Huawei Tau Scaling Law: China's Answer to ASML Sanctions

Samantha: Welcome to The State of Tech, The European Edition, Wednesday May twenty-seventh, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: a record-breaking seven hundred billion dollar AI infrastructure splurge, Huawei tries to rewrite the rulebook for chips, Beijing locks down its AI talent, Washington puts AI to work in healthcare, and to finish, two things you can actually try out today, a new agentic AI platform and a hard look at what AI is doing to your car's security. Let's get into it.

Big Tech to pour nearly seven hundred billion dollars into AI infrastructure in 2026.

Samantha: We're starting with a number that is genuinely hard to wrap your head around. Meta, Alphabet, Microsoft and Amazon are collectively set to spend close to seven hundred billion dollars on AI infrastructure this year. That's data centres, chips, and the networking gear to tie it all together.

Bob: And Meta is making the sharpest jump. The company has lifted its planned spending by eight percent, to somewhere between one hundred and twenty-five and one hundred and forty-five billion dollars. Microsoft and Alphabet have also revised their forecasts upward, and Amazon is sitting on one of the largest tech budgets in the industry at around two hundred billion.

Samantha: To put that in perspective, seven hundred billion is roughly the size of the entire Dutch economy. Four American companies, spending the GDP of a mid-sized European country, in a single year, on one technology.

Bob: And it's almost all going into the picks and shovels of AI. New data centres, the high-end chips that train and run the models, and the fibre and switches that connect everything. It is a race to build out capacity before competitors do.

Samantha: Which raises the obvious question for Europe. There is no European hyperscaler in that league. The biggest cloud providers on the continent are spending a fraction of these sums. Brussels has been talking about AI sovereignty for years, but the gap in raw infrastructure spending keeps widening.

Bob: For European businesses there are two sides to this. On one hand, more capacity globally means cheaper, faster access to AI services. If you are running a startup in Berlin or Lisbon, you benefit from the build-out, even if it's happening in Virginia or Oregon. On the other hand, it deepens the dependency. Every euro a European company spends on AI tooling tends to flow to an American cloud bill.

Samantha: And there are knock-on effects that don't show up in the spreadsheets. Energy demand. Grid pressure. Water for cooling. Several European countries are already wrestling with data centre permits because of strain on local power supply. If American hyperscalers keep doubling down at this pace, that pressure won't stay confined to Dublin or Frankfurt.

Bob: The other thing worth flagging is what this says about confidence. These companies are betting, very publicly, that AI demand keeps growing fast enough to justify hundreds of billions in fixed assets. If that demand softens, we're looking at one of the largest write-downs in tech history. For now though, the cheques keep getting bigger.

Huawei unveils Tau Scaling Law to reshape chip industry around US sanctions.

Bob: Onto our second story. Huawei has just introduced what it calls the Tau Scaling Law, at a major circuits and systems conference in 2026. Informally it's already being nicknamed Her's Law, after the head of Huawei's semiconductor business, He Tingbo. The pitch is bold: this is meant to be a new foundational principle for chipmaking, an alternative to Moore's Law.

Samantha: Moore's Law, very briefly, is the decades-old idea that you keep shrinking transistors to pack more power into the same chip. Huawei's argument is that this approach is running into both physical limits and political ones, because China is cut off from the most advanced lithography machines, the ones made by Dutch firm ASML.

Bob: So instead of trying to win the shrinking race, the Tau approach proposes a multi-layered design. Performance gains through architecture, packaging, and integration, rather than through ever smaller transistors. In plain terms, if you cannot make the bricks smaller, build smarter walls.

Samantha: Whether the engineering actually delivers is another question. Naming a scaling law after yourself is a bold move. But the strategic intent is unmistakable. China is signalling that it will route around the export controls rather than wait them out.

Bob: And this matters directly for Europe. ASML, based in Veldhoven, is at the centre of the export restrictions. If Huawei and the broader Chinese chip industry succeed in building competitive chips without cutting-edge European lithography, that changes the leverage Brussels and Washington thought they had.

Samantha: It also opens up a question for European chip designers and researchers. If a parallel chip ecosystem emerges around different design principles, do European firms engage with it, or stay aligned with the Western supply chain? That is a strategic choice that's been looming for a while, and Huawei just made it more concrete.

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.

China expands travel restrictions to top AI researchers at private firms like Alibaba and DeepSeek.

Bob: Story three, and it's connected to the last one. China is reportedly expanding travel curbs to cover leading AI professionals at private companies, including Alibaba and DeepSeek. Government agencies are requiring approval before strategically important researchers, founders and executives can travel abroad.

Samantha: This is a notable shift. Travel restrictions on staff at state firms and sensitive research bodies have been around for years. What's new is the focus on private sector AI talent. Startup founders. Researchers at independent labs. The kind of people who, until recently, moved fairly freely between Beijing, San Francisco and Zurich.

Bob: The motivation, according to the reporting, is twofold. Protect critical technology from leaking out, and keep top talent inside the country to accelerate domestic AI development. Beijing clearly sees its AI workforce as a strategic asset, on par with rare earth minerals or advanced manufacturing know-how.

Samantha: For European universities and research institutes, this has practical consequences. Conferences in Paris, Amsterdam or Munich rely on Chinese researchers showing up. Joint research projects depend on people being able to travel. If approval becomes the default barrier, collaboration slows down or moves online only.

Bob: And it cuts in both directions. European companies hiring senior AI staff from China may find that those candidates simply cannot relocate. Recruitment pipelines that have existed for a decade could narrow significantly.

Samantha: There's also a quieter signal here about how Beijing sees the next phase of AI competition. Models can be copied. Chips can be reverse-engineered. People are harder to replace. Restricting their movement says the talent itself is now considered the bottleneck.

