Last month, the NVIDIA GTC Conference brought together the global AI developer community. Nebius — a specialized AI cloud provider and preferred NVIDIA partner — was there all week, showcasing its solutions to developers worldwide.
A lot happened during the event. Nebius not only delivered several talks about its platform and the broader AI landscape, but also took the opportunity to interview representatives from some of its most innovative customers.
In this article, I’ll break down the key takeaways from those conversations to explore why developers choose Nebius and how its solutions stack up against the competition.
Announcements
Before diving into the main topic, let’s quickly go over the key announcements Nebius made during GTC 2025.
Nebius revealed that it will be among the early AI cloud providers to adopt the newly unveiled NVIDIA Blackwell Ultra AI Factory platform, announced by Jensen Huang during his keynote. This platform is set to become the world’s most advanced compute-on-demand infrastructure, enabling AI developers and enterprises to build the next generation of agentic, reasoning, and physical AI systems.
As part of this initiative, Nebius will offer customers access to instances powered by NVIDIA GB300 NVL72 — accelerated by 72 NVIDIA Blackwell Ultra GPUs — by the end of 2025. To support this rollout, Nebius announced two major infrastructure moves:
The newly announced New Jersey data center will be fully dedicated to NVIDIA Blackwell-architecture GPUs.
Its Kansas City data center will be upgraded with NVIDIA HGX B200 systems.
“Access to world-class AI infrastructure is the key to realizing the full potential of AI. With dedicated NVIDIA Blackwell capacity available from next quarter, Nebius is giving AI innovators and enterprises everywhere access to the world’s most powerful AI compute through our AI-native, multi-tenant cloud. This is the future of AI, and we’re building it today.”
— Arkady Volozh, Founder and CEO of NBIS
Additionally, Nebius announced it is joining the ecosystem of partners for NVIDIA Dynamo, an open-source inference serving framework designed to scale GenAI deployment across large distributed environments.
Interviews with Team Members
Interview with the Head of Hardware R&D and the Head of Cloud Platform Foundation
One of the most common questions customers ask Nebius is: “How are you different from other cloud providers?” The short answer: A deep focus on reliability, customization, and engineering talent.
When Nebius set out to rebuild its cloud platform from scratch in just 12 months, it was the people who made it possible. Most of the engineers who joined the company came from Yandex, following the founding team to help build a new kind of cloud business from the ground up.
This engineering depth, combined with long-standing partnerships, is what allows Nebius to build a competitive moat. While the original idea was to simply sell compute hours, the market made it clear: especially for AI training, developers don’t just want raw compute — they want “good compute” that reliably contributes to model progress.
That’s why reliability has become a core differentiator. In today’s large-scale clusters, hardware failures are a major concern. To tackle this, Nebius designs its own equipment and heavily invests in “auto-healing” infrastructure. Their system is built to detect and recover from failures at every level — hardware, monitoring, compute management, Kubernetes orchestration, inference services, etc. The deeper a customer goes into the Nebius stack, the more “auto-magical” the experience becomes.
Use cases vary widely. Some customers train foundational models — for tasks like brand image or logo generation. Others work on weather prediction models to help manage energy costs, video generation, or inference at scale (often in multi-cloud environments).
“We’re really starting to see inference picking up.”
Ultimately, Nebius doesn’t just want to sell compute — it wants to be a trusted AI cloud partner, understanding each customer’s needs and delivering tailored infrastructure to support their goals.
Interview with the Senior AI Product Marketing Manager
Nebius also introduced its AI Studio, an Inference-as-a-Service platform designed to simplify access to leading open-source models like LLama, Mistral, Qwen, DeepSeek, and more.
The goal is to eliminate infrastructure complexity and reduce the cost and learning curve typically involved in deploying and scaling these models. AI Studio delivers a wide range of core services, such as:
Text generation (e.g., reasoning models like DeepSeek)
Text-to-image synthesis
Vision models for image recognition
Embedding models for RAG implementations
Custom-function models tailored to specific use cases
And more.
One of the standout features is the platform’s speed of model integration. Whenever a new model becomes popular, the Nebius team can onboard it in just days — or sometimes even hours. A great example was their rapid integration of DeepSeek.
Interview with the Head of Solutions Architecture
Nebius started as a group of HPC engineers selling GPUs, and turned into an AI-centric cloud platform.
Today, what sets Nebius apart is its hands-on experience with large-scale training and inference workloads. The team understands the unique challenges AI developers face and works closely with customers to solve them.
Like AWS in its early days, Nebius initially catered to startups. Many of those early clients are now frontier AI labs — some of the biggest names in Generative AI. The company has several different workloads, whether training, fine-tuning, pre-training, inference, data preparation, etc, but the “bread and butter” is its specialization in large-scale distributed training, arguably the most expensive and complex part of Generative AI.
