OpenAI reportedly eyes Google AI chips, challenging Nvidia

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

Bob: And I'm Bob Russell. Today: OpenAI reportedly turns to Google for AI chips, a direct challenge to Nvidia. Sam Altman tells companies to put AI security ahead of productivity. OpenAI signs an AI safety deal with Poland's national research institute NASK. A Taiwanese drone maker teams up with two American firms on drones that coordinate on their own. And two practical items: what Travelport's new AI-native booking platform means for your next trip, and how to protect your photos from DeepNude-style image tools. Let's start with the chips.

OpenAI reportedly plans to buy AI chips from Google, challenging Nvidia's grip on AI hardware.

Samantha: OpenAI is reportedly planning to buy artificial intelligence chips from Google. That decision lands right in the middle of the most important supply chain in tech. Nvidia has been the dominant name in the hardware that trains and runs large AI models. If one of the biggest AI developers in the world starts sourcing chips elsewhere, that dominance faces a real challenge. The reporting frames it as a strategic move: OpenAI wants a broader base for training and deploying its models.

Bob: The word to hold on to is diversification. Relying on a single supplier for your most critical component is a risk for any company, and for an AI lab it can mean waiting in line for hardware while competitors move ahead. Adding Google's chip technology spreads that risk. This is reported as a plan, not a delivered order, so the scale and timing are still unclear. But the signal matters, because every big buyer that looks elsewhere strengthens the case for alternatives.

Samantha: There's also a twist in who the supplier is. Google is not a neutral parts vendor. It builds its own AI and competes with OpenAI for users and business customers. So OpenAI would be leaning on a competitor's hardware to power its own products. That shows how intense the race for computing power has become: securing chips now outweighs the awkwardness of buying from a rival. The source places it in a broader push by tech giants to secure their own AI hardware.

Bob: For the market, the bottom line is more competition in AI chips, and the reporting suggests that could speed up innovation in a sector shaped by one company for years. More credible alternatives can mean more pressure on pricing and supply, though that depends on how much capacity Google offers outside its own walls. The source doesn't cover that yet, so for now this is a direction of travel rather than a finished shift.

Samantha: Which brings us to Europe. European AI developers compete for the same scarce hardware as the American giants, often with smaller budgets and less negotiating power. If the chip market opens up, they could end up with more options and less dependence on a single vendor. That matters for startups training their own models and for companies that want to run AI inside Europe. Key takeaway: when the biggest buyers shop around, everyone further down the queue may get more choice.

Sam Altman warns companies that AI security must come before productivity gains.

Bob: Sam Altman has a message for companies rushing into AI: look at security first. Speaking at Oracle's AI World event, the OpenAI chief said understanding and mitigating AI security threats should be the "top short-term priority" for any organisation using AI. That puts it ahead of the productivity gains most companies are buying AI for in the first place. Coming from the head of one of the leading AI developers, that is a notable ordering of priorities.

Samantha: The reason he gave stands out. Altman said he was surprised by how many software vulnerabilities AI systems are uncovering. In other words, these models are getting very good at finding weak spots in code. That cuts both ways. A company can use that ability to find and fix its own holes before anyone else does. But the same capability in the wrong hands means attackers can find those holes faster too. That's the tension behind his warning.

Bob: According to the reporting, it reflects a wider worry in the AI community: that advanced models could both expose and create new security challenges. This is important because many AI rollouts are judged on speed. Hours saved, emails drafted. Altman's warning suggests that's the wrong place to start. If AI is plugged into internal systems without anyone checking what it can reach and what it can reveal, the productivity gain can turn into a security bill later.

Samantha: For European companies, the message lines up with the EU's focus on responsible AI and managing risk. The source frames it as a push for robust AI security frameworks. That doesn't create any new legal obligation today. But a European board asking whether its AI rollout is secure now has backing from one of the people building the technology. The bottom line: before measuring what AI saves you, check what it might expose.

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.

OpenAI signs an AI safety agreement with Poland's National Research Institute NASK.

Bob: OpenAI has signed an AI safety agreement with Poland's National Research Institute, known as NASK. The deal is about shared methods: how to evaluate what an AI model can do, how to assess its risks, how to test whether its safeguards hold up, and how to report security incidents. That sounds procedural, but these are the building blocks of any oversight system. Without common methods, every lab and government measures safety its own way, and the results can't be compared.

Samantha: That comparability is the explicit goal. According to OpenAI, the agreement is part of its push for international technical standards for frontier AI, the most powerful models at the cutting edge. The idea is a framework where researchers and public authorities can line up their findings across organisations and national borders. So a test done in Warsaw could, in principle, be read and checked by researchers elsewhere, and the other way round.

Bob: Notable is the choice of partner. This isn't a deal with an EU institution in Brussels, but with a national research body in Poland. That gives one member state a direct technical line into how a leading AI developer evaluates its models. It also shows the AI safety debate is no longer limited to Washington, London or San Francisco. The source doesn't spell out a timeline or which models get tested first, so those details remain open.

Samantha: The European angle is twofold. First, it strengthens Europe's seat in global AI safety discussions through hands-on technical work, not only legislation. Second, the source suggests the agreement could influence future EU rules and standards for AI. That's a possibility, not a done deal. But if shared testing methods prove workable in Poland, they could become a reference point for other European regulators. Key takeaway: AI safety is shifting from principles on paper to concrete testing.

