The Next Web TNW

German startup Marvel Fusion and Colorado State University have broken ground on a $150M laser facility in a bid to commercialise fusion energy. Dubbed ATLAS, the facility will use three ultra-high intensity lasers to fire 7 petawatts of power — over 5,000 times the electrical generation capacity of the US — at a target roughly the width of a human hair.   The blast will last approximately 100 quadrillionths of a second. However, it will produce enough heat and pressure to fuse atoms together, initiating the same reaction that powers the Sun and stars. For decades, scientists have been experimenting with lasers to create fusion reactions. A major breakthrough came in 2022, when the US government’s National Ignition Facility (NIF) successfully achieved a first in the sector: a net energy gain in a fusion reaction.  Calling all Scaleup founders! Join the Soonicorn Summit on November 28 in Amsterdam. Meet with the leaders of Picnic, Miro, Carbon Equity and more during this exclusive event dedicated to Scaleup Founders! Net energy gain essentially means that the reaction produced more energy than went into it. The feat ignited hopes that fusion’s promise of abundant, clean, and limitless energy may not be too far off. However, there’s a huge difference between achieving net energy gain and making a commercial fusion power plant that generates continuous clean energy. For that you’d need to produce these fusion blasts several times per second.   But that’s exactly what Marvel Fusion plans to do.  Commercial fusion reactors ATLAS, located at the university’s campus in Fort Collins, will aim to repeat the laser blasts ten times per second. That will be enough to generate an ongoing fusion reaction and, hopefully, a stable supply of clean energy.  The facility will be similar to NIF, but use cutting-edge technology designed to improve the laser’s power and efficiency, while bringing down costs.  Colorado State University will develop one of the lasers. Marvel will build the other two, in a bid to prove its core technology. The partners aim to complete the laser facility in 2026.  “This groundbreaking (feels that a word’s missing) marks an exciting new chapter in the partnership between Marvel Fusion and Colorado State University as we move forward with constructing a facility that will drive the future of fusion energy,” said Heike Freund, chief operating officer at Marvel Fusion.   While Marvel Fusion has established a subsidiary in Colorado to support this collaboration, the company’s headquarter remains in Munich, Germany.  When asked why he chose the US, Marvel’s CEO Moritz von der Linden previously told the Financial Times that it was the “fastest, most capital-efficient way for us to move on building this facility.” There is simply more funding and an appetite for this kind of technology across the pond, he said.  Nevertheless, he doesn’t necessarily intend on building a full-scale commercial plant in the US. “It could very well, maybe hopefully, be in Europe,” he said.  Marvel recently secured €62.8mn in a Series B funding round, bringing its total raised  to date to €100mn.  Alongside fusion energy, ATLAS will also support research in medicine, semiconductors, and X-ray imaging. For instance, lasers could be used to deposit energy in a very localised region for tumour treatment.  Both Colorado State University and Marvel will fund the construction of the new facility. The US government has also put $28mn toward the project. source

Debates about AI are everywhere these days. Families are chatting about its impacts on their lives. Politicians are deliberating over the laws that oversee it. Workers are talking about the risks of job automation. Entrepreneurs are chewing over the business opportunities. And the tech world is discussing what everyone else will be discussing next. Unfortunately, these conversations have also attracted countless scoundrels. They join the chat with wild promises and heavy doses of AI snake oil. But behind their breathless hype, remarkable innovations are emerging. Don’t believe me? Well, just take a look at this week’s Nobel Prize winners.  Calling all Scaleup founders! Join the Soonicorn Summit on November 28 in Amsterdam. Meet with the leaders of Picnic, Miro, Carbon Equity and more during this exclusive event dedicated to Scaleup Founders! On Tuesday, a Nobel was awarded to AI pioneers for the first time. Just a day later, a Nobel was awarded to AI pioneers for the second time. These scientific achievements are a taste of what’s to come. AI is rapidly approaching an inflection point. “This goes beyond just the latest gadgets or GenAI breakthroughs,” says Chris Duffey, the head of AI innovation and strategy at Adobe. “It’s about a fundamental shift in how we think, create, and work.” Duffey has a unique perspective on these development. Alongside his work at Adobe, he’s written an award-winning book about AI and become a renowned futurist. He’s also a member of TNW’s advisory board. The board have been having their own debates about artificial intelligence. Their discussions led AI and Deeptech to become one of six themes for TNW Conference 2025. A gateway to the future of AI Over two jam-packed days in Amsterdam next June, we’ll explore the innovations set to shake up the world. Tech luminaries will be dissecting and demoing breakthroughs in AI, quantum, robotics, and