Interview with the VP of Product Marketing
"We're a very customer-centric cloud, so the cloud experience matters to us and is a top priority almost every day.
One thing most people don’t know about our platform is that we have in-house expertise and a dedicated AI team building our own LLM model. This model performance-tests the platform and provides feedback on any issues or areas for improvement.
I see this as a huge value-add for the customer experience because it allows us to continuously enhance what we deliver to the end user."
Interview with the Chief Communications Officer
According to Nebius’ CCO, hyperscalers build infrastructure primarily for their own internal needs, leaving the broader AI industry underserved. Nebius is in the perfect position to fill that gap: building the entire stack from scratch — from data centers and GPUs to its AI-optimized cloud and software layer.
Few companies can do this at scale, which gives Nebius a unique position in the market.
Right now, most of its customers are AI-native startups, but that’s expected to change. Over the next year or two, the company anticipates a wave of enterprise and corporate adoption of AI, which represents a massive opportunity. Nebius is already preparing for this shift by developing the right mix of services to support broader enterprise needs.
Interviews with Nebius Customers and Industry Experts
Interview with Higgsfield AI
Higgsfield AI is one of the best AI video tools out there, and in this short interview, its Founder explained why they chose Nebius over other GPU cloud providers.
Summary:
- Nebius is very friendly to start-ups. If you talk to developers off the record, you'll often hear that larger cloud providers have unclear terms, while Nebius offers full transparency and fair terms with its on-demand GPU offering.
As I explained before, this is particularly attractive for start-ups and smaller developers because they can't predict exactly how their workloads will change. "Nebius has much better flexibility than any other cloud."
- Larger cloud providers typically aren't interested in start-ups — they prefer higher-volume customers. "Nebius provides infrastructure and reliability at the same level as Google or Amazon."
Again, just like I said in my last article, hyperscalers are not a threat because they often serve different needs.
- Nebius uses a pay-per-token model, and the AI Studio is highly versatile in terms of model offerings, supporting everything from ChatGPT to DeepSeek and other LLMs.
- It's very easy to work with Nebius. The team is highly responsive, unlike larger cloud providers.
- "Nebius' flexibility, combined with its strong support, will likely attract many media companies."
- Other cloud providers have lengthy agreements — sometimes up to 200 pages — while Nebius keeps it simple with 3-5 page contracts. Transparency matters.
- Most cloud providers require commitments for a fixed number of GPUs per year, which most start-ups can't handle. Nebius, however, offers resource-based commitments, where a minimum spend unlocks discounts instead of requiring a set GPU quota.
- Nebius provides a level of customization and flexibility that no other cloud provider offers.
Interview with Dylan Patel
Dylan Patel, Founder of SemiAnalysis, on how Nebius differentiates itself from other neoclouds:
“The end market for cloud is splintering significantly. We used to have just a few major cloud providers, but now we have dozens of neoclouds, and the user experience across them varies widely. You might get just bare metal instances, or virtual machines, or Slurm, or Kubernetes with Slurm on top—maybe something like a SUNK-type stack. Some offer inference services. There are all sorts of different engagement models with the cloud, and it’s definitely not a one-size-fits-all scenario.
A cloud provider has to be able to deliver across every vector, because different teams within a company are going to want different things. Some will only use an API. Others will want direct access to bare metal. Most teams want something in the middle.
Where Nebius stands out is in its ability to offer that full spectrum. Most clouds focus on just a few things, but Nebius supports the whole gamut. That lets them serve a broader base of customers, and as those customers evolve and start asking for X, Y, or Z, Nebius doesn’t say, ‘Sorry, we don’t offer that.’ Their answer is, ‘Let’s go.’
And then there’s scale. The cloud market is really evolving because of all the splintering, and I think there will be some consolidation again in the coming years because not all clouds will be able to justify their existence – Nebius having the platform that is the broadest is very useful.”
He also touched on Nebius’ engineering strength — something I’ve covered before and which continues to set them apart:
As NVIDIA’s pace of innovation accelerates, many believe that the massive CapEx investments from neoclouds like Nebius and CoreWeave risk becoming obsolete too quickly.
What makes the difference? Engineering resources.
Nebius has hundreds of AI/ML engineers with over a decade of cloud experience. Unlike most cloud providers, they design their own racks — a key advantage. Importantly, they directly collaborate with NVIDIA to design racks that are future-proof, ensuring seamless integration with next-generation chips already in development.
This also gives Nebius a structural cost advantage over other cloud providers. Instead of buying from Dell or Supermicro, they source hardware from ODMs in Taiwan and customize it for efficiency. This approach:
• Lowers costs (some savings are passed to customers)
• Improves reliability (rigorous in-house testing)
• Reduces downtime (better liquid cooling, optimized performance)
Less downtime = fewer wasted resources = significant cost savings.