Taiwan's AerosoarX teams up with Rajant and SiMa.ai on drones that coordinate on their own.

Bob: Taiwanese drone developer AerosoarX has teamed up with two American companies, Rajant and SiMa.ai, to build drones that can think and talk for themselves. Rajant brings mobile mesh networking, which lets drones pass signals to each other like a relay, instead of each one needing a direct line to a ground station. SiMa.ai brings the onboard computing, a small AI chip that processes sensor data right on the drone.

Samantha: Put those together and you get drones that analyse what their sensors pick up locally, then coordinate with each other autonomously. The aim, according to the companies, is less reliance on constant ground control. That matters when the link to the operator is weak, distant or interrupted. A group of drones that keeps sharing information among themselves is more resilient than one that depends on a single connection.

Bob: The hardware choice reveals the engineering challenge. AerosoarX chairman Max Lo said SiMa.ai's processor was picked for its performance per watt. In plain terms: how much computing you get for every unit of battery power. On a drone, energy spent on calculation is energy not spent on staying in the air. So a chip that does more with less power means longer flights or more capability on board.

Samantha: There's a geopolitical layer too. This is a Taiwanese company deepening ties with American tech firms in uncrewed systems, a field with defence as well as commercial uses. The source describes it as part of a growing trend of international partnerships in AI-powered drones. The announcement doesn't say which customers or missions the systems are meant for, so it's not yet clear whether early deployments will be military, civilian, or both.

Bob: For Europe, the source points to a potential impact on both the defence and commercial drone markets. Partnerships like this raise the bar for what a competitive system looks like: local processing, networking between drones, and long endurance. European suppliers will be measured against that. The bottom line: drones are becoming less like remote-controlled cameras and more like small, cooperating networks.

Travelport and Cognizant build an AI-native platform to reshape how travel gets booked.

Samantha: To close out, two practical things, starting with how you'll book your next trip. Travelport and Cognizant are expanding their partnership to build what they call an AI-native platform for global travel retailing. After two decades of working together, the next phase embeds AI across Travelport's technology, products and operations. The plan, announced in Bengaluru, aims to modernise how the systems are built, automate workflows, and improve booking for airlines, travel agencies and travellers.

Bob: The reason behind it is about you. The companies say travellers increasingly use AI tools to plan trips, and existing booking systems need to evolve to keep up. You can ask a chatbot for a weekend itinerary in seconds, but the system behind the actual booking was often built long before anyone talked to their computer. Travelport sits in the middle of that chain, connecting airlines with the agencies and sites where people book.

Samantha: To be clear about what this is: there's no app to download today. It's a platform under construction, and the announcement gives no launch date or says which features reach travellers first. What you can take from it now is a direction: the booking flow behind agencies and airline sales is being rebuilt with AI in mind. If you already plan trips with an AI assistant, the gap between planning and booking is what the industry is now trying to close.

Bob: For European travellers and businesses, the source suggests this could mean more efficient and more personalised booking. That fits a continent where one trip easily crosses several countries and carriers, and where business travel is often run through agencies. Practically: next time a travel agency or booking site offers an AI assistant, check whether it can actually complete and change bookings, or only suggest options. That distinction is where platforms like this aim to make the difference.

DeepNude-style AI image tools grow more realistic, making photo protection a practical priority.

Samantha: The second item is about protecting yourself rather than trying something new. DeepNude started as a single application, but by 2026 the name stands for a broader class of AI systems that produce strikingly realistic synthetic images. They modify existing photos, estimating and reconstructing visual details that were never in the original. The tools have become more accessible and their output more realistic, which is why they're back in the spotlight.

Bob: The core risk is that the raw material is an ordinary photo. Any picture of a person can, in principle, be turned into a fake image that person never agreed to. As realism improves, these fakes get harder to tell apart from real photos, which deepens the harm to the person shown. That's why the discussion has moved towards regulation, ethical safeguards built into AI tools, and practical protections against misuse.

Samantha: The one thing you can act on this afternoon is visibility. Take ten minutes to review who can see the photos on your social media accounts, and whether old public albums still need to be public. That won't make anyone immune, and the source promises no single fix. But it reduces how easily a stranger can grab pictures of you or your kids. Understanding how these systems work, as the guidance stresses, is the first step to not being caught off guard.

Bob: In Europe, regulators continue to grapple with the ethical and legal implications of this kind of image manipulation. According to the source, that's driving discussions on stricter rules for how such content can be generated and used. Those rules are still under discussion, not finalised. Meanwhile, responsibility sits partly with platforms and AI developers, and partly with users deciding what they share publicly. Key takeaway: as these tools get more realistic, the protections around them matter more.

Samantha: Today we covered: OpenAI reportedly buying AI chips from Google and challenging Nvidia, Sam Altman putting AI security ahead of productivity, OpenAI's safety pact with Poland's NASK, and AerosoarX's self-coordinating drones with Rajant and SiMa.ai. And two practical items: Travelport's AI-native booking platform, and protecting your photos from DeepNude-style image tools.

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.