Arkady Volozh, founder of Yandex, the Russian Google, on July 16 announced the creation of a new company – called Nebius Group – that aims to build one of the world’s largest commercially available AI infrastructure businesses out of Europe.
Volozh is leading 1,300 employees, mostly ex-Yandex staff, to build Nebius, which, among other things, is developing a cloud computing platform specifically designed to support the training and running of large-scale AI models.
Nebius AI is targeting three groups of clients: companies training foundational models, companies – including large corporates – who need to fine-tune and run AI models, and AI application developers. Nebius’ headquarters and main R&D presence is in Amsterdam, with additional R&D hubs in Europe, North America and Israel. It is already serving customers globally, including Europe’s best-known AI start-ups in France and Germany, according to Volozh. Some 80% of its current clients are based in Silicon Valley.
The plan is to provide hundreds of megawatts of capacity in Europe for training LLMs – mostly in Northern Europe – where electricity is green and cheap- and then build smaller data centers all over the world that are closer to clients that need to run models.
The ability to secure GPUs governs how quickly companies can develop new artificial-intelligence systems. The chips are so valuable that they are delivered to the networking company Cisco by armored car. Demand is outstripping supply. What’s more, access to power is becoming a constraining issue the U.S, as artificial intelligence and data centers and are pushing America’s aging power grid to the brink.
“Demand is high,” Volozh said in an interview with The Innovator. “Whatever we build is booked weeks and months in advance.”
Goldman Sachs research details four waves in AI. The first is AI chips. Nvida, has been the chief beneficiary of a boom in demand for chips that can train and run powerful generative AI models such as OpenAI’s ChatGPT. Its shares have surged 162 % since the start of the year, pushing the chipmaker’s market value above $3 trillion — up 20-fold on the roughly $150bn the company was worth in August 2018. The second wave predicted by Goldman Sachs is AI infrastructure, Nebius’s core business.
“Overcoming the current AI infrastructure deficit is a challenge for the entire industry if the AI revolution is to deliver on its promise,” Volozh said in a press release. “This is something foundational, like the railways in the industrial revolution.”
The global GPU-as-a-service market was estimated at $3.2 billion in 2023 and is projected to reach about $50 billion by 2032, a CAGR of more than 35%, according to Fortune Business Insights. This growth is being driven by increased penetration of machine learning and AI-based applications in business and consumer contexts, leading in turn to increased demand for data center capacity and higher spending on public Cloud. Gartner expects an increase of 20% in worldwide end-user spending on public Cloud services to $675 billion in 2024, primarily as a result of the adoption of Generative AI.
Exit From Russia
The launch of Nebius follows news earlier in the week that Yandex had successfully sold its Russian assets in a $5.4 billion deal, in what constitutes the largest corporate exit from the country since the start of the full-scale invasion of Ukraine over two years ago.
Yandex, which was founded in 1997 and counted U.S. Internet maven Esther Dyson as an early investor, was a rare Russian tech success story. It became known as “The Google of Russia,” given that it sold products broadly similar to its U.S. counterpart including search, e-commerce, advertising, maps, transportation and more. While Yandex’s primary market was Russia, the company went public on the Nasdaq in 2011 via a holding company called Yandex N.V. registered in the Netherlands, followed by a secondary listing three years later on the Moscow Exchange.
Yandex hit a peak market cap of $31 billion in November, 2021. However, in the months that followed, Yandex’s shares plunged as Russia invaded neighboring Ukraine, with the Nasdaq putting a temporary halt on trading before delisting Yandex (alongside several other Russian-affiliated companies).
The EU imposed sanctions on Volozh in 2022 over what it described as Yandex’s complicity in the war. Volozh resigned as chief executive, transferred the voting rights from his controlling stake to the board, and released a statement a year later saying the invasion “is barbaric, and I am categorically against it”. He has regained control of voting rights in Nebius following the split with Yandex.
Yandex N.V. — the parent holding company — has now offloaded all remaining assets linked to Russia. The sale of Yandex’s core Russian assets came after a protracted two-year negotiation that required President Vladimir Putin’s personal approval over an asset the Kremlin considers strategically important. The deal faced various setbacks and hurdles, from Kremlin demands for a discount of at least 50% on foreign asset sales to a nationalization risk.
“Now we are finally free to do something new,” Volozh told The Innovator. It is a moment that Volozh has been preparing for over the last few years from his home base of Israel. He is so confident in the company’s success that he says Nebius could soon go public on the Nasdaq Global Select Market.
Core AI Infrastructure Business
Some 400 of former Yandex AI engineers are focused on developing Nebius’ specialized Cloud infrastructure. With proprietary Cloud software architecture and hardware designed in-house, including servers, racks and data center design, Nebius says it is well-positioned to give AI and ML developers the compute, storage, managed services and tools they need for training inference and data processing.
Nebius primarily competes with other Cloud providers, including Coreweave, Lambda Labs, Together.ai and DeepInfra, as well as hyperscalers such as AWS, Microsoft Azure and Oracle.