Nebius is stacked with top-tier talent at every level, and that's what will drive its success.
Interview with Lynx Analytics
Lynx Analytics is an AI and machine learning company with a strong focus on life sciences. Headquartered across three global hubs — San Francisco, Singapore, and Budapest — roughly 70% of its revenue is generated from the U.S., with the remainder coming from Asia and Europe.
The company supports the full drug development lifecycle, spanning preclinical research, clinical trials, and certain commercial phases. Lynx collaborates with industry leaders like Roche, Genentech, Novo Nordisk, and AstraZeneca, along with emerging biotech firms such as Hummingbird Bioscience.
Why Nebius?
“We chose Nebius because it’s one of the few providers that makes it easy to set up multi-GPU clusters. This significantly speeds up matrix multiplication operations and boosts overall computational efficiency. That was the key driver for us — though we appreciated many other aspects of the platform too.”
Lynx had evaluated other cloud providers but faced challenges with performance and reliability.
“We lost our data twice in seven months with our previous provider. We needed something more stable, scalable, and GPU-compatible. The ability to run Spark on GPU clusters and connect CPUs and GPUs in the same environment was crucial. Nebius checked all those boxes.”
Their team also emphasized how Nebius’ global infrastructure — particularly its European clusters — was vital for R&D, not only for performance but also to maintain GDPR compliance for internal experiments.
Chief Architect's Take:
“It’s been a great experience working with the Nebius team. Honestly, we didn’t need much support — the online documentation was enough. We spun up instances with just a few clicks. Setting up GPU clusters is also very simple — unlike with other cloud platforms where it’s often a painful process.”
On Competing Hyperscalers:
“In my experience, AWS works — but I dread opening the UI. You always have to edit some JSON file just to get basic things going. Nebius is different. It was designed from the ground up for AI workloads. It’s simpler, more intuitive, and overall just more enjoyable to use.”
Interview with CentML
CentML builds infrastructure that improves the cost and performance efficiency of AI and machine learning models, including LLMs and computer vision workloads. The company optimizes every stage of the pipeline — from training and inference to deployment.
As a bare-metal customer, CentML runs its proprietary software directly on Nebius hardware. This partnership model allows CentML to support other customers in leveraging Nebius as well.
One of the biggest challenges for companies like CentML is finding a cloud solution that combines high performance, reliability, and affordability. While many new providers have entered the market promising cheaper alternatives to major players like AWS, GCP, and Azure, most fail to deliver the consistent performance and reliability required for enterprise-grade AI workloads.
Nebius stands out by offering a powerful combination of performance, reliability, and cost-effectiveness. Its infrastructure has proven to be both robust and responsive, making it a strong partner for running demanding AI workloads efficiently.
Another critical advantage Nebius offers is early access to cutting-edge hardware, such as NVIDIA’s latest GPUs. For CentML, having access to chips like the H100, H200, and now the B200 — as soon as they’re available, and sometimes even before public release — is essential. This early access enables the team to prepare the entire software stack in advance, ensuring they stay ahead of the curve in terms of performance and innovation.
Another standout feature that customers appreciate about Nebius is, again, its flexibility. A key advantage of using Nebius is not just its competitive pricing, but its adaptable approach. For companies like CentML, this is crucial — especially when working with new customers, where long-term commitments are uncertain.
Unlike many providers that require year-long reservations to unlock better pricing, Nebius offers flexible terms, such as 3- or 6-month reservations, without compromising on cost. They also provide attractive on-demand pricing, allowing teams to scale as needed without overcommitting early on.
This flexibility — combined with strong performance — makes Nebius “the best performance-per-dollar solution currently available in the market”.
Interview with KissanAI
KissanAI is a pioneering generative AI startup focused on transforming agriculture. Its mission is to bridge the knowledge gap between farmers and agribusinesses using large language models. In many regions, language and literacy barriers make it difficult for farmers to access vital information — like product manuals and usage instructions — produced by agri companies.
The company tackles this by developing agriculture-specific AI models built on curated knowledge bases. These models support key workflows such as advisory services, conversational commerce, customer support, and sales co-pilots, helping agribusinesses communicate more effectively with farmers.
KissanAI chose Nebius for its reliable, flexible, and cost-effective cloud infrastructure, which provided both the computing resources needed for AI model training and the lowest-cost inference options. As a bootstrapped startup working in agriculture — a sector that typically doesn't attract high levels of investment — securing large-scale clusters for model training was a significant challenge. However, Nebius delivered, offering uninterrupted access to GPUs and cloud resources that were vital for the company’s growth.
The ability to handle both training and inference tasks within the same platform proved especially valuable. Nebius enabled KissanAI to manage hands-on model training while simultaneously running low-cost inferences and hosting models — all without needing to juggle resources across multiple providers. This all-in-one solution saved the team both time and resources.
