字号·· | 护眼
thenextweb

Nebius首席技术官Danila Shtan:执行成本低廉的当下,创意再度重要Nebius CTO Danila Shtan says ideas matter again now that execution is cheap

点「原文对照」整页切到原文,或双击某段只看那段的原文。

TNW的克里斯蒂安·迪纳(左)与Nebius首席技术官丹尼拉·什坦在阿姆斯特丹举行的HumanX活动上,2026年9月24日。图片来源:TNWTNW’s Cristian Dina (left) and Nebius CTO Danila Shtan at HumanX in Amsterdam, 24 September 2026 Credit: TNW

当被问及一家公司像 Nebius 这样快速增长时,什么会最先崩溃,CTO Danila Shtan 给出了一个单字回答:一切。他告诉 TNW,规模和增长速度时刻考验着他团队构建的一切。

Asked what breaks first when a company grows as fast as Nebius, Danila Shtan gave a one-word answer: everything. The scale and the pace of growth test everything his team builds, all the time, the Nebius CTO told TNW.

“你开始构建,对增长、客户需求或他们所需的性能有一些预期,但它们每天都会让你感到意外。” Nebius CTO Danila Shtan 说道。

“You start building, you have some expectations about the growth or the client’s needs or the performance they need, and they come to surprise you every day.”Danila Shtan, CTO of Nebius Shtan described Nebius as a Dutch company and one of the largest AI clouds in the world. He spoke to Cristian Dina for the TNW Podcast at HumanX in Amsterdam on 24 September. That day he also joined a fireside chat on the main stage about making AI agents work at scale.

Shtan 将 Nebius 描述为一家荷兰公司,也是全球最大的 AI 云服务商之一。他在 9 月 24 日于阿姆斯特丹举行的 HumanX 大会上,接受了 Cristian Dina 为 TNW Podcast 进行的采访。当天,他还登上主舞台参加了一场炉边谈话,探讨如何让 AI Agent 大规模落地应用。

Earlier episodes featured Alexander Rinke of Celonis, Kristin Fracchia of Gamma, and Daniel Dines of UiPath. The full conversation is in the video below.

此前几期节目的嘉宾分别是 Celonis 的 Alexander Rinke、Gamma 的 Kristin Fracchia 以及 UiPath 的 Daniel Dines。完整对话收录在下方视频中。从 50 美元到数亿美元Shtan 表示,Nebius 为训练和运行 AI 模型、以及在模型基础上构建应用提供基础设施。他负责软件开发,因此其团队构建运行在服务器上的软件。

From $50 to hundreds of millions Nebius provides the infrastructure for training and running AI models and for building applications on them, Shtan said. He leads software development, so his team builds the software that runs on the servers.

“它们需要接入高性能计算,这本质上是当前 AI 革命、炒作、无论你怎么称呼它的核心、血液和一切。” Nebius 的 Danila Shtan 说道。

“They need access to high performance computing, which is essentially the heart and the blood and everything of the current AI revolution, hype, whatever you call it.”Danila Shtan, Nebius Some customers spend hundreds of millions on the platform, he said. Others spend $50 to run a small slice of compute for half an hour of internal experiments.

他透露,一些客户在平台上花费数亿美元,另一些客户则花费 50 美元,运行一小块算力,进行半小时的内部实验。与人交谈,重写代码Shtan 称,他的团队约有 500 到 550 人。他的一天主要花在与他们交谈上。他会检查项目进展,并在出现严重问题时给予反馈。Nebius 构建平台,同时也建设数据中心,他称后者为一个并行的组织。

Talking to people, and coding again Shtan’s team is about 500 to 550 people. His day is mostly talking to them, he said. He checks on projects and gives feedback when something goes very wrong. Nebius builds a platform, but it also builds data centres, which he called a parallel organisation.

