Nvidia China Export Controls: EU Impact Explained

Samantha: Welcome to The State of Tech, The European Edition, Tuesday June second, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: Microsoft's new Surface laptop with Nvidia's most powerful chip yet, Washington tightens the screws on Nvidia chips heading to China, NVIDIA launches an open world model for physical AI, Microsoft Build puts AI agents and local Windows AI centre stage, an open-source tool that shrinks AI to run on your laptop, and a Vietnamese humanoid robot making its European debut. Let's start with Microsoft's biggest hardware bet in years.

Microsoft unveils Surface Laptop Ultra with Nvidia RTX Spark superchip at Computex 2026.

Samantha: So Microsoft has gone big. At Computex in Taipei they unveiled the Surface Laptop Ultra, and this is the company's first real attempt to build a premium laptop that directly takes on Apple's MacBook Pro. The headline feature is Nvidia's new RTX Spark superchip inside, which combines a twenty-core processor with a Blackwell graphics chip and a hundred and twenty-eight gigabytes of shared memory. It's a fifteen-inch machine, and Microsoft is positioning it squarely at creative professionals and AI developers.

Bob: And what makes this interesting isn't just the spec sheet, it's the partnership. Microsoft and Nvidia putting their names on the same piece of hardware is a statement. For years the high-end creative laptop space has belonged to Apple. With the Surface Laptop Ultra, Microsoft is saying you can run serious AI workloads locally, on Windows, without ever touching a cloud server. Think video editing with AI effects, three-D rendering, running large language models on the device itself.

Samantha: And that local angle matters because the whole industry is shifting. Up to now, if you wanted real AI horsepower you sent your data to a cloud provider. With a chip like this in a laptop, you can keep everything on the device. That changes the economics for freelancers, studios, and small companies who don't want to pay cloud bills every month.

Bob: It also changes things for Intel and AMD. Both have been pushing AI PCs for over a year now, but Nvidia muscling into the laptop processor market with Microsoft as launch partner is a different category of threat. The question is price. Microsoft hasn't put a clear number on it yet, but with this kind of silicon inside, this is not a mainstream machine. It's a flagship.

Samantha: For European listeners, there's a specific angle here. European creative industries, from Berlin's game studios to Paris's animation houses to London's post-production sector, have been some of the heaviest users of high-end Apple hardware. A credible Windows alternative with this kind of local AI power gives European creative businesses a real choice for the first time in years. And for sectors with strict data rules, like healthcare research or legal work, on-device AI means sensitive material never leaves the building. That aligns neatly with how European data protection rules want things handled. The catch is availability. European launches of Surface flagships often trail the US by months, and pricing in euros tends to add a premium. So while the announcement is global, the practical impact for European buyers may not land until late this year.

US tightens export controls to block Nvidia AI chips reaching China via Malaysia.

Bob: Staying with Nvidia, but a very different story. The US Commerce Department has tightened export controls to stop Nvidia's most advanced AI chips reaching Chinese companies through Malaysian subsidiaries. Washington believes hundreds of thousands of high-end chips have been routed through Malaysia to China over the past year, and these new rules are designed to close that loophole.

Samantha: The timing is pointed. This came down just before Jensen Huang's keynote at Computex. Nvidia has been walking a tightrope for years, trying to stay compliant with US rules while keeping access to the Chinese market, which is one of its biggest. Every time Washington tightens the screws, Nvidia has to redesign chips, lose revenue, or both.

Bob: And the Malaysia angle is the new twist. Until now most enforcement focused on direct shipments. Going after subsidiaries means the US is willing to police the whole supply chain, not just the border. That has knock-on effects for any country that hosts chip distribution, including parts of Southeast Asia and the Middle East.

Samantha: For Europe, this is consequential in two ways. First, European chip equipment makers, particularly ASML in the Netherlands, sit in the middle of this geopolitical tug-of-war. Every new US rule puts pressure on Brussels to either align with Washington or carve out its own position. Second, European AI companies that buy Nvidia chips could face supply tightness or price increases if Nvidia loses China revenue and reprices globally to compensate.

Bob: There's also the question of retaliation. China has been steadily restricting exports of rare earths and other materials that European industry needs. Each escalation makes the European position more uncomfortable, caught between two superpowers neither of which is particularly interested in European preferences.

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.

NVIDIA launches Cosmos 3, an open foundation model for physical AI and robotics.

Bob: NVIDIA has launched Cosmos 3 at GTC Taipei, and this is aimed at what the industry calls physical AI, which basically means AI that controls things in the real world, robots, autonomous vehicles, factory equipment. Cosmos 3 is what NVIDIA calls an open world foundation model, and it can handle text, images, video, ambient sound and physical actions all in one system.

Samantha: The key claim is speed. NVIDIA says Cosmos 3 can cut AI training cycles for robotics from months down to days. For anyone who's followed humanoid robotics, that's the bottleneck right now. Training a robot to handle real-world physics, gravity, friction, unpredictable objects, takes enormous amounts of compute and time. If NVIDIA can genuinely compress that, the whole field moves faster.

Bob: And the model is open, which matters. NVIDIA also announced what they're calling the Cosmos Coalition, with AI labs and robotics companies collaborating on these world models. Open in this context means researchers and smaller companies can build on it without licensing fees, which is how you get an ecosystem rather than just one company's product.