Nebius says its key competitive advantages include:
*A sole focus on AI, allowing it to provide solutions precisely tailored to meet the specific AI needs of AI developers and businesses.
*Full control over the whole value creation chain, from in-house server design to proprietary Cloud platform, ensuring maximum productivity and quick scaling, as well as higher cost efficiency and low downtime.
*A team of AI/ML and Cloud engineers with a track record of building Cloud services and infrastructure from scratch.
*20% to 25% lower cost of GPU ownership and operations compared to an average GPU provider.
*A long-standing collaboration with Nvidia across hardware and Cloud which gives it preferential access to the latest and most powerful GPUs. Nebius is a launch partner for Nvidia’s next generation Blackwell platform, which will be available to Nebius’ clients as soon as it is launched.
*More than 15 years of collaboration experience with leading Original Design Manufacturers for co-designing, manufacturing and optimizing cost structure on services.
*A highly energy-efficient data center in Finland that is home to the powerful commercially available supercomputer in Europe , which is among the 20 most powerful in the world. In addition to expanding the capacity of its Finnish data center, Nebius says it plans to build greenfield data centers at new locations primarily in Europe, as well as deploying additional GPU capacity at colocation data centers with modified architecture to reduce installation time.
In addition to its tech and talent advantages Volozh says Nebius has a capital advantage as well. “We can go public on Nasdaq and have access to cheaper capital as a public company,” he says. For investors, this will create an opportunity to invest in a pure-play AI infrastructure stock outside of traditional U.S. Big Tech, he says. For Nebius, which has about $2.5 billion in cash and no debt, being a public company will give it long-term access to capital on favorable terms as it seeks to scale its existing businesses and creates new ones.
Beyond The Cloud
The Nebius Group is also developing three other businesses:. Toloka AI, TripleTen and Avride, which will operate under their separate brands.
Toloka AI provides high-quality training data for Generative AI, focusing on three stages of LLM productin- fine-tuning, alignment and evaluation, which require large columes of human-annotated data. For successful model development, Generative AI needs huge amounts of high-quality data, which needs to be accurate, diverse, legally compliant and produced at scale. This is a complex task and one of the bottlenecks of modern AI development. Toloka tackles this by using highly skilled contractors instead of drawing from a mass audience and is based on a hybrid approach that combines a human backed by LLM instead of purely human input.The company said it currently has more than 5000 expert contractors onboarded and is able to grow capacity by about 1,500 new experts per month.Toloka’s customers include prominent names in Big Tech, AI startups and enterprises who are starting to integrate GenAI solutions.
TripleTen is an edtech platform that capitalizes on growing demand for reskilling and online education to fill the skills gap causes by rapid digitalization. Its offerings include software engineering bootcamps.
Avride develops autonomous driving technology for self-driving cars and delivery robots. The team has been delivering and commercializing projects across North America and Asia for over seven years. It is headquartered in Austin, Texas, with officers in Tel Aviv, Belgrade and Seoul.
IN OTHER NEWS THIS WEEK
INFORMATION TECHNOLOGY
Tech Outage Causes Disruptions Worldwide
One of the biggest-ever IT outages has disrupted companies across the world, from airlines to financial services and media groups, after what should have been a routine software update cascaded into crisis. Thousands of flights were cancelled on Friday, workers in cities from Tokyo to London were unable to log on to their computers, hospital operations were postponed and some television channels went off air. The outage has been blamed on a security update from US group CrowdStrike, which caused a problem with Microsoft’s Windows. PCs and servers are affected.
ARTIFICIAL INTELLIGENCE
Meta Won’t Offer Future Multi-Modal Models In Europe
Meta will withhold its next multimodal AI model — and future ones — from customers in the European Union because of what it says is a lack of clarity from regulators there, according to a an Axios story. Meta’s move highlights a growing conflict between the U.S.-based tech giants and European regulators.
Meta plans to incorporate the new multimodal models, which are able to reason across video, audio, images and text, in a wide range of products, including smartphones and its Meta Ray-Ban smart glasses.Meta says its decision means that European companies will not be able to use the multimodal models even though they are being released under an open license. It could also prevent companies outside of the EU from offering products and services in Europe that make use of the new multimodal models.The company is also planning to release a larger, text-only version of its Llama 3 model soon. That will be made available for customers and companies in the EU, Meta said.
Crusoe, Lancium Team Up to Enable AI and Reduce Impact to the Grid
MOBILITY
Saudi Airline Hands Germany’s Lilium Biggest Order Yet For Flying Taxis
Germany-based air taxi developer Lilium on July 17 clinched its biggest order yet in a deal for Saudi airline Saudia Group to purchase up to 100 of its all-electric flying shuttles, which are still in development.
The state-owned airline made a firm order for 50 of Lilium’s electric Vertical Take-Off and Landing (eVTOL) Jets with an option for an additional 50 later, Lilium said.
The first of the shuttles, which have four-to-six seats and are designed to replace road trips or short hops by aircraft or helicopters, are expected to join Saudia’s fleet in 2026.
To access more of The Innovator’s News In Context articles click here.