so much more. Expect a mix of mind-bending panels, keynotes from digital leaders, and exclusive tips. Debates about AI will permeate the event — and we want you to join them. “The innovators who gather here are poised to challenge the status quo, to push boundaries in ways that ripple far beyond technology itself,” Duffey says. “What happens next isn’t inevitable, it’s the result of deliberate key moments, decisions, and ideas that will change everything.” To keep you on the right track, the AI and Deeptech theme is underpinned by four pillars: AI Revolution: Explore how AI is transforming different industries Leaping into Quantum: Discover how businesses can tap the power of quantum and why investors should pay close attention Rise of the Machines: Find out which cutting-edge robotics trends you should keep on your radar The Cyber Arms Race: What are the latest cybersecurity technologies you should know about? And what are the ethical dilemmas when it comes to securing your business’ systems against threats? Duffey has big expectations for the programme. “We’re not just predicting the future,” he says. “We’re inventing it, one groundbreaking idea at a time.” If you want to hear all about these ideas, you can grab a ticket forTNW Conference now. Use the code TNWXMEDIA2025 to get 30% off your pass. See you on June 19 and 20 in Amsterdam! source

Danish startup Landfolk has launched a visual search engine for holiday homes. Named Daisy, the feature invites users to search for their dream destinations. They can enter endless prompts, from “a luxurious room overlooking the ocean” to “a cosy cabin in the mountains.” After ingesting a query, the feature spits back images linked to properties on the platform. Users can then click-through for further details on the home. Landfolk believes the tool is the first visual search engine in the travel industry. The startup expects it to inspire not only holiday choices, but also home design ideas. Chris Sørensen, Landfolk’s co-founder and CTO, said the search engine harnesses a multi-modal approach. Calling all Scaleup founders! Join the Soonicorn Summit on November 28 in Amsterdam. Meet with the leaders of Picnic, Miro, Carbon Equity and more during this exclusive event dedicated to Scaleup Founders! “We use vision and language models trained on our listing images,” he told TNW. “This setup allows us to do image-to-image or text-to-image similarity searches without the need for us to extensively tag our images.” Landfolk’s holiday homes plan Landfolk was founded in 2021 by seven former colleagues at Airbnb. The startup introduced a new marketplace for premium holiday homes. Each home is handpicked based on four criteria: aesthetics, personality, quality, and location.  After launching in Denmark, the platform expanded across Scandinavia and Germany. The next target is southern Europe. In July, the startup raised €10.3mn in Series A funding. The company said the investment would “fuel Landfolk’s commitment to innovation.” The new search engine is testament to that ambition. Landform now plans to built on this machine learning foundation. Sørensen envisions AI finding the best seasonal cover photos and curating collections of holiday homes. “We also believe this technology will assist us internally, making it easier to acquire new homes that match Landfolk’s unique vibe, and to sort images according to customer preferences,” he said. source

The walking, talking, dancing Optimus robots at the recent Tesla demonstration generated huge excitement. But this turned to disappointment as it became apparent that much of what was happening was actually being controlled remotely by humans. As much as this might still be a fascinating glimpse of the future, it’s not the first time that robots have turned out to be a little too good to be true. Take Sophia, for instance, the robot created by Texas-based Hanson Robotics back in 2016. She was presented by the company as essentially an intelligent being, prompting numerous tech specialists to call this out as well beyond our capabilities at the time. Similarly we’ve seen carefully choreographed videos of pre-scripted action sequences like Boston Dynamics’ Atlas gymnastics, the English-made Ameca robot “waking up”, and most recently Tesla’s Optimus in the factory. Obviously these are still impressive in different ways, but they’re nowhere near the complete sentient package. Let Optimus or Atlas loose in a random home and you’d see something very different. A humanoid robot capable of working in our homes needs to be capable of doing many different tasks, using our tools, navigating our environments and communicating with us like a human. If you thought this was just a year or two away, you’re going to be disappointed. Building robots able to interact and carry out complex tasks in our homes and streets is still a huge challenge. Designing them even to do one specific task well, such as opening a door, is phenomenally difficult. There are so many door handles with different shapes, weights and materials, not to mention the complexity of dealing with unforeseen circumstances such as a locked door or objects blocking the way. Developers have actually now created a door-opening robot, but robots that can deal with hundreds of everyday tasks are still some way off. Behind the curtain The Tesla demonstration’s “Wizard of Oz” remote operation