Google Gemma 4, Microsoft Japan AI investment & Chinese AI models surge

Samantha: Welcome to The State of Tech — The European Edition. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: Google drops Gemma 4, its most powerful open-weight AI models yet. Microsoft pours ten billion dollars into Japan for AI and cybersecurity. Chinese AI companies are surging globally with a booming token economy. Data center spending hits astronomical levels as Elon Musk pitches a "Terafab." TSMC's massive revenue growth confirms the AI supercycle is very real. And OpenAI buys a security firm to tackle the risks of agentic AI. Let's start with Google.

Google releases its most capable open-weight AI model family to date — and it could be a game-changer for developers who can't afford proprietary tools.

Samantha: So Google has officially launched Gemma 4, the newest generation of its open-weight AI models. And Bob, the key word here is open-weight. This isn't locked behind an API paywall. Developers can actually download these models and run them locally.

Bob: And what makes this release stand out is the range. They've got smaller models designed for edge devices — think laptops, phones, embedded systems — all the way up to large models for serious compute workloads. Google is calling it "unprecedented intelligence-per-parameter," which is marketing speak, sure, but the benchmarks back it up. These models are built on the same research foundation as Gemini 3, their flagship proprietary system. So you're getting frontier-level reasoning and code generation in a package that doesn't require a massive server farm.

Samantha: NVIDIA jumped in immediately too. They've announced optimizations for Gemma 4 across their entire GPU lineup, from data center chips down to RTX desktop cards. So if you've got a reasonably modern NVIDIA GPU, you can run these models efficiently on your own hardware.

Bob: The bottom line for Europe is pretty significant. European startups and research institutions often operate on tighter budgets than their Silicon Valley counterparts. Open-weight models like Gemma 4 dramatically lower the barrier to entry. A research lab in Munich or a startup in Lisbon can now access capabilities that were exclusive to well-funded companies just eighteen months ago. And running models locally? That's a data sovereignty advantage too. You keep your data on your own infrastructure, which aligns neatly with European privacy expectations.

Samantha: It also creates competitive pressure on proprietary model providers to justify their pricing. More options, more competition — that's healthy for everyone building AI products in Europe right now.

Microsoft bets big on Japan with a ten-billion-dollar commitment that signals how global AI infrastructure is being strategically distributed.

Bob: Let's talk about Microsoft and Japan. Microsoft has announced a 1.6 trillion yen investment — roughly ten billion US dollars — to be spent between 2026 and 2029. The focus is AI infrastructure and cybersecurity cooperation with the Japanese government.

Samantha: And there's a workforce component too. Microsoft is committing to training one million engineers and developers in Japan by 2030. That's not a small number. This follows similar announcements they've made in Singapore and Thailand, so there's a clear pattern here. Microsoft is building a global web of AI infrastructure partnerships with national governments.

Bob: What's interesting is the cybersecurity angle. This isn't just about Azure data centers. Microsoft is positioning itself as a strategic national security partner. They're offering AI infrastructure and cyber defense as a bundled package, essentially making themselves indispensable to government operations.

Samantha: And here's where Europeans should pay attention. When Microsoft builds out massive infrastructure in Japan and Southeast Asia, that strengthens the resilience and capacity of its global cloud network, which European businesses rely on heavily. But it also raises a question. Europe has been talking about digital sovereignty and building its own AI capacity for years now. Meanwhile, Microsoft is locking in deep partnerships across Asia at remarkable speed.

Bob: The precedent this sets is important. If governments in Asia are partnering directly with American tech giants for AI and cybersecurity, European leaders need to think carefully about their own strategy. Do they pursue similar partnerships? Do they invest more aggressively in homegrown alternatives? Probably both. But the clock is ticking, and these bilateral deals are reshaping the global AI landscape in real time.

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.

Chinese AI companies are quietly building massive global user bases, and the numbers are getting hard to ignore.

Bob: Chinese AI models are expanding their global footprint fast. Companies like MiniMax and Zhipu AI are now ranking among the top globally by token consumption — that's essentially a measure of how much AI processing power their models are burning through.

Samantha: And what's driving this is the rise of agent-based AI. There's a system called OpenClaw that lets users delegate complex, multi-step tasks to AI assistants. You're not just asking a chatbot a question anymore — you're handing it a workflow. That dramatically increases the number of tokens consumed per interaction, and it's pushing these Chinese platforms up the global rankings.

Bob: The scale is remarkable. We're seeing a fundamental shift in where AI consumption happens and which companies are serving it. For a long time, the narrative was dominated by OpenAI, Google, Anthropic — the usual Western players. Now Chinese companies are competing head-to-head on the global stage, and in some metrics, they're winning.

Samantha: For Europe, this is a complicated picture. On one hand, more competition means more choices and potentially better pricing for European businesses looking to integrate AI. On the other hand, European regulators are already grappling with the AI Act and data protection. Adding widely adopted Chinese models into the mix creates new questions about data privacy, where information flows, and who has access to it.