Samantha: For Europe, this lands at an interesting moment. European robotics has real strengths, German industrial robots, Swiss precision engineering, Nordic warehouse automation, but the foundation models powering the next wave of intelligence in those machines have mostly come from US companies. An open model lowers the barrier for European robotics firms to add advanced AI without having to build everything from scratch.

Bob: The flipside is dependency. If European robotics ends up running on NVIDIA's world model, that's another layer of US technology embedded in European industry. Brussels has been talking about tech sovereignty for years. Open models help with access, but they don't solve the underlying question of who controls the foundation.

Microsoft Build 2026 puts AI agents and local Windows AI at the centre.

Samantha: Microsoft Build 2026 also kicked off, and the theme is AI agents and local AI on Windows. Microsoft is rolling out new tools and interfaces that let developers build AI features directly into Windows applications, without sending everything to the cloud. There's also an update to GitHub Copilot, which is shifting to token-based billing, and developers have mixed feelings about that.

Bob: The token-based billing is the part that's stirring up the developer community. Until now Copilot has been a flat monthly fee. Switching to charging based on how much you use it means heavy users pay more, light users pay less, and budgets become harder to predict. For solo developers and small studios, that's a real shift.

Samantha: The broader push, though, is local AI on Windows. Microsoft wants developers to build apps where the AI runs on the user's own machine, which ties directly into that Surface Laptop story we opened with. It's all part of one strategy, sell more powerful Windows hardware, get developers to build AI apps that take advantage of it, and reduce the dependency on cloud costs that have been eating into margins industry-wide.

Bob: For European developers, there's a meaningful upside. Local AI means data stays on the device, which makes building applications that comply with European privacy rules much simpler. You don't need cloud agreements, regional data centres, or complex transfer mechanisms if the AI never leaves the user's laptop. That's a genuine advantage for European software businesses serving regulated sectors like health, finance and legal.

Samantha: The catch is the billing change. European developers tend to be more cost-sensitive than their US counterparts, and unpredictable monthly bills are a harder sell to European finance departments. Microsoft will need to explain very clearly what an average user actually pays before this lands well across the continent.

Bob: To close out, two things you can actually use today.

Tether AI releases open-source TurboQuant SDK to run large AI models on your own device.

Samantha: First one is for anyone who wants to run bigger AI models on their own hardware. Tether's AI research group has just released an open-source toolkit called TurboQuant. It's based on a memory compression technique from Google Research, and the whole point is to shrink how much memory an AI model needs to run.

Bob: In plain terms, AI models eat memory. The bigger the model, the more memory it needs, and that's why running serious AI on a laptop or phone has been impossible until recently. TurboQuant compresses the memory footprint, especially for long conversations or processing big documents. That means a model that used to need a beefy server can now run on a regular laptop, or even a phone, without falling over.

Samantha: It's free, it's open source, and it's available to download today on the usual developer platforms. If you're a developer or a hobbyist who's been experimenting with running local AI on your own machine, this is genuinely worth a look this week. You don't need new hardware, you just need to integrate the toolkit into your existing setup.

Bob: For European users specifically, the appeal is privacy. Running AI locally means your prompts, your documents, your conversations never touch a cloud server. For anyone handling client data, medical records, or just personal information they'd rather keep private, that's a real advantage. It's also useful if you're somewhere with poor internet, because once the model is loaded, you don't need a connection at all.

Samantha: The practical bit. You'll need a laptop with at least sixteen gigabytes of memory to get meaningful results, more if you want to run the bigger models. It works across Windows, Mac and Linux. And because it's open source, there's no licence fee and no account to set up. Just download, install, and try it with whatever local AI setup you already have. A weekend project for the curious.

VinDynamics debuts Dyno humanoid robot at ICRA in Vienna and Computex Taipei.

Bob: And the second find, this one's not something you can download, but it's something you can go and see. Vietnamese tech company VinDynamics, part of the Vingroup, just debuted its first humanoid robot, called Dyno, at the IEEE robotics conference in Vienna and at Computex in Taipei.

Samantha: Dyno is pitched as a household and security assistant. It can move objects with dexterous arms, do surveillance patrols, and act as a guide with multilingual narration. The demos in Vienna had it walking visitors around exhibitions, answering questions in several languages. It's the first time Vietnam has put a serious humanoid robot on the global stage.

Bob: The reason this is worth flagging as something to check out is the videos. The conference demos are being shared widely online this week, and if you've been following the humanoid robot space, this is a useful benchmark for where the field actually is in 2026. Not science fiction, but also no longer just a Boston Dynamics curiosity.

Samantha: And there's a European angle that's quite practical. ICRA being held in Vienna means European robotics researchers, students and industry people can go and see Dyno in person this week. If you're anywhere near Austria and you've been curious about humanoid robotics, the conference floor demos are open at certain times. Worth checking the ICRA schedule.

Bob: It also signals something broader. Humanoid robotics has been dominated by US and Chinese firms. A Vietnamese entrant showing up at the top European robotics conference is a sign that the field is genuinely globalising. For European robotics companies, that means more potential partners, more competition, and a faster pace overall.

Samantha: Today we covered: Microsoft's Surface Laptop Ultra with Nvidia's new superchip, tighter US export controls on Nvidia chips to China, NVIDIA's Cosmos 3 open model for physical AI, Microsoft Build's focus on AI agents and local Windows AI, the TurboQuant toolkit for running AI on your own device, and VinDynamics' Dyno humanoid robot debuting in Vienna.

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.