technique is a commonly used control method in this field, giving researchers a benchmark against which to test their real advances. Known as telemetric control, this has been around for some time, and is becoming more advanced. One of the authors of this article, Carl Strathearn, was at a conference in Japan earlier this year, where a keynote speaker from one of the top robotics labs demonstrated an advanced telemetrics system. It allowed a single human to simultaneously operate many humanoid robots semi-autonomously, using pre-scripted movements, conversation prompts and computerised speech. Clearly, this is very useful technology. Telemetric systems are used to control robots working in dangerous environments, disability healthcare and even in outer space. But the reason why a human is still at the helm is because even the most advanced humanoid robots, such as Atlas, are not yet reliable enough to operate completely independently in the real world. Another major problem is what we can call social AI. Leading generative AI programs such as DeepMind’s Gemini and OpenAI’s GPT-4 Vision may be a foundation for creative autonomous AI systems for humanoid robots in the future. But we should not be misled into believing that such models mean that a robot is now capable of functioning well in the real world. Interpreting information and problem solving like a human requires much more than just recognising words, classifying objects and generating speech. It requires a deeper contextual understanding of people, objects and environments – in other words, common sense. To explore what is currently possible, we recently completed a research project called Common Sense Enhanced Language and Vision (CiViL). We equipped a robot called Euclid with commonsense knowledge as part of a generative AI vision and language system to assist people in preparing recipes. To do this, we had to create commonsense knowledge databases using real-world problem-solving examples enacted by students. Euclid could explain complicated steps in recipes, give suggestions when things went wrong, and even point people to locations in the kitchen where utensils and tools might typically be found. Yet there were still issues, such as what to do if someone has a bad allergic reaction while cooking. The problem is that it’s almost impossible to handle every possible scenario, yet that’s what true common sense entails. This fundamental aspect of AI has got somewhat lost in humanoid robots over the years. Generated speech, realistic facial expressions, telemetric controls, even the ability to play games such as “rock paper scissors” are all impressive. But the novelty soon wears off if the robots are not actually capable of doing anything useful on their own. This isn’t to say that significant progress isn’t being made toward autonomous humanoid robots. There’s impressive work going on into robotic nervous systems to give robots more senses for learning, for instance. It’s just not usually given the same amount of press attention as the big unveilings. The data deficit Another key challenge is the lack of real-world data to train AI systems, since online data doesn’t always accurately represent the real-world conditions necessary for training our robots well enough. We have yet to find an effective way of collecting this real-world data in large enough quantities to get good results. However, this may change soon if we can access it from technologies such as Alexa and Meta Ray-Bans. Nonetheless, the reality is that we’re still perhaps decades away from developing multimodal humanoid robots with advanced social AI that are capable of helping around the house. Maybe in the meantime we’ll be offered robots controlled remotely from a command centre. Will we want them, though? In the meantime, it’s also more important that we focus our efforts on creating robots for roles that can support people who urgently need help now. Examples would include healthcare, where there are long waiting lists and understaffed hospitals; and education, to offer a way for overanxious or severely ill children to participate in classrooms remotely. We also need better transparency, legislation and publicly available testing, so that everyone can tell fact from fiction and help build public trust for

Stockholm-based node.vc has closed a €71mn fund to back early-stage startups in the Nordics. “The Nordic tech ecosystem is thriving, especially in areas like AI, gaming, fintech, and climate tech,” John Elvesjö, managing partner at node.vc, told TNW. “We’re seeing experienced talent, particularly from companies like Klarna, Spotify, Voi, Kry, and Pleo, stepping up to become founders,” Elvesjö said. The devaluation of employee stock options and increasing layoffs have sparked “fresh entrepreneurial energy,” he added. The fund is sector-agnostic. This means that any startup with “innovative technology” can apply. The size of initial investment per company will be around €1-2mn, although this amount can rise up to €10mn over time. Calling all Scaleup founders! Join the Soonicorn Summit on November 28 in Amsterdam. Meet with the leaders of Picnic, Miro, Carbon Equity and more during this exclusive event dedicated to Scaleup Founders! Node.vc aims to build a portfolio of 22-26 investments. To date, it has backed three companies: no-code startup Lemonado, data management platform Starhive, and