On October 21, a new platform was opened to Nasdaq traders: NBIS, a spin-off of Nebius, an emerging player in the AI cloud infrastructure space.
Casual observers could be forgiven for wondering where this company was coming from, as there has been little in the way of the usual fanfare that surrounds most startups’ journeys to IPO – no roadshow; no horn tootin’; no confetti-filled celebrations; nothing, not a peep. That’s because Nebius is an unusual beast: a public company, but a startup in almost every sense of the word.
Nebius has actually been public for 13 years, floating in May 2011 as Yandex NV – the Dutch company that owns the Russian internet giant Yandex (often referred to as the “Google of Russia”). At the end of 2021, Yandex NV reached a peak value of 31 billion dollars, but after the Russian invasion of Ukraine at the beginning of 2022, everything changed. Nasdaq suspended trading in Yandex NV shares in February due to sanctions imposed on Russian-linked companies, and a year later Nasdaq said it would delist Yandex entirely. But Yandex successfully appealed on the grounds that it was restructuring – a process that would take an additional 16 months to fully complete.
Part of this included divesting all of its Russian assets, which was where the real value of the business lay. What remained under the ownership of Yandex NV was a random assortment of infrastructure and business units that happened to be located outside of Russia. This rollout was completed in July, when Yandex NV changed its name to Nebius AI, an AI cloud platform complete with its Finnish data center.
The new venture would be led by Arkady Volozh (pictured above), the Russian Yandex co-founder and former CEO who was removed from Europe’s sanctions list in March after publicly criticizing Russia’s invasion of Ukraine.
Nebius’ core business is selling GPUs (graphics processing units) as a “service” to companies that need “computing” – that is, processing power and resources to perform computing tasks such as running algorithms and building machine learning models. Last month, the company released a complete cloud computing platform designed for the “full machine learning lifecycle,” which includes data processing, training, optimization, and interpretation.
With the restructuring complete, and Volozh free to run the show from the company’s new HQ in the Netherlands, Nasdaq green-lighted Nebius to resume trading last month. The situation was unprecedented, however: a public company whose trading was temporarily suspended, then restarted almost three years later under a new name and a completely different business proposition?
In many ways, it would have made sense to have already delisted and grown with private equity, a good old-fashioned way to start. But as Volozh explained to TechCrunch earlier this year, building infrastructure costs a lot of money, and the easiest and cheapest way to access money in what is currently one of the hottest areas in technology is to use the public markets. But it’s never been certain how the public markets will react to this strange new entity. No one really knew what to expect.
The moon is in, and Nebius enjoys a somewhat gentle re-entry into public life; it fell sharply to $18 billion before the February 2022 trading halt, which was expected, and has since reached between $3.5 billion and $4.75 billion, with some signs of stabilization.
“We couldn’t predict what would happen, whether it would be $5 per share, or whether it would be $50 per share — this has never happened before, no one really knows how to do it,” Volozh told TechCrunch in an interview in London this month. . “It is still changing but it is stable, the good thing is that it is more stable than the cost of goods, which means that the market believes that we will be able to build a business here. How big the business is, we will see.”
Nebius competes with all the usual hyperscaler cloud behemoths, though arguably its direct competitors are other cloud startups like CoreWeave, which raised a ton of money this year. As CoreWeave is in the midst of expanding from the US to Europe, Nebius is going in a different direction, announcing plans this week to expand its presence in the US with a new GPU cluster in Kansas City (on the Missouri side) scheduled to go live. in early 2025. The company has also opened “customer hubs” in San Francisco and Dallas, with a third planned in New York by the end of the year.
But while the cloud infrastructure business is its bread and butter (accounting for two-thirds of its profits, according to its first earnings report last month), there is a triumvirate of additional businesses under the Nebius Group umbrella. These include a self-driving car company called Avride, based in Texas; a Swiss AI and LLM manufacturing company called Toloka; and edtech platform TripleTen, based in Wyoming.
