Snapdragon 8 Elite Gen 6: Agentic AI on Phones + CUDA Rivals

Samantha: Welcome to The State of Tech, The European Edition, Sunday October four, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: a new chip from Qualcomm that pushes AI agents straight onto your phone, a US-Taiwan drone partnership tested on the Ukrainian battlefield, DeepSeek and Huawei chipping away at Nvidia's software monopoly, a UK-Taiwan pact on satellites and silicon photonics, and two things you can actually use today, the open-source humanoid robot platform RP1 and an Indian AI cancer-diagnosis system. Let's get started.

Qualcomm pushes agentic AI straight onto your phone with the Snapdragon 8 Elite Gen 6.

Samantha: Qualcomm has pulled the wraps off its next flagship mobile chip, the Snapdragon 8 Elite Gen 6, and the headline is not raw speed. It is agentic AI running directly on the device. That means the kind of autonomous AI assistants that until now mostly lived in the cloud will start handling tasks locally on your phone, from drafting messages to managing your calendar to editing photos, without constantly phoning home to a data centre.

Bob: And that shift matters for two concrete reasons. First, latency. An AI agent that lives on your phone can act in a fraction of a second instead of waiting on a round trip to a server. Second, privacy. If your assistant is reading your email and your location to decide what to do next, keeping that processing on the device means less sensitive data flowing across networks. Qualcomm is also pitching better battery life, because specialised AI silicon uses less power than general-purpose chips doing the same work.

Samantha: There is a competitive angle too. Apple has been doing on-device AI for a couple of generations now with its own silicon, and Google's Tensor chips go the same direction. Qualcomm supplies most of the Android world outside Google and Samsung's in-house parts, so this launch sets the baseline for what a premium Android phone will do next year. Phone makers from Xiaomi to OnePlus to Honor will build around it.

Bob: For European users, this lands at an interesting moment. Brussels has been pushing hard on data minimisation under the GDPR, and the AI Act adds extra scrutiny for systems that make decisions about people. On-device agents sidestep part of that pressure, because sensitive processing never leaves the handset. That is a selling point European carriers and retailers can genuinely lean on when the phones hit shelves next year. It also raises fresh questions about what these agents are allowed to do autonomously, and how users give or withdraw consent.

Samantha: The bottom line is that the smartphone is quietly turning into a small AI computer, and Qualcomm wants to make sure Android handsets keep pace with Apple on that front. Expect the first devices with the chip in the first quarter of next year.

AEVEX and Taiwan's CIRC unveil battle-tested drones Vandal and Onyx after Ukraine trials.

Bob: Moving on. US-based AEVEX Aerospace and Taiwanese manufacturer Coretronic Intelligent Robotics, usually shortened to CIRC, have unveiled two new drones called Vandal and Onyx. What makes them stand out is not the spec sheet, it is where they have been. Both models have been tested in Ukraine, which means they have been through the single harshest proving ground for unmanned systems anywhere in the world right now.

Samantha: And that pattern of Ukraine testing has become a kind of certification in itself for drone makers. Electronic warfare in that conflict is brutal, with jamming and spoofing on both sides. A drone that still functions there has demonstrated something a lab cannot replicate. AEVEX says it will expand operations in Taiwan under this technology-transfer partnership, which signals a longer-term commitment rather than a one-off product launch.

Bob: The geopolitics here are hard to miss. Taiwan has been racing to build up its own drone industry, both for self-defence and as an export sector that reduces reliance on Chinese components. Pairing with a US firm gives CIRC access to Western markets and defence contracts, while AEVEX gets manufacturing scale and engineering talent. For European defence buyers, this creates another non-Chinese supply option at a moment when that choice is politically essential.

Samantha: European armed forces have been scrambling to catch up on cheap, attritable drones after watching how they reshape ground combat in Ukraine. Several EU member states have launched procurement programmes aimed at buying tens of thousands of units. A product that already carries combat validation shortens that purchasing cycle considerably. It also puts pressure on European drone startups like Germany's Helsing and Quantum Systems to show equivalent battlefield credentials.

Bob: The broader shift is that drones have moved from a niche capability to a core piece of modern militaries, and the supply chain behind them is being redrawn in real time. Where those drones get built, and under which jurisdiction, is now a strategic question for every European defence ministry.

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.

DeepSeek and Huawei turn up the heat on Nvidia's CUDA software stronghold.

Bob: Nvidia's dominance in AI chips is well known. Less discussed is why that dominance has been so hard to break: it is not just the chips, it is the software layer called CUDA that sits on top. CUDA is the toolkit developers have used for nearly two decades to build neural networks on Nvidia GPUs, and switching away from it has historically meant rewriting enormous amounts of code. That is exactly what DeepSeek and Huawei are now attacking.

Samantha: Both companies are developing alternative software stacks that let AI developers run their models on non-Nvidia hardware without starting from scratch. For Huawei, that means its own Ascend AI accelerators, which are increasingly central to Chinese AI infrastructure after US export controls cut off access to Nvidia's top chips. DeepSeek has made its name building capable AI models at a fraction of the usual cost, and part of that story is squeezing more out of lower-tier hardware through better software.

Bob: The ripple effects reach well beyond China. Every major cloud provider, from Amazon to Microsoft to Google, has been building its own AI chips precisely to escape Nvidia's pricing power. A credible alternative software ecosystem makes those in-house chips far more attractive to outside customers, because developers do not have to learn a new world for every vendor. That is the real competitive lever here.