creative play studio Roro. Led by former founders and startup veterans, the VC firm boasts what it labels a “founder-first” approach. This, said Elvesjö, translates into “being there for the entrepreneur, getting hands-on, and helping remove roadblocks.” Some examples include support with fundraising, recruitment, and networking, as well as crafting product and go-to-market strategies. The aim is to accelerate growth and increase a startup’s probability of success. Despite a more, according to Elvesjö, “cautious” funding environment, node.vc’s fund attracted the backing of a series of investors, including Saminvest — a venture capital firm formed by the Swedish government in 2016. Other backers include several Nordic institutional investors and founders of a number of companies, such as Axis Communications and Yubico. Overall, norde.vc has built an investor base that counts over 70 entrepreneurs. “Early-stage funding still feels accessible,” Elvesjö said. “And the Nordics’ global outlook and collaborative nature keep it competitive, even in tighter markets.” source

From the outside, it looks like a clean, modern social housing block. You can tell it’s new — there are 84 shiny solar panels on the roof and the fresh paint has barely dried on the walls. But it’s how this 56-dwelling building in Barcelona, complete with ground floor nursery, was designed and built that really matters.  “Our software is our superpower,” says Lucas Carné, co-founder of 011h, a construction tech firm, as he describes how his company has designed digital tools to help architects plan buildings like this one. “We use a lot of prefab,” he adds. “That reduces the need for on-site labour.” Prefabricated construction relies on factory-made parts or components of buildings that are joined together on-site. It is far from a new concept but 011h has tried to make it much easier for architects to use this approach. The firm offers architectural software plugins containing libraries full of these prefab components. It makes the process of designing a building a bit like playing with digital Lego.  Improving productivity to solve housing shortage So far, 011h has collaborated with architectural and construction firms on several Spanish apartment blocks of differing designs — roughly completing one such block per year. The company is now planning to scale up to multiple projects annually, encompassing a total of around 200 dwellings per year. The firm has raised more than 35 million euros to date and has 90 employees.  This push to streamline design and construction is sorely needed, suggests Carné. As in many European countries, demand for housing in Spain is currently outpacing supply. Across the continent, the picture varies but a lack of housing, and a shortage of construction workers, are common themes. Materials costs also rocketed during the worst years of the pandemic. Now, various European startups are coming up with a raft of ideas to try and mitigate these problems. “Our goal is to improve productivity,” says Carné. “Probably the biggest problem this industry is facing is that productivity is stagnant.” Build with digital Lego Traditionally, construction is a woefully disjointed business. Roughly speaking, an architect will come up with a design and then hand it to a builder who, separately, will work out how to put that design together, for real. 011h takes a different stance. “We run a design and build process together with the suppliers and solutions we are going to use,” says Carné.  All the information about who will supply each building part, the costs involved, and the carbon footprint is collated from the very start in the company’s software. His team focuses on using sustainable materials — responsibly sourced timber rather than concrete, for instance, since the latter is high in embodied carbon. 011h’s buildings have achieved embodied carbon emissions lower than 400 kg of CO2 equivalent per square metre — significantly better than current averages. Plus, construction time is also shortened, by around 30% versus the average, claims Carné. Construction conservatism can hinder new technology adoption Revolutionising construction with digital technology has promise but it is difficult to do, says Sam O’Gorman at McKinsey’s Real Estate practice. There is often resistance to new ways of doing things in the construction sector, he notes. Plus, given the large amount of capital involved, there is significant risk if a project goes awry: “One mess-up could cost the business.” However, if firms entering this space can establish a good track record — perhaps through significant self-funding, at least initially — then they might be able to convince potential partners that they are worthy of collaboration or investment, adds O’Gorman. Another company that says technology can help us build homes smarter and faster is AUAR (Automated Architecture), in the UK. Gilles Retsin, co-founder and CTO, says his firm’s approach is to supply building firms with “micro-factories” — boxes with big robot arms inside. These arms work tirelessly to produce modular building units. Imagine a roughly 4×3 metre timber floor or wall panel, thick enough to be filled with insulation, which can be joined together with other panels to make a building.  