The user-friendly experience also stood out. Even team members without technical backgrounds were able to easily access and utilize the platform. Additionally, KissanAI found that Nebius offered the most competitive pricing for their needs, along with added services and features that typically come at a premium with other providers.
Customer support was another highlight. The Nebius team consistently delivered quick responses and hands-on assistance whenever needed — unlike other providers, where support is often slow or unresponsive.
For KissanAI, Nebius has proven to be a trusted partner, offering the performance, flexibility, and customer care necessary for a startup to thrive.
Interview with Captions
Captions is a generative video and video editing platform that combines AI technology with creative storytelling. The company develops its own models and provides a suite of tools for consumers to use AI-driven video editing features. The goal of Captions is to empower users to tell stories through video, leveraging the latest in AI to enhance the creative process.
Nebius has played a crucial role in helping Captions develop its large-scale audio-to-video foundation model, Mirage. By providing reliable and scalable cloud infrastructure, Nebius has enabled Captions to significantly improve the stability and performance of its training process. Compared to their previous provider, the improvement in both model size and training reliability has been an order of magnitude better. Nebius' support allowed Captions to overcome previous limitations, reducing the need for low-level debugging and enabling the team to focus on solving core problems. This stability has been essential as the company pushes the boundaries of what these models can achieve.
“Their sales team and technical support have been very dependable and proactive in providing the solutions we need. As we think about scaling out inference, they've been very flexible, offering us easy ways to scale up our model and quickly test it in different settings.”
“We know we can trust them. Having a partner we can truly rely on has simplified our lives. We know that if something crashes in the middle of the night, their team is there to quickly recover it. Our team is based solely in New York, so we’re time zone restricted. Just knowing that there’s support around the clock for any technical issues or expansion needs has been a huge relief for us.”
Bonus: Other Customer Stories
Nebius has supported many other companies in enhancing their AI capabilities:
Chatfuel: Leverages Nebius’ infrastructure and Llama-405B models to improve chatbot performance, optimizing training costs and real-time efficiency while enabling quick deployment.
London Institute for Mathematical Sciences (LIMS): Uses Nebius’ scalable, reliable infrastructure for advanced research on LLMs, enhancing model training and data processing capabilities.
Positronic Robotics: Uses Nebius' virtual machines with NVIDIA H100 GPUs to train AI models for robotic control systems, aiding in the development of intelligent cleaning robots.
SynthLabs: Partners with TractoAI (owned by Nebius) to simplify training infrastructure, accelerating the release of the Big Math dataset and improving model training using high-end GPUs.
Krisp: Reduced model training time by 50–80% by switching to Nebius’ NVIDIA H100 GPUs, enhancing AI models for noise cancellation and speech recognition.
Dubformer: Relies on Nebius for AI dubbing and localization, handling vast audio datasets and ensuring continuous 24/7 model training for improved efficiency.
Unum: Streamlined the training of multimodal models and successfully open-sourced several models, advancing research in compact AI models.
TheStage AI: Optimizes inference performance with Nebius’ GPU instances, ensuring scalability and reliability for model evaluation and deployment.
Recraft: Uses Nebius to train a generative AI model with 20 billion parameters, overcoming challenges and achieving benchmark-breaking performance in AI design.
Conclusion
Nebius is clearly a standout choice for developers, and it’s easy to understand why. Their focus on flexibility, reliability, and top-notch customer support sets them apart in a crowded field. By building their platform from the ground up specifically for AI, they avoid the pitfalls of retrofitting older cloud technologies. This approach allows them to deliver exactly what AI developers need, whether it’s startups looking for fair terms and transparency or enterprises seeking cutting-edge infrastructure.
Their early adoption of NVIDIA’s latest innovations, combined with their commitment to engineering excellence, ensures developers always have access to the best tools available. Plus, their customer-centric mindset means they’re not just selling compute power — they’re partnering with developers to help them succeed.
Whether it’s startups pushing the boundaries of generative AI or established companies scaling their operations, Nebius proves time and again that they’re one of the go-to choices for anyone serious about AI development.
Make sure to read my latest article about the company to see my updated Valuation Model:
That’s it! Thanks for reading.
Disclaimer: As of this writing, M. V. Cunha holds a position in Nebius Group (NBIS) at $26.27/share.
Important Communication:
In a few weeks, I’ll start restricting part of my content to paid subscribers. As such, the current price (which was set lower on purpose, since it was voluntary) will change. If you choose to support me before then, you’ll get a 50% discount on the yearly plan. I spend countless hours working to bring you the best content I can, so I appreciate your understanding. Don’t worry — I’ll still publish plenty of free articles as well.
Hi! I enjoy your content but, after reading your Deep dives on Nebius, I still have a doubt: When Nebius is expected to reach profitability?