作者一年前搬到伦敦，曾被同事讽刺地问会写哪些「大型欧洲科技公司」。他上个月前往荷兰拜访 ASML：这家公司生产制造半导体所必需的晶片制造设备，服务对象涵盖 Nvidia、Apple 等领先企业。执行长 Christophe Fouquet 是受过物理训练的法国工程师，经常与台积电、三星、英特尔等客户高层讨论延伸到 5 年与 10 年的产品路线图，工作核心高度依赖极紫外光（EUV）等硬核技术细节。
Fouquet 直言欧洲在科技产业的位置令人失望：市场与人才都在，但「先监管再行动」的取向，尤其对 AI 的过早规范，正在压抑创新并错失机会。他也批评西方对中国的科技出口管制过度：ASML 被禁止向中国出售最先进机台，只允许输出十多年前卖给西方客户的同代设备。他认为这类限制反而会迫使中国建立本土替代者；面对 14 亿人口的需求，中国不会接受被切断技术。
他谈到 ASML 对法国 AI 新创 Mistral 的 €13 亿（约 $15 亿）投资，强调目的不是打造「欧洲 AI 冠军」或铺路收购，而是把 AI 用来缩短 ASML 的研发周期并改善产品。以资料量来看，一台扫描仪每天可产生数个 TB 的资料，而支援该扫描仪的量测系统也会产生大致相同规模的资料；若能用 AI 驾驭这些资料以提升良率、降低停机，回报可观。他反对直接收购，理由是并购会让对方从专注 AI 转向服务 ASML。
A year after moving to London, the writer was sarcastically asked which “big European tech companies” he would cover. Last month he visited ASML in the Netherlands, whose machines are essential to making semiconductors for Nvidia, Apple, and most market leaders. CEO Christophe Fouquet, a French engineer with a physics master’s, spends heavily technical time with customers like TSMC, Samsung, and Intel, reviewing product road maps that run 5 and 10 years ahead and hinge on EUV and other deep-physics constraints.
Fouquet argues Europe is underperforming despite a huge market and coveted engineers because regulation leads and innovation follows—especially with AI being regulated first. He also criticizes Western export controls aimed at slowing China: ASML cannot ship its most advanced tools to China, and only equipment comparable to what it sold to Western customers more than a decade ago is allowed. He warns such constraints will accelerate Chinese self-sufficiency and competitors; with 1.4 billion people, China will not accept a technology cutoff.
He defended ASML’s €1.3 billion (about $1.5 billion) investment in French AI startup Mistral as a way to use AI inside ASML, not to create a European AI champion or set up an acquisition. He sees AI speeding development cycles and improving customers’ operations: one scanner can generate several terabytes of data per day, and its supporting metrology system produces roughly the same. If AI turns that data into higher yields and less downtime, the payoff is large—yet buying Mistral outright would distract it from AI to serving ASML.