Time to drive
Avride is part of the international division of Yandex’s self-driving unit, which exited the joint venture with Uber in 2020. While Alphabet’s Waymo is now a leader in the burgeoning robotics space, recently receiving a $45 billion valuation, Yandex was the first startup. trailblazer in Russia, Volozh noting that the company has been beating Waymo will introduce the first fully autonomous cars on public roads, before the war put the kibosh on the plans.
“See [Yandex] “They were going to launch the first taxis on public roads without anyone driving, in a real city (Moscow), a few months before Waymo launched in San Francisco,” said Volozh. “Journalists were invited to a big event on March 22, but that didn’t happen. People had to pack up all their things and leave after a few weeks.”
The team working on Yandex’s autonomous car project switched to Avride, a new brand it launched last year, eventually moving to Austin via Tel Aviv.
“These are the same 250 people,” added Volozh.
Last month, Avride announced a significant multi-year partnership with Uber, which saw Avride’s food-delivery robots live at Uber Eats starting in Austin, though the partnership will also bring Avride’s self-driving cars to Uber’s premises later (Uber signed other similar deals, including Google’s sister company Waymo).
While Yandex has pockets deep enough to finance autonomous car projects, Nebius doesn’t — it has billions of dollars in the bank from its divestment in Russia, and is focused on building its cloud infrastructure business. And that’s why Volozh says Avride will need to find more partners over time.
“They have enough budget for this year and next year,” said Volozh. “We finance them, but they must use this time to find new partners, because it costs a lot of money to build cars. It needs real investment.”
Obvious partners could include automakers, but it could be any company ready to invest billions, Volozh added, adding that he would be willing to cede control to Avride if needed.
Toloka, on the other hand, is a platform that focuses on data labeling and quality control of large-scale linguistic models (LLMs) and related AI systems – very similar to Scale AI, which was recently valued at more than 13 billion dollars. Toloka has clear synergies with Nebius’ core infrastructure business, but the customers are not the same. Nebius works mainly with productive AI startups looking for computing, while Toloka works with large companies like Amazon and Hugging Face looking to develop their LLMs.
Both Toloka and Avride could end up following the same path as ClickHouse, the creators of an anonymous open source management system that spun off from Yandex in 2021. While ClickHouse’s commercial organization secured big-name backers like Index Ventures, Benchmark Capital, and Coatue, Nebius retained a small stake.
“ClickHouse became very popular, and we were asked by investment funds to build a business about an open source project. Now they have income, and they are growing,” said Volozh.
TripleTen, on the other hand, is something of an outlier in the Nebius group of businesses, because it’s pretty much a direct-to-consumer product that offers online coding bootcamps for those aspiring to transition into the tech industry. One idea Nebius discussed is to position itself as a “bundle of services” provider to AI companies, from data centers and GPU infrastructure, to education. And this highlights the situation Nebius finds himself in: He’s drawing lines between the different businesses he’s left with, and trying to make sense of it all.
For now, TripleTen is failing, and Volozh admits that it won’t be the major revenue driver its infrastructure business is – but it has the potential to provide meaningful revenue and will remain part of the Nebius Group.
“Nebius is a billion-dollar business,” Volozh said. “TripleTen – it’s a good model, but maybe a business of tens or hundreds of millions of dollars. It is not a multi-billion dollar business.”
Parallel computing
As for Nebius AI’s core cloud business, the company already has its own fully managed data center in Finland, with plans to triple its capacity to 75 megawatts. In parallel, the company is building more sites in the conglomerates, a move designed not only to increase its capacity, but also to reduce latency by bringing processing closer to its customers. In addition to the Kansas location announced this week, Nebius had already unveiled a new GPU cluster in Paris that went online this month.
In the future, Nebius plans to build its other data centers, both in Europe and in the US, but given the time it takes, it is faster to close the gap with collaborative areas, which is why it is moving forward in a hybrid way.
“It works well if we build it ourselves, but building means a year and a half or two years – it’s a long process, and we can’t wait,” said Volozh. “That’s why we have these joint locations in Paris and Kansas City.”
