TSMC Delays High-NA EUV: ASML Impact & AI Chip Race

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

Bob: And I'm Bob Russell. Today: Google closing in on Nvidia for the world's most valuable company crown, Beijing pushing its AI governance model across Southeast Asia, Armenia launching a four billion dollar AI mega-hub, TSMC's Arizona packaging push and an EUV delay that hits ASML, the energy companies quietly cashing in on the AI boom, and finally the AI startup playbook for 2026. Let's start with the race to the top of the market cap leaderboard.

Google closes in on Nvidia to become the world's most valuable public company.

Samantha: Alphabet is breathing down Nvidia's neck. According to a Barron's report, Google's market cap is approaching four point six five trillion dollars, and analysts say it could overtake Nvidia within weeks. The shares are up roughly one hundred and forty percent over the past year. That's a level of outperformance we have not seen from Google in over a decade.

Bob: And the interesting part is what's driving it. For two years the AI story was almost entirely a chip story, and Nvidia was the obvious winner. Now investors are betting that the layer above the silicon, the models, the cloud platforms, the actual AI products people use, is where the next leg of value sits. Google has Gemini, it has search, it has YouTube, it has Workspace, and it has its own custom TPU chips. That's a very wide moat.

Samantha: Nvidia is not collapsing here, let's be clear. Their daily gains are still solid, the data center demand is still enormous. But the gap is narrowing fast, and whichever one crosses six trillion first will be a real milestone.

Bob: It also tells you something about market psychology. Picks-and-shovels plays dominate early in a gold rush. As the rush matures, attention shifts to who's actually finding gold. Google is being repriced as a company that turns AI into recurring revenue, not just a company that buys GPUs.

Samantha: For European investors this matters in two ways. First, most pension funds and ETFs in Europe hold significant positions in both names, so the rotation reshapes returns directly. Second, it shifts where European venture capital looks. If applied AI and platform AI are where the value accrues, then European startups building vertical applications get more attention than yet another infrastructure play.

Bob: There's a regulatory wrinkle too. A more dominant Google means renewed scrutiny in Brussels. The Digital Markets Act already designates Google as a gatekeeper, and a four point six five trillion dollar Alphabet weaving AI into search, Android, and Chrome will sharpen every antitrust conversation in Europe.

Samantha: And the bottom line is that the AI economy is consolidating around a very small group of hyperscalers. That's the real story underneath the share price.

China exports its AI governance playbook to Southeast Asia through training and open source.

Bob: Beijing is making a serious diplomatic play around AI. President Xi Jinping is calling for stronger AI governance and for China to keep the initiative in AI development, and that vision is being rolled out across ASEAN through training programs, open-source model sharing, and bilateral cooperation deals.

Samantha: The framing is interesting. China is positioning what it calls its DeepSeek moment, the rise of cheaper, capable open-source models, as proof that there's a real alternative to the American closed-model approach. And for many Southeast Asian governments that pitch lands well, because they don't want to choose sides and they certainly don't want to be locked into one foreign vendor.

Bob: Open source is the lever here. By making capable models freely available and then training local engineers to fine-tune them, Beijing builds dependency without writing huge cheques. It's a softer form of tech diplomacy than the Belt and Road infrastructure model, but arguably more durable, because once your developers are trained on a stack, switching costs get high.

Samantha: And it directly competes with the EU's approach. Brussels has the AI Act, with strict risk tiers, transparency obligations, and fundamental rights protections. China is essentially offering a lighter-touch governance package focused on state stability and economic growth. Two very different visions for the same region.

Bob: For European companies operating in Southeast Asia this gets complicated. If local rules start aligning with Chinese norms on data, content moderation, and model approval, European firms face a compliance split. They build to the AI Act at home, then to a different regime in Jakarta or Bangkok.

Samantha: And there's a standards battle in the background. Whoever sets the technical baselines, evaluation methods, safety benchmarks, audit procedures, ends up shaping the global market. Europe has been a rule-maker before, with GDPR. The question is whether the AI Act travels as well.

