Texas, California, Arizona, New York and Ohio. All of them have huge new facilities either under construction or planned. Washington has thrown enormous public money at bringing chip production back onto American soil, but as this report makes painfully clear, you can pour concrete much faster than you can train a process engineer. And chip plants without the right people don't produce chips, they produce very expensive empty buildings.

Samantha: The knock-on effect for Europe is direct. Europe has its own Chips Act, its own ambitions with new fabs in Germany and elsewhere, and it's fishing in exactly the same talent pool. Industry groups on both sides of the Atlantic have been warning for years that the pipeline of trained technicians is too thin. If the US struggles to staff its plants, that either pulls talent away from Europe with higher salaries, or it delays global chip supply, or both. It also strengthens the case that Taiwan and South Korea will remain absolutely central to the market, no matter how many billions get invested elsewhere.

Bob: There's a slightly awkward truth in this story too. Building chip capacity was sold politically as a jobs story, but the jobs on offer require years of specialist training. Community colleges and technical universities in the affected states are ramping up programmes, but the arithmetic still doesn't work out for 2030 without a serious surge in enrolments. For Europe, the takeaway is to start now, not later, on training pipelines, because the workers being trained today are the ones staffing fabs at the end of this decade.

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.

Identity security firm Keeper hits two hundred twenty-five million dollars as AI agents multiply.

Bob: Third story. Keeper Security, an identity security company, announced today that it's crossed two hundred twenty-five million dollars in annual recurring revenue. On its own that's a business milestone, but the reason behind the growth is what makes it a story worth telling. Keeper says it now protects over ninety-five thousand organisations, including Fortune 500 companies and government agencies. And the driver is something happening quietly inside every large company: an explosion of what the industry calls non-human identities.

Samantha: Let's unpack that, because it sounds abstract. A non-human identity is basically a login used by a machine, a script, or increasingly an AI agent, rather than a person. Every time a company deploys an AI assistant that can read emails, book meetings, query databases or trigger actions, that assistant needs credentials. Multiply that across hundreds of processes and thousands of employees, and suddenly your organisation has far more machine logins than human ones. Each of those is a potential door for attackers.

Bob: And this is where the numbers get uncomfortable. Machine identities are already outpacing human identities in many enterprises, and AI agents are accelerating that curve. If you don't know what all your bots and agents have access to, you can't defend it. That's the pitch behind Keeper's growth, and it's why competitors are piling into the same space. For European businesses this is a very concrete issue. The AI Act, NIS2 and existing data protection rules all assume that companies know exactly who, or what, has access to sensitive systems.

Samantha: The bottom line is that identity has quietly become the new perimeter. Firewalls used to define where your company ended, now it's credentials. For European IT and security teams the practical question is simple: do you have a full inventory of every AI agent, script and service account touching your data, and can you rotate or revoke them quickly? If the answer is no, that's the gap attackers will find first.

Google launches Gemini for Science to speed up research across biology, medicine and climate.

Bob: Fourth story, and it's a big move from Google. The company is rolling out something called Gemini for Science, an initiative aimed squarely at researchers. The idea is to bundle Google's AI models together with a set of tools to help scientists sift through the mountain of published research, spot patterns in complex data, and even generate and test hypotheses. Google is putting some of its greatest hits under this umbrella: AlphaFold for figuring out how proteins fold, AlphaGenome for identifying disease-linked genes, and new research engines for fields like solar forecasting and epidemiology.

Samantha: The pitch is that no human researcher can keep up with the sheer volume of science being published every week. Gemini for Science is meant to act as a kind of tireless research assistant, reading everything, drawing connections, and suggesting experiments worth actually running in the lab. Testing thousands of hypotheses in software before spending money on physical experiments could genuinely shift how research is done. AlphaFold already did that for structural biology and won its creators a Nobel Prize, so this isn't hype without a track record.

