Nebius is one of the most successful hyperscalers and it seemed to come out of nowhere. But it didn’t — it spun out of Yandex after the 2022 geopolitical chaos, started essentially from scratch with their brand, and built a GPU cloud business that’s now valued at over $4 billion. They went from zero US employees in July 2024 to 40+ in nine months. They’re building a 300-megawatt data center in New Jersey.
Andrei Meganov, their Head of GTM, came to SaaStr AI Annual to share how they did it. Here are the real lessons that apply to any startup, not just if you’re building infrastructure.
And come see 300+ sessions like this at SaaStr AI 2026, May 12-14 in SF Bay.
1. Play to Your Strengths First
Nebius inherited world-class engineers from their Yandex days. So they started with a complex product for sophisticated customers rather than trying to build something accessible for everyone.
This is counterintuitive. Most advice says “start simple, go broad.” But if your team’s superpower is deep engineering, lead with that. If your team’s superpower is sales, lead with that. Build from strength, then expand.
They’re now ranked in the top tier of GPU clouds and used by leading AI companies globally. Engineering-first worked because that’s who they actually were.
2. Your Network Is Your First Pipeline
They leveraged their VC relationships and the Russian-speaking tech community in the Valley to find early adopters. Not random customers. People who would give them real feedback and help refine the product.
This is something I see founders mess up constantly. They want “scale” before they have product-market fit. Nebius went the other way: find people who actually know you, who will tell you the hard truths, who will stick with you through the rough patches.
Then lean into their criticism. The feedback that’s hardest to hear is usually the most useful.
3. Challenge the “Rules” Everyone Accepts
They were told: “You can’t sell overseas capacity to US customers.”
Turns out that’s not true. You can’t sell it for all purposes, but you absolutely can sell training capacity located in Europe to US-based companies. They just had to actually talk to customers instead of accepting received wisdom.
How many constraints in your business are real versus just things everyone assumes? Worth questioning.
4. When You Commit, Commit with Conviction
Once they decided to enter the US market, they moved fast. First data center opened in months. 300MW facility already in progress.
But here’s the nuance: they spent a year on the ground in the Valley before committing to that first data center. Research deeply, then execute with total conviction. Not half-measures.
5. Billboards Actually Work (Sometimes)
I know, I know. We all roll our eyes at billboards. But they actually closed deals from people who saw their billboard on 101 and reached out.
San Francisco is weird. Tech founders drive that highway constantly. They’re thinking about their problems. A well-placed billboard plants a seed.
Even better: they made typos on their swag. People took photos, made fun of them, and it went viral. The lesson? Most publicity is good publicity when you’re building awareness from zero. Own your mistakes. They become your story.
6. Go the Extra Mile for A-Players
They built their US sales and marketing team from scratch. Their approach:
Use in-house recruiters if you can afford it. External recruiters spray and pray. Internal recruiters can focus on exactly the profiles you need.
A-players are expensive and demanding. Lean into their demands. The best people in the industry know their worth. Meet them where they are.
Be ready to change yourself. Research shows most organizations can’t grow headcount more than 30% per year while maintaining culture. If you’re growing faster, your culture will change. That’s not bad. It’s inevitable. Merge the incoming culture with your own and build something new.
7. Manual First, Then Automate Everything
Their process: figure out the right way to do something by hand, then automate it to make it seamless for customers.
Too many startups try to automate before they understand what they’re automating. You can’t automate a process you haven’t figured out yet.
And one more thing: think about where the puck is going. The AI industry right now is, in Andrei’s words, “overblown investment and underblown revenue.” That won’t last forever. Spend some resources preparing for what the industry will look like, not just what it looks like today.
The Meta-Lesson
Nebius borrowed ideas freely. They moved into inference-as-a-service because they saw it working in the market. Their take: your product development doesn’t have to be unique. You just have to do it better than everyone else.
Tech is a time-based advantage game. The more feedback and iterations that go into your product, the better it gets. Start iterating faster than competitors and you’ll catch up and pass them.
One regret they shared: they made their inference product and GPU cloud product entirely separate. That created customer confusion. Don’t cut so many corners that you create fragmentation.
The real story here isn’t about GPU clouds. It’s about a team that found themselves in an impossible situation, played to their strengths, challenged assumptions, and executed with conviction. That playbook works whether you’re building infrastructure, B2B, or anything else.