AI 编程 Agent 让他重新开始写代码。他可以在几小时或几天内测试一个想法,向团队展示一个粗糙的原型,而不必只靠口头讲解。他表示,他的思维过程一直比实现技能快,而 Agent 现在承担了那部分工作。

AI coding agents have him building again. He can test an idea in hours or days and show the team a rough prototype instead of talking them through it. His thought process was always faster than his implementation skills, he said, and agents now take on that part.

主权问题有多个角度。什坦表示,Nebius相当一部分客户是美国公司。美国客户使用欧洲的算力,欧洲客户使用美国的算力。

Sovereignty has many angles A significant portion of Nebius’s customers are American companies, Shtan said. American customers use European capacity, and European customers use American capacity.

他说,欧洲各国政府把主权当作基础设施来谈。美国政府谈的是模型和能力。大公司谈的是不依赖单一模型供应商。他称这些角度都是正确的。

European governments talk about sovereignty as infrastructure, he said. The US government talks about models and capabilities. Large companies talk about not depending on a single model vendor. He called all of these angles correct.

“我觉得有时候它们被过度炒作了,这对谁都没好处。对生意也没好处。”达尼拉·什坦,Nebius独立参与者什坦表示,Nebius希望成为世界上最重要的独立AI云公司之一。如今AI算力很大一部分掌握在垂直整合的参与者手中,它们拥有芯片、数据中心、模型和应用。他希望在每一层都有替代方案,包括模型供应商和消费级应用。

“I feel like sometimes they’re over-hyped, and this is not good for anyone. It’s not good for business.”Danila Shtan, Nebius An independent player Nebius wants to be one of the most important independent AI cloud companies in the world, Shtan said. Much of today’s AI capacity sits with vertical players that own the chips, the data centres, the models and the apps. He wants alternatives at every layer, including model vendors and consumer apps.

“所以我们完全支持多元化,我们完全支持打造这一独立的AI层,因为权力过度集中在一家公司手中,即便那是一家好公司,也从来不是好主意。”达尼拉·什坦,Nebius他说,公司最大的挑战是增长。对大模型、小模型和专用模型的需求持续上升,瓶颈不再只是GPU,还包括内存、磁盘、服务器、电力、土地和高压设备。

“So we are all for diversity, and we are all for, like, building this independent layer of AI because too much power in one company’s hands, even if it’s a good company is never a good idea.”Danila Shtan, Nebius Its biggest challenge is growth, he said. Demand keeps rising for large, small and specialised models, and the bottleneck is no longer only GPUs. It is also memory, disks, servers, electricity, land and high-voltage equipment.

想法再次重要起来什坦担任工程经理约25年。他从软件开发者做起,很快转向人员管理。他先后在仓储自动化、零售、金融科技、生物科技、一家新闻媒体和一个分类广告网站工作过。后来他转向开发者工具、云和Nebius。

Ideas matter again Shtan has been an engineering manager for about 25 years. He started as a software developer and moved quickly into managing people. He has worked in warehouse automation, retail, fintech, biotech, a news outlet and a classifieds website. Later he moved into developer tooling, clouds and Nebius.

他说,对年轻的创业者来说,这是一个很好的时代,因为推出一款成功的产品不再需要庞大的团队。如今更重要的是产品品味和把它卖出去的能力。他说,五年前的信条是想法不重要,执行才重要。

For young builders, he said, this is a great time, because a large team is no longer needed to ship a successful product. Taste in product and the ability to sell it now matter more. Five years ago, he said, the maxim was that ideas don’t matter and execution does.

“现在,执行成本相当低廉,所以创意再次变得重要。”Nebius 的 Danila Shtan 说。他说,企业在 AI 上失败,是因为把它当成魔法。他的建议是了解其局限,进行实验,失败后再尝试。

“Now, execution is quite cheap, so ideas matter again.”Danila Shtan, Nebius Companies fail at AI when they treat it as magic, he said. His advice is to learn its limits, experiment, fail and try again.

“AI 不是什么神奇的东西。它是一种工具。”Nebius 的 Danila Shtan 说。

“AI is not a magical thing. It’s a tool.”Danila Shtan, Nebius

阅读媒体原文 ↗