A bot job “The robot will essentially grip raw materials, cut them, and put them on an assembly table where it nails them together,” explains Retsin. “Then they’re craned onto site by humans.” AUAR has 17 employees and has raised £2.6 million. The company is currently targeting the European and US markets, both of which are suffering from labour shortages in the construction sector. “We just built a two-storey building in Belgium,” says Retsin. In that case, the microfactory robot took about three days to make the building block units and human workers took a further three days to connect those units together and complete the building’s main structure. Initially, AUAR provided its microfactory robots at a cost of £250,000 each but Retsin says the firm is shifting to a hardware-as-a-service model in which builders can pay a smaller fee to have the microfactory delivered to site. They will then pay a further small charge per square metre of building produced. Although the firm is yet to scale, Retsin emphasises the huge potential of running the robot arms continuously — and perhaps using many of them in parallel. “The capacity of one robot is 200 homes per year, if you run it eight hours per day,” he says, adding that the next step for the firm is to tackle a 30-home project in the US next year. Robo brickie Finally, even in 2024, bricklaying remains a crucial skill required for housebuilding in Europe, as brick-based construction remains prized by European homebuyers. However, bricklayers are among the workers currently in very short supply. Monumental, based in Amsterdam, has an alternative.  “You tell me, ‘I want to build a façade for a house, it’s X square metres’,” says Salar al Khafaji, founder and CEO. “I will quote you a price and do it for you — but I will do it with robots.” The firm, which has raised $25 million to date and has 32

In 2020, Google identified that more than 70% of its Chrome browser’s severe security bugs were in fact caused by memory safety issues. “That is,” the Chrome team said, “mistakes with pointers in the C or C++ languages which cause memory to be misinterpreted.” In 2022, the NSA weighed in on memory safety with Neal Ziring, its cybersecurity technical director saying that “Memory management issues have been exploited for decades and are still entirely too common today. We have to consistently use memory safe languages and other protections when developing software to eliminate these weaknesses from malicious cyber actors.” That wasn’t the end of the matter, however. Memory safe programming languages have continued to be under an intense spotlight. In February of this year, the US White House Office of the National Cyber Director (ONCD) issued a report advising that all programmers should move to memory-safe programming languages. The 💜 of EU tech The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now! 5 jobs to discover this week Cybersecurity Coordinator France M/F, MBDA France, Le Plessis-Robinson Data Scientist (F/H), Novencia, Lyon Software Architect, GDV Dienstleistungs-GmbH, Hamburg Software Developer, InTraffic, Utrecht Software Architect, Capgemini, Eindhoven The report pointed out that the burden of cybersecurity threat protection is currently placed on end users, and that, “efforts must be made to proactively eliminate entire categories of software vulnerabilities.” The report elaborated further, saying that, “Experts have identified a few programming languages that both lack traits associated with memory safety and also have high proliferation across critical systems, such as C and C++.” Memory safety matters now more than ever, because so much more of what we do happens online. The pandemic accelerated the rapid adoption of ecommerce, online payments, and digital advertising, according to the World Economic Forum. As a result there are a lot more potential vulnerabilities to exploit. Stack Overflow points out that some of the biggest vulnerability events of the past were memory-safety issues. These include 2014’s Heartbleed, which affected OpenSSL software allowing bad actors to steal X.509 certificates, usernames and passwords, instant messages, and emails. In 2017, the WannaCry ransomware attack garnered massive attention as it spread globally, infecting more than 230,000 computers. A new Consumer Security and Financial Crime Report from Revolut points to Meta platforms as the biggest source of all scams (62%) globally during the first half of 2024. Revolut identified that Facebook had fraud volumes (39%) which were more than double that of WhatsApp (18%). Making C++ safe Memory safe languages do exist and include Rust, Go, Java, Swift, and Python. C++ is under particular scrutiny because of the amount of critical code that has been written in it. Given the context, it isn’t so surprising that the C++ community has reacted, announcing the Safe C++ Extensions proposal in September of this year. ​​The work is being done via the C++ Alliance, and its president and executive director Vinnie Falco said that this was, “a revolutionary proposal that adds memory safety features to the C++ programming language.” Falco added that: “the need for safe code has never been more pressing. With the increasing importance of software security and reliability, developers are facing mounting pressure to adopt safer coding practices. The Safe C++ Extensions aim to address this critical need by introducing novel features that prevent common memory-related errors.” So will this fix the issue? Some