Bob: It is worth saying, the reporting describes this as an escalation of existing practice rather than a sudden new policy. But the inclusion of names like Alibaba and DeepSeek puts it on a different scale than before.

US federal agencies push ahead with AI in healthcare, from fraud detection to clinical trials.

Samantha: Story four takes us to Washington, where federal agencies are moving on several fronts to shape how AI is used in healthcare. The White House has postponed a planned executive order on AI and cybersecurity, which suggests there is still real debate inside the administration about how heavy the oversight should be.

Bob: But other parts of government are not waiting. The Department of Health and Human Services has launched a programme called AERO, which uses AI to review audits of federally funded health programmes. The aim is to spot waste, fraud and abuse faster than human auditors can.

Samantha: And the Food and Drug Administration has issued a request for information for a pilot programme on AI-enabled tools in early-phase clinical trials. The framing emphasises trustworthiness and alignment with established AI risk management frameworks. In other words, the FDA wants to understand how AI is already being used in drug development before writing firm rules.

Bob: For European listeners this matters because the FDA and the European Medicines Agency tend to watch each other closely. If the FDA defines what acceptable AI use looks like in clinical trials, the EMA's eventual framework will likely rhyme with it, even if the specifics differ.

Samantha: The AI Act in Europe already classifies AI in healthcare as high-risk, with strict documentation and oversight requirements. But the AI Act sets the floor. How regulators actually inspect and approve AI-driven medical tools in practice is still being worked out, and US precedent will feed into that.

Bob: There's also a competitive angle. American pharmaceutical firms are running AI pilots inside FDA processes right now. If European regulators move more slowly, European drugmakers risk falling behind in adopting these tools, which could affect how quickly new medicines reach patients here.

Samantha: One thing the news makes clear: the era of AI in healthcare being a research topic is over. It is now an operational reality, used to fight fraud, screen trials, and assist doctors. The question for Europe is whether oversight keeps up without strangling adoption.

Alibaba opens its agentic AI platform to global developers, with free training in Singapore.

Bob: To close out, two things you can actually use today. First up: Alibaba Cloud has just launched what it calls its Agentic AI Ecosystem at its first international Qwen Conference in Singapore. The headline product is Qwen3.7-Max, the latest version of Alibaba's large language model, which the company says competes with leading international models.

Samantha: Alongside the model, there is a new platform called Qwen Cloud. This is designed to make it easier for developers and businesses to build AI applications and so-called agents, which are AI tools that can carry out multi-step tasks on your behalf. Booking a meeting room, drafting a report, pulling data from different systems. That kind of work.

Bob: What makes this a practical find rather than just corporate news is that the Qwen models are available to try directly. If you are a developer, or even a curious user, you can sign up and test the model through Alibaba's cloud interface. There is a free tier for small-scale experimentation, which is the level most people will want to start at.

Samantha: And for anyone based in or near Singapore, Alibaba is partnering with local groups to train more than a thousand small businesses and students in generative and agentic AI. So if you are reading this from Asia, that is a free upskilling opportunity worth checking.

Bob: For European users, the angle is different. Most of us are deep in the OpenAI and Anthropic ecosystems by default. Qwen gives you a credible third option to compare against, and for some tasks, especially anything involving Chinese language or Asian markets, it can actually outperform the Western models. Worth half an hour of your afternoon to set up an account and run a few test prompts.

Samantha: A useful trick: take the same prompt you normally use in ChatGPT or Claude, run it through Qwen, and see where the outputs diverge. That tells you something about each model's biases, which is genuinely useful when you are choosing tools for serious work.

Bob: Just be mindful of where you are sending your data. Qwen Cloud runs on Alibaba infrastructure, so for anything sensitive, the usual data residency questions apply. For experimentation and learning though, it is a low-cost way to broaden your AI toolkit.

Ransomware attacks on cars more than double, with AI accelerating the threat.

Samantha: And our final story is one to take seriously if you drive a connected car, which by now is most new cars sold in Europe. According to the 2026 Global Automotive and Smart Mobility Cybersecurity Report by Upstream, ransomware incidents in the car and smart mobility sector more than doubled in 2025. They now account for nearly half of all recorded cybersecurity incidents in the industry.

Bob: And the report points to AI as part of the reason. AI is letting attackers operate faster, at greater scale, and with more automation. The defending side, meanwhile, is struggling to adapt. The really alarming part is that attackers are getting access to remote command-and-control functions in vehicles. We're talking ignition systems. Door locks. The physical controls.

Samantha: The report describes cases where attackers have demanded payment after seizing those controls. That blurs the line between a traditional cyber incident and a physical safety problem. A locked laptop is a nuisance. A locked car in a motorway service station is a different category of crisis.

Bob: So what can you actually do about it today, as a driver? A few practical things. Keep your car's software updated. Manufacturers push security patches the same way phones do, and many drivers ignore them. Check your vehicle's connected app, accept the updates.

Samantha: Review which apps and services have access to your car. If you connected a third-party service two years ago and never use it, revoke access. Every connection is a potential entry point.

Bob: And use strong, unique passwords on your manufacturer's account. The account that controls your car remotely should not share a password with your old shopping site. Multi-factor authentication, if your carmaker offers it, is worth turning on today.

Samantha: For European regulators, this report adds pressure to the existing UN cybersecurity rules for vehicles that came into force a few years ago. The numbers suggest those rules are not yet keeping pace with the threat. Expect more scrutiny on carmakers in the months ahead.

Samantha: That's it for today. We covered: the seven hundred billion dollar AI infrastructure race, Huawei's Tau Scaling Law, China's travel curbs on AI talent, US federal moves on AI in healthcare, Alibaba's new agentic AI platform, and the doubling of ransomware attacks on connected cars.

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.