Bob: There's also a geopolitical dimension that's hard to ignore. European policymakers have been cautious about dependency on American tech platforms. Widespread adoption of Chinese AI models adds another layer of complexity to that conversation. It's not just about choosing between European and American technology anymore — there's a third major player at the table, and they're moving aggressively.

Samantha: And let's be honest, most European businesses will care primarily about performance and cost. If a Chinese model does the job better or cheaper, adoption will happen whether regulators are ready or not.

The world is spending staggering amounts on data centers, and Elon Musk wants to build semiconductor capacity on a scale that sounds like science fiction.

Bob: Data center spending has gone completely vertical. US hyperscalers alone are projected to spend around seven hundred billion dollars this year. That's nearly six times what they spent in 2022. Google is building a clean energy-powered facility in Texas. Meta has revised its investment in El Paso to ten billion dollars. The numbers just keep climbing.

Samantha: And then there's Elon Musk, who never does anything small. He's proposed something called a "Terafab" — a semiconductor manufacturing initiative in Austin designed to support one terawatt of compute capacity annually. One terawatt. That's an almost absurd scale, but given the trajectory of AI demand, maybe not as absurd as it sounds.

Bob: The challenge for Europe is catching up. European data center capacity has been growing, but nowhere near the pace of the US. Permitting processes are slower. Energy grids need significant upgrades. Clean energy supply is a bottleneck in many regions. If Europe wants to host meaningful AI workloads domestically rather than depending entirely on US infrastructure, these issues need urgent attention.

Samantha: And it's not just about building the buildings. You need the power, the cooling, the network connectivity, and the skilled workforce to operate these facilities. Every one of those is a constraint that European countries are dealing with right now. The gap between ambition and execution is real.

Bob: What projects like the Terafab signal is that the biggest players are planning for a future where AI compute demand is orders of magnitude beyond where it is today. Europe needs to be honest about whether its current infrastructure trajectory meets that future. Because if it doesn't, the continent risks becoming a consumer of AI rather than a producer.

TSMC's stellar revenue forecast tells us the AI chip boom isn't slowing down — it's accelerating.

Samantha: TSMC is projecting thirty-eight percent revenue growth year-over-year for the first quarter of 2026. That is a massive number for a company of that size. And it's being driven by their 3nm and 5nm chips, plus the early ramp-up of their next-generation 2nm technology using Gate-All-Around transistor architecture.

Bob: This is the clearest signal that the so-called AI supercycle is not hype — it's real and intensifying. TSMC makes the most advanced chips on the planet. Their client list includes NVIDIA, Apple, AMD, and virtually every major hyperscaler. When TSMC posts numbers like this, it means demand for cutting-edge silicon is relentless across the entire tech ecosystem.

Samantha: And they're maintaining high gross margins despite rising costs, which tells you their pricing power is enormous. Nobody else can manufacture at these nodes at scale. Samsung is trying, Intel is trying, but TSMC remains the dominant force.

Bob: For European tech companies, this is both good and concerning. Good because TSMC's continued investment ensures the chips powering European AI innovation keep getting better and more available. Concerning because Europe has essentially zero capacity to manufacture at these advanced nodes. The EU Chips Act is working to change that, but building fabs takes years. In the meantime, Europe's entire AI hardware supply chain runs through Taiwan, and that geopolitical concentration of risk hasn't gone away.

Samantha: It's one of those situations where the technology news is fantastic but the strategic vulnerability is sobering.

OpenAI acquires a security testing firm, acknowledging that as AI agents get more powerful, the risks get more serious.

Bob: OpenAI has acquired Promptfoo, a company that specializes in security testing for agentic AI systems. This is about finding and fixing vulnerabilities in AI that can autonomously execute complex, multi-step tasks.

Samantha: And this makes total sense when you think about where AI is heading. We just talked about agent-based systems driving the token economy. As these agents get more capable — booking flights, managing code deployments, handling financial transactions — the attack surface grows exponentially. A prompt injection vulnerability in a chatbot is annoying. A prompt injection in an AI agent managing your cloud infrastructure is catastrophic.

Bob: Promptfoo built tools specifically for identifying these kinds of risks. Bringing that expertise in-house means OpenAI can integrate security testing directly into its development pipeline rather than treating it as an afterthought. It's a signal that the industry is maturing — security is no longer optional, it's foundational.

Samantha: The European angle here ties directly into the AI Act. European regulators have been ahead of the curve on demanding accountability and safety in AI systems. This acquisition validates that approach. If the biggest AI company in the world is buying security firms, it confirms that the risks Europe has been legislating around are very real.

Bob: And as European businesses adopt more agentic AI tools — from customer service to supply chain management — they'll need assurance that these systems are secure. Expect European demand for AI security auditing and certification to grow significantly. This could actually become a competitive niche for European cybersecurity firms who understand both the technology and the regulatory landscape.

Samantha: Today we covered: Google's Gemma 4 open-weight launch, Microsoft's ten-billion-dollar Japan investment, the global rise of Chinese AI models, skyrocketing data center spending and Musk's Terafab, TSMC's blockbuster revenue growth, and OpenAI's move into AI security with the Promptfoo acquisition.

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.