Article Directory
Nebius Token Factory: Is This AI's Assembly Line or Just More Hype?
Nebius Group, formerly Yandex, is making waves again, this time with its Nebius Token Factory. The pitch? Production-scale AI inference using open-source models. They're promising speed, efficiency, and enterprise-grade security. But let's dissect this claim, shall we?
Peeling Back the Layers: Infrastructure and Claims
The foundation is Nebius AI Cloud 3.0 "Aether," built for intensive AI workloads. They're boasting sub-second latency and 99.9% uptime, even with hundreds of millions of requests per minute. That's a bold statement. Prosus, a user of Nebius Token Factory, claims up to 26x cost reductions compared to proprietary models. Impressive if true, but context is crucial. What models were they comparing, and what was the specific workload? These are the types of details that often get glossed over (or conveniently omitted) in press releases.
Hugging Face is also collaborating, aiming to improve access and scalability. This partnership lends credibility, but it doesn't guarantee success. The devil, as always, is in the implementation.
Nebius is touting support for a range of open-source models (NVIDIA Nemotron, DeepSeek, Llama, etc.) and the option for customers to host their own. This flexibility is a definite plus. The ability to fine-tune and distill models, potentially cutting inference costs and latency by up to 70%, is also enticing. But achieving those savings requires expertise and optimization. It's not a plug-and-play solution.
The focus on governance, with Teams and Access Management, Single Sign-On (SSO), and enterprise-focused billing, is a smart move. Enterprise clients demand control and compliance. The zero-retention inference in EU or US datacenters caters to data-residency requirements, another key concern for larger organizations. And, SOC 2 Type II including HIPAA, ISO 27001 and ISO 27799 certifications are table stakes for this market.
The CoreWeave Comparison: A Reality Check
Nebius is positioning itself against competitors like CoreWeave. But how do they stack up? Nebius's revenue soared 462% to $118 million in 2024, but its adjusted EBITDA was negative $266 million. CoreWeave, on the other hand, saw revenue skyrocket 738% to $1.92 billion in 2024, with an adjusted EBITDA of $1.22 billion. Both are unprofitable on a GAAP basis, but CoreWeave is clearly further ahead. Better Artificial Intelligence Stock: Nebius Group vs. CoreWeave
Analysts project Nebius' revenue to reach $4.34 billion by 2027, with adjusted EBITDA turning positive in 2026. CoreWeave's revenue is projected to hit $18.1 billion during the same period. Nebius is growing faster on a percentage basis, but that's because it's starting from a much smaller base.
Nebius's market cap is around $31.4 billion, valuing it at 7 times its projected 2027 sales. CoreWeave, with a market cap of $72.5 billion, trades at just 4 times its estimated 2027 sales. These are speculative valuations, to be sure (especially given the volatility in AI), but CoreWeave appears to be the more reasonably priced option.
And this is the part of the report that I find genuinely puzzling: Nebius only owns a single data center in Finland, leasing the rest. CoreWeave operates 33 first-party data centers. That infrastructure gap is significant. The Microsoft deal is a positive, but relying heavily on a single partner carries risk. CoreWeave's revenue is also heavily reliant on Microsoft (around 70%), which isn't ideal either.
More Promise Than Proof?
Nebius Token Factory sounds promising. The focus on open-source models, enterprise features, and cost optimization is appealing. But the company needs to demonstrate it can scale its infrastructure and achieve profitability. The partnership with Microsoft is a major boost, but it's not a guaranteed success. Until Nebius shows concrete results, it remains a speculative play.
So, What's the Real Story?
Nebius Token Factory is an interesting development, but it's not a clear-cut winner. The company needs to execute its plans and prove it can compete with established players like CoreWeave. For now, approach with cautious optimism.