critics are skeptical, and the developer from the C++ Alliance, Sean Baxter points out that: “There’s only one popular systems level/non-garbage collected language that provides rigorous memory safety. That’s the Rust language. Although they play in the same space, C++ and Rust have different designs with limited interop capability, making incremental migration from C++ to Rust a painstaking process.” A number of actions are suggested to ensure performant C++ code, including prohibiting developers from writing operations that might result in lifetime safety, type safety, or thread safety undefined behaviors. Additionally, there are other challenges, with Baxter pointing out that, “Although they play in the same space, C++ and Rust have different designs with limited interop capability, making incremental migration from C++ to Rust a painstaking process.” Moving code to memory safe status will be painstaking and time-consuming, but the Defense Advanced Research Projects Agency (DARPA) is seeking to bridge this gap using AI. It is developing a programmatic code conversion vehicle called TRACTOR (Translating All C TO Rust). It says that, “the goal is to achieve the same quality and style that a skilled Rust developer would produce, thereby eliminating the entire class of memory safety security vulnerabilities present in C programs.” Ready to find your next software role? Check out The Next Web Job Board source

 Welcome to the new episode of the TNW Podcast — the show where we discuss the latest developments in the European technology ecosystem and feature interviews with some of the most interesting people in the industry. In today’s special episode, we’re happy to present an interview with Anita Schjøll Abildgaard, co-founder and CEO at Iris.ai. The startup has been around for almost a decade and saw several significant pivots, while Anita and the team have been through all the highs and lows imaginable, including balancing on the brink of bankruptcy. Anita and Andrii talked about it all, as well as the future of academic publishing, the prospects of generative AI, the inconvenience of chat interfaces, and much more. Here are the links for this episode: Music and sound engineering for this podcast are by Sound Pulse. Feel free to email us with any questions, suggestions, and opinions at [email protected]. source

“Welcome on board. I have been tasked with taking you to Mäntsälä — in the middle of nowhere,” the minivan driver greets us in the characteristic clear and unhurried intonations of a Finnish native speaker. Mäntsälä is, indeed, in the middle of nowhere. But this kind of location is often where you find collections of some of the most powerful machines of today, humming away behind doors along unpretentious corridors. This includes Nebius’ AI data centre, taking shape in the small community an hour’s drive or so north of Helsinki. Amsterdam-based Nebius is labelling itself an AI cloud infrastructure company. Its proprietary platform, it says, has been optimised for AI training and inference without performance bottlenecks. “The AI cloud is different from the ‘regular’ cloud. In the set of tools, in the applications, yes, but the people who use it are also different,” says Nebius’ head of product and infrastructure, Andrey Korolenko. The would-be European AI infrastructure force has begun amassing a tremendous amount of compute. Today, it announced that it will triple the Mäntsälä capacity to up to 60,000 Nvidia GPUs. Precisely, this entails deployment of Nvidia’s H200 GPUs, available from November, in addition to already installed H100s. Nebius is also one of the launch partners for Nvidia’s upcoming Blackwell platform. The Blackwell GB200 will enter mass production in December. Korolenko states that whenever the first unit ships (currently slated for Q1 2025), Nebius will also have it “in a matter of weeks.” Filling gaps in the AI training and inference market While others are toiling away to close the gap to Nvidia, the latter’s hardware (along with its CUDA platform) is still the gold standard for AI training and inference. Nebius’ core business is to offer time on its GPUs to everyone from app developers and companies optimising foundation models for their own businesses to AI model tuners and builders, from pre-training to inference, with different levels of support for different levels of skill. The company has 400+ engineers in its employ, and customers already include the likes of Mistral AI, Genesis Therapeutics, Recraft, and Jetbrains. Nebius has built custom racks for its Nvidia hardware. “We are doing it [building the data centre] from the ground up,” Korolenko says. “If you build it, you can just adjust it,” he adds, addressing the evolving nature of chips along with all the infrastructure requirements this entails. During our visit, a large batch of racks has just arrived from the manufacturer in Taiwan, and Nebius crew are in the process of unpacking them from their boxes. At the mention of Taiwan, red flags concerning supply chains immediately pop up, and it is not long before the question of “What will happen to you in the event of a ‘reunification’ in the South China Sea?” arises. “What will happen to the world in this case?” Korolenko muses, sounding rather stoic. Nebius is busy setting up new server halls to host its GPUs. Credit: Linnea Ahlgren/TNW The data centre also houses ISEG, which