Bob: For European science, this cuts two ways. On one hand, universities from Cambridge to ETH Zurich to Karolinska now have access to some serious firepower they don't have to build themselves. Smaller labs in particular could punch well above their weight. On the other hand, there's a growing dependence question. If the tools that generate hypotheses, filter literature and design experiments all come from one American company, that shapes what science gets done and how. European research funders have been talking about sovereign AI for science, but so far the practical tools sit mostly with Google, Microsoft and a handful of US labs.

Samantha: One area to watch is medicine. Combining AlphaFold-style protein work with genomics tools could genuinely accelerate drug discovery, and European pharma companies are among the biggest users of these platforms. The next few years will show whether AI-assisted science delivers on the promise, or whether the productivity gains are more modest than the headlines suggest. Either way, the direction of travel is now very clear.

AI eye screening is booming, and EyeArt is one tool bringing it into everyday clinics.

Bob: To close out, two things you can actually check out today. First, a look at where AI is quietly showing up in healthcare. New market figures out today project that the global AI eye screening market will grow from around one hundred fifty million dollars this year to roughly one point one seven billion by 2032. That's a compound growth rate close to thirty-seven percent a year, driven by hospitals and clinics adopting AI tools that scan retinas to catch diseases like diabetic retinopathy early.

Samantha: One of the concrete tools behind that trend is called EyeArt, developed by a company called Eyenuk. It's an AI system that analyses retinal photos taken with a standard camera and flags signs of diabetic eye disease within seconds, without a specialist having to be in the room. It's been cleared by regulators in both the US and Europe, and it's already used in clinics, GP surgeries and even some pharmacies in a handful of countries.

Bob: For listeners, this isn't something you install on your phone this afternoon, but it is something to ask about. If you have diabetes, or a family history of eye disease, an AI-assisted retinal scan is increasingly available at your regular check-up rather than requiring a separate hospital appointment. Reviewers and clinicians report that EyeArt catches many early cases that would otherwise slip through, simply because the scan takes minutes rather than a full specialist consult.

Samantha: The wider point is that AI in healthcare is finally moving beyond flashy demos into routine screening. Asia-Pacific is leading the growth, but European health systems are quietly rolling these tools out too, especially in the Nordics, the Netherlands and the UK. If you're curious whether your local practice offers it, the phrase to ask about is AI-assisted retinal screening. Ten minutes of your time could catch something years earlier than the traditional route.

Bob: Not every day is a new app, sometimes the story is that AI is quietly showing up at your optician.

Rome2Rio is the free travel planner that maps any journey across Europe in seconds.

Samantha: And for the second one, something you can genuinely open on your phone this afternoon. With the 2026 FIFA World Cup pulling attention to major travel and events, and summer holiday planning in full swing across Europe, a useful and often overlooked tool is Rome2Rio. It's a free travel planning website and app that shows you every way to get from point A to point B, whether that's train, bus, ferry, flight or car, all on one screen.

Bob: What makes it particularly handy in Europe is that it pulls together operators most single apps ignore. Regional trains in Italy, ferries between Greek islands, night buses across the Balkans, the small budget airlines that don't show up in the big booking engines. You type in Amsterdam to Lisbon, and it gives you the direct flight, the two-day train combination, and everything in between, with estimated prices and travel times. It's the sort of tool that quietly saves you money because it shows options your usual booking site hides.

Samantha: It's free to use, works in a browser or as an app on iPhone and Android, and doesn't require an account for basic searches. Users often mention that it's especially useful for planning trips that cross multiple countries, or for reaching smaller towns where the obvious flight route doesn't exist. It won't always book the ticket for you directly, but it points you at the right operator, which is half the battle solved.

Bob: A quick, free, non-AI tip that just makes European travel a bit less of a puzzle. Give it five minutes and see what routes you didn't know existed.

Samantha: Today we covered: OpenAI and xAI launching new frontier models on the same day, the US chip industry running out of skilled workers, Keeper Security crossing two hundred twenty-five million as AI agents multiply, Google launching Gemini for Science, and two things to try yourself: EyeArt as the face of AI eye screening in clinics, and Rome2Rio for planning any journey across Europe.

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.