Bob: The honest answer is that Europe needs partners. Going it alone with strict rules but no allied bloc is a recipe for irrelevance. Expect more outreach from Brussels to ASEAN, Japan, and Korea in the coming months.

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.

Armenia launches a four billion dollar AI mega-hub powered by Nvidia Blackwell GPUs.

Bob: Armenia is making an audacious bet. A four billion dollar, one hundred and twenty-five megawatt data center is being built in Hrazdan by San Francisco-based Firebird AI, packed with thousands of Nvidia Blackwell GPUs. For a small landlocked country of under three million people, this is a transformational project.

Samantha: It's also a clear pivot. Armenia's tech sector has historically been about IT outsourcing, software services, some fintech. Moving into heavy infrastructure AI, where you compete on power, cooling, and chip access, puts them in a different league.

Bob: The geopolitics is impossible to ignore. The United States is facilitating access to advanced chips and technology, which signals that Yerevan is being pulled westward. That has consequences for Russia, which has long treated the Caucasus as its sphere, and for the broader competition over where AI compute physically lives.

Samantha: One hundred and twenty-five megawatts is also genuinely big. That's enough to power tens of thousands of homes. Building it in Hrazdan, near existing power infrastructure, makes sense, but Armenia will need to scale renewable generation fast if this is the start of more such projects.

Bob: For Europe there are real opportunities. Armenia has an EU partnership agreement, strong ties to the Armenian diaspora across France and other member states, and a well-educated engineering workforce. European researchers and startups could find an attractive compute partner closer to home than the United States, and one that's politically aligned.

Samantha: It also creates an alternative for European firms worried about over-reliance on a handful of American hyperscalers. Sovereign AI capacity in friendly nearby countries is becoming part of the strategic picture, alongside Europe's own data center buildouts.

Bob: The risk is execution. Four billion dollars and thousands of Blackwells is one thing on paper. Delivering, staffing, securing, and actually filling that capacity with paying customers is another. We'll watch how fast Firebird can ramp.

TSMC delays next-gen ASML EUV adoption while expanding Arizona packaging.

Samantha: TSMC is reshaping its capital plan. The world's biggest contract chipmaker is reportedly expanding advanced packaging in Arizona with Amkor for deployment ahead of 2029, while pushing back adoption of ASML's next-generation High-NA EUV lithography until at least 2029.

Bob: Two messages in one decision. On packaging, TSMC is acknowledging that for AI chips, advanced packaging is now as much of a bottleneck as the transistors themselves. CoWoS capacity has been the constraint behind every Nvidia accelerator shortage, so building more of it on US soil with Amkor is both commercial and political.

Samantha: On EUV, the message is cost discipline. High-NA machines from ASML cost around three hundred and fifty million dollars each. TSMC is signalling that current generation EUV plus clever process engineering is enough for the next node, and they'd rather not absorb that capex now.

Bob: That has direct consequences for ASML in the Netherlands. ASML is the only company in the world making these systems, and TSMC is their most important customer. A delay doesn't kill the order, but it pushes revenue out and makes forecasts harder. ASML's share price tends to react sharply to exactly this kind of news.

Samantha: It also raises the question of whether Samsung or Intel might move on High-NA earlier and try to leapfrog. Intel has already taken delivery of High-NA tools. If they execute, the leadership map at the leading edge could shift.

Bob: For Europe the strategic concern is broader. The European Chips Act was meant to bring more semiconductor manufacturing onshore. ASML is the crown jewel of that ecosystem, and its health depends on a small group of customers spread across Taiwan, Korea, and the United States. Concentration risk is real.

Samantha: And the Arizona packaging move shows the geographic shift continuing. Advanced AI silicon is increasingly built and packaged in the US. Europe still has design strength and lithography, but the assembly stages of the AI supply chain are not coming here.