currently ranks as the 19th fastest supercomputer in the world, and the fastest commercially available in Europe. With 35.26 GFlops per watt-second, it has also made it to number 24 on the Green Top500 list. One billion USD in European AI infrastructure The build-out of Mäntsälä is part of a plan to splurge a total of $1bn across Europe by mid-2025, including opening an additional three data centres across the continent (as well as one in an as-of-yet undecided location in North America). This includes a recent addition in Paris, a colocation deployment based at Equinix’s PA10 campus. The site, located in the Saint-Denis district, has an urban farm on the roof, and heat from its servers was used to heat the Olympic pool during the 2024 summer games. At its Mäntsälä data centre, Nebius provides heating for about 2,000 homes. A feature that, if replicable, makes the company’s presence an attractive proposition for other remote locations eager to up their green credentials. Nebius’ data centre currently only utilises air to cool its servers. This will change with the addition of the later GPU models. What will not change is the amount of power that will go back towards heating the neighbouring town — the percentage, approximately 70%, may even increase when deploying a liquid cooling system. In fact, with the expansion, Nebius will export more heat than the village of Mäntsälä requires. The Yandex legacy You may wonder how come you have not heard of a company with 400 engineers able to get a hold of tens of thousands of highly coveted Nvidia GPUs up until now. Nebius recently emerged from the European remnants of Yandex (which had a long-term relationship with the GPU maker) following the company’s high-profile divestment from Russia. One of the things to come out of that laborious process, other than the Mäntsälä data centre, was a few billion dollars in cash. This is currently fueling the rollout of what could end up being a global force in AI infrastructure. As a result of the legacy from Yandex, Nebius Group is listed on NASDAQ. The group also encompasses data business Toloka, upskilling edtech platform TripleTen, and autonomous driving technology unit Avride. However, its shares are not currently trading. Earlier this month, it announced it had enlisted Goldman Sachs as its financial advisor with a view to recommence trading further down the line. Yandex founder Arkady Volozh is Nebius’ CEO, who has publicly criticised Russia’s war in Ukraine. However, along with the Mäntsälä data centre, Nebius has likely also inherited a certain degree of suspicion due to its origins. The company has moved its employees out of Russia, essentially evacuating thousands of people including whole families to locations outside of the country in 2022. The company does not allow its employees to work from inside Russia should they go back and visit. It has also had to go through rigorous vetting from EU authorities to receive approval for the divestment deal. Its execs, who had to

DeepL, the Cologne-based AI translation unicorn, today launched its first tech hub in the US, in New York City. The move follows the company’s increasing growth and investment in the US market, where it already counts customers such as Coursera and Morningstar. DeepL opened its first US office, in Austin, Texas, earlier this year. The New York tech hub will focus on research, product innovation, and engineering, aiming to boost the startup’s expansion in the region. “[It] positions us at the centre of one of the largest talent pools in the market and brings us closer to our customers, including many Fortune 500 companies,” Jarek Kutylowski, CEO and founder of DeepL, said in a statement. Calling all Scaleup founders! Join the Soonicorn Summit on November 28 in Amsterdam. Meet with the leaders of Picnic, Miro, Carbon Equity and more during this exclusive event dedicated to Scaleup Founders! The company is actively hiring for product and engineering positions, planning to double the hub’s size within the next 12 months. To further support growth in the US, DeepL has also appointed a new chief technology officer, Sebastian Enderlein, who previously led software development and engineering teams in Uber and Salesforce. Strategic focus Since its founding in 2017, DeepL has managed to successfully rival machine translation giants such as Google Translate. The company reached unicorn status in 2022, and rose to a $2bn valuation in May this year, following a $300mn investment round. One element behind DeepL’s significant growth is its focus on AI translation alone, supported by its proprietary neural network technology. “Translate isn’t the core business of Google — it’s one of the 100 side gigs” Kutylowski told TNW in a previous interview. “The same goes if you consider LLMs and the OpenAIs of this world as our competition; translation is only one thing of what they’re doing and their GPU is doing a tonne of different things. We’re focused on one particular area.” Another element is DeepL’s strategy to mainly target the B2B market, where it sees the biggest demand for its services. To date, the startup offers translation services for over 30 languages and has a customer network of more than 100,000 businesses. In the past few months, DeepL has launched a series of products for business users, including an AI writing assistant and an AI glossary generator. source