Samantha: For European AI companies, which have smaller budgets than their American rivals, cheaper and more diverse options at the software layer could meaningfully lower the cost of training and running models. European startups building sovereign AI stacks, like France's Mistral or Germany's Aleph Alpha, benefit from any erosion of a single-vendor lock-in. It also fits with the EU's broader push for strategic autonomy in critical digital infrastructure.

Bob: The caveat is that CUDA has a nearly twenty-year head start, and most AI research still assumes it as the default. Breaking that habit takes years, not months. But the direction of travel is clear: the software monopoly that quietly underpinned Nvidia's valuation is finally facing serious challengers.

Taiwan and the UK sign a science pact on satellites, silicon photonics and chip talent.

Samantha: Taiwan and the United Kingdom have signed a Science, Innovation, and Technology Arrangement, SITA for short, with three concrete projects attached. The UK Space Agency and the Taiwan Space Agency will work jointly on advanced satellite communication technology. Universities in both countries will launch a shared silicon photonics research initiative. And there is a programme on semiconductor talent development aimed at training engineers for both nations.

Bob: Silicon photonics is the one worth pausing on. It is the technology that uses light instead of electricity to move data inside chips and between them, and it is becoming essential for AI data centres where the bottleneck is no longer processing power but the bandwidth between chips. Taiwan dominates advanced chip manufacturing through TSMC, and the UK has strong academic research in photonics at places like Southampton and Cambridge. Pairing those strengths is a sensible, low-drama play.

Samantha: The satellite communications piece lands in a crowded field. Europe already has Eutelsat OneWeb and the planned IRIS Squared constellation, while Taiwan has been building out its own capacity partly to reduce dependence on undersea cables that could be cut in a conflict. Collaboration on resilient satellite links has obvious strategic value for both sides.

Bob: What is notable about this deal is that it institutionalises something that was previously ad hoc. The two governments have been cooperating informally for years, but a formal arrangement signed in London gives researchers and companies a predictable framework to build on. That matters when projects run over multiple years and change of government cycles.

Samantha: For the broader European picture, this reinforces a pattern of individual EU and European countries building bilateral semiconductor and space ties with Taiwan, in parallel to collective EU efforts under the Chips Act. Germany has TSMC's Dresden fab under construction, the Netherlands has ASML as a critical supplier, and now the UK adds a research dimension. The underlying message is that semiconductor resilience is no longer purely an American conversation.

RoboParty unveils RP1, a full open-source humanoid robot anyone can build on.

Bob: To close out, two things you can genuinely explore today. First up, at the International Conference on Intelligent Robots and Systems, a group called RoboParty has unveiled RP1, also called ROBOTO 01. It is a full-stack, open-source bipedal humanoid robot. That phrase matters, because most humanoid robot projects today are closed products from companies like Tesla, Figure or Agility Robotics. RP1 publishes its software, its hardware designs and its roadmap openly.

Samantha: For researchers, educators and robotics developers, that changes the economics of getting started. Instead of buying a sealed commercial humanoid for hundreds of thousands of dollars, a university lab or a serious hobbyist team can study, modify and extend RP1 directly. RoboParty says the platform is designed for dynamic motion research, for training embodied AI models, and for collecting the kind of real-world movement data that general-purpose robots need to learn from.

Bob: What you can actually do with it today depends on your appetite for building. The design files and software roadmap are rolling out over the fourth quarter, with details on mass-production options to follow. If you are a developer, the practical first step is to check the project's documentation and the IROS presentation materials, which lay out exactly which parts of the stack are available now and which arrive later. The name to search for is RP1 from RoboParty.

Samantha: For European universities, this is the kind of platform that fits the open-science tradition well. Projects like the Eclipse Foundation's Sovereign AI Initiative have been pushing for more open infrastructure in AI, and robotics is the obvious next frontier. A shared humanoid platform means European labs can collaborate on the same hardware without each buying a different proprietary machine.

Bob: One honest caveat: building a bipedal robot is not a weekend project, even with full designs. This is a tool for teams and institutions more than individual tinkerers. But the fact that the blueprint exists openly at all is the shift worth noting.

Indian researchers combine neural networks with whales-and-swarms algorithms for cancer diagnosis.

Samantha: And second, something from the medical research world that reads a bit like science fiction but is very concrete. A team in India has built an AI system that combines two different types of neural networks with nature-inspired optimisation algorithms, named after the hunting patterns of whales and the collective behaviour of swarms. They used it to analyse histopathology slides for colorectal cancer, and the diagnostic accuracy they report is striking.

Bob: The whales-and-swarms part sounds whimsical, but it is a serious branch of computer science. These algorithms borrow mathematical patterns from how animals forage or hunt collectively, and apply them to tune the many thousands of settings inside a neural network. The result, according to the researchers, is a system that detects cancer patterns in tissue slides more reliably than the standard models alone.

Samantha: What you can do with this today is read the research itself, which has been published openly, and watch for how it feeds into clinical trials. For now this is a research tool, not something a hospital can download tomorrow. But similar AI-assisted pathology systems from companies like Paige and Ibex are already being approved in Europe, so the pipeline from paper to clinic is faster than it used to be.

Bob: For European healthcare systems facing pathologist shortages, AI diagnostic aids are becoming genuinely practical. The United Kingdom's National Health Service has already piloted AI tools for breast cancer screening, and the EU's new regulation on AI in medical devices creates a clearer path to approval. Research like this feeds into that pipeline.

Samantha: Today we covered: Qualcomm's agentic AI chip for phones, Vandal and Onyx drones battle-tested in Ukraine, DeepSeek and Huawei challenging Nvidia's CUDA, the UK and Taiwan signing a satellites and photonics pact, and two things to explore: the open-source humanoid robot RP1, and the Indian whales-and-swarms AI cancer diagnosis system.

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.