Bob: Which means the next industrial policy debate in Brussels will probably be about packaging and assembly, not just fabs.

Energy and utility firms cash in as AI's power demand explodes toward 85 trillion dollars.

Bob: This one is unglamorous but huge. Companies like Williams in pipelines and Eaton in electrical equipment are reporting strong profits driven by AI infrastructure demand. Some analysts now estimate the total cost of building out global AI infrastructure could hit eighty-five trillion dollars over the long term, a number that dwarfs current global GDP.

Samantha: That eighty-five trillion figure is a long-horizon estimate, not a near-term spend, but it captures the scale of the buildout. The way to think about AI is as a five-layer cake: energy at the bottom, then data centers, then chips, then models, then applications. The base layer is where the physical bottleneck sits right now.

Bob: Williams moves natural gas, which increasingly feeds the gas turbines powering data centers. Eaton makes the switchgear, transformers, and power management gear that every hyperscaler needs by the truckload. Both are seeing record backlogs.

Samantha: For Europe the equation is harder. We have higher energy prices, stricter emissions targets, and slower permitting. Building gigawatt-scale data centers here means going through grid connection queues that are years long in markets like Ireland, the Netherlands, and Germany.

Bob: But European energy companies stand to benefit too. Iberdrola, Engie, RWE, all of them are signing long-term power purchase agreements with cloud operators. Renewables plus storage plus nuclear is the European mix, and the AI buildout is accelerating investment across all three.

Samantha: There's also a societal question. If a single AI campus consumes as much power as a small city, do households pay more, do industrial users get squeezed, do emissions targets slip? Regulators are starting to ask whether AI data centers should face specific efficiency rules.

Bob: Ireland is already there, effectively pausing new data center connections in parts of the grid. Expect that conversation to spread.

Samantha: The takeaway is simple. The AI race is a power race. Whoever has cheap, clean, reliable electricity wins.

The 2026 AI startup playbook reveals what works and what to avoid.

Bob: And to close, something a bit lighter, a kind of field guide for AI founders. A new analysis lays out what's actually working for AI startups in 2026, and what's quietly dying. The good news for serious builders is that the hype phase is over and real patterns have emerged.

Samantha: The winners are vertical AI for regulated professional work, so legal, medical, finance, where deep workflow integration and compliance create defensibility. Then developer tools, anything that makes engineers faster, and voice as infrastructure, embedding voice into call centers, healthcare intake, logistics.

Bob: And the losers list is brutal. Generic ChatGPT-for-X wrappers, plain API resellers with no extra value, and consumer AI hardware. That last one stings, because we've watched several high-profile AI gadgets flop hard. Turns out asking people to wear a new device just to talk to an AI is a tough sell when their phone already does it.

Samantha: The deeper insight is that compute access, control, and governance are now central business risks. If your startup depends entirely on one model provider's API, you don't really own your product. A pricing change or a policy update can wipe out your margins overnight.

Bob: That's why investors are favoring teams that own their data pipeline, fine-tune their own models, or build genuine workflow lock-in. Being a thin layer on top of someone else's model is no longer a venture-scale story.

Samantha: For Europe this is honestly encouraging. Vertical AI in regulated industries plays directly to European strengths. We have world-class legal and medical institutions, sophisticated financial services, and the AI Act actually gives compliant European startups a credibility edge when selling into regulated buyers.

Bob: And voice infrastructure suits Europe's multilingual reality. A startup that can do natural, accurate voice AI across French, German, Italian, Polish, Dutch has a moat American competitors struggle to match.

Samantha: So the message to European founders is, don't try to out-OpenAI OpenAI. Pick a niche, go deep, own the workflow.

Bob: And maybe skip the AI pin.

Samantha: Today we covered: Google closing in on Nvidia, China's AI governance push into Southeast Asia, Armenia's four billion dollar AI mega-hub, TSMC's Arizona packaging and EUV delay, the energy boom behind AI, and the 2026 AI startup playbook.

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.