The numbers whispered a story of explosive growth. The market heard a warning. On August 13, Nebius (NBIS) reported Q2 2025 revenue of $582.3 million, a 454% year-over-year surge that would make any growth investor salivate. Adjusted EBITDA hit $236.2 million, a 40.6% margin that screams operational efficiency. Yet the stock dropped 9.5% in the following session. The code whispered what the pitch deck screamed: growth at scale is beautiful, but the architecture of profitability is still under construction.
This is not a technology company that failed to deliver. This is a neocloud – a GPU-as-a-service infrastructure provider born from the ashes of Yandex’s Russian exit. Nebius sells AI compute to developers and enterprises, competing with CoreWeave, Lambda, and the hyperscalers. But the stock’s reaction reveals a deeper truth: the market is no longer buying growth at any cost. It is demanding proof that the unit economics hold under the weight of capital intensity.
Let me be clear. I have audited crypto projects that raised $20 million on a whitepaper full of broken hash functions. I have seen the gap between press release and reality. Nebius is not a rug pull. But the financials demand a forensic dissection, not a celebratory headline.
Context: The Neocloud Thesis
Nebius operates in the AI infrastructure layer – the physical and software stack that powers model training and inference. Unlike traditional cloud providers (AWS, Azure, GCP) that offer general-purpose compute, neoclouds specialize in GPU clusters with high-bandwidth networking and optimized scheduling. The thesis is simple: AI workloads are compute-intensive and growing exponentially. Companies building the next GPT or Stable Diffusion need raw GPU power, and they are willing to pay a premium for performance and availability.
Nebius emerged from the restructuring of Yandex N.V., which divested its Russian assets in 2023. The remaining entity, now rebranded as Nebius, retained the core engineering team that built Yandex’s massive distributed systems. This is a hidden asset. In an industry where talent is scarce, having a team that scaled a search engine to hundreds of millions of users gives Nebius a structural advantage over capital-driven competitors.
But there is a catch. Neoclouds are capital-intensive. Every dollar of revenue requires a dollar of upfront GPU procurement. The financial model is akin to a hotel chain: you build rooms (buy GPUs) before you can rent them. The timing mismatch between capex and revenue creates a volatile earnings profile. Nebius’s Q2 revenue of $582.3 million is impressive, but it reflects investment decisions made in 2024. The market is now asking: what did you buy, and at what price?
Core: The Systematic Teardown of Nebius’s Financial Architecture
Let me walk you through the numbers that matter, not the ones in the press release.
Revenue Growth: Real but Base-Effect Distorted
Q2 revenue of $582.3 million represents a 454% year-over-year increase. But the base year (Q2 2024) was only $105.1 million. That is a small number. Sequentially, the growth is more telling: Q1 2025 revenue (implied from H1 2025 total of $981.3 million minus Q2) is approximately $399 million. That gives a sequential growth of 46% quarter-over-quarter. This is still exceptional for a B2B infrastructure business. Most cloud providers grow 10-20% QoQ in their early stages. Nebius is doubling that. The sales pipeline is clearly strong.
However, this growth rate is not sustainable. It is the result of a massive GPU fleet coming online after a long procurement and deployment cycle. The law of large numbers will slow this down. The question is whether the company can maintain a 30%+ sequential growth for the next few quarters, or if this is a one-time catch-up.
EBITDA Margins: The Mask of Profitability
The adjusted EBITDA margin of 40.6% is the headline figure that makes Nebius look like a cash machine. But I have seen too many crypto projects use adjusted EBITDA to hide the cost of equity. Let me dissect.
Adjusted EBITDA excludes stock-based compensation (SBC), depreciation, and amortization. For a capital-intensive business like a neocloud, depreciation is a real cost. GPUs depreciate rapidly – the NVIDIA H100 has a useful life of 3-5 years, but with the rapid release of Blackwell, the effective depreciation is likely shorter. If Nebius is using straight-line depreciation over 5 years, they are understating the true economic cost. The difference between adjusted EBITDA of $236.2 million and the net loss from continuing operations of $190.4 million is $426.6 million. That gap is primarily depreciation and amortization. Let me put that in perspective: the company is generating $236 million in cash flow before capital expenditures, but it is consuming $426 million in non-cash charges. This is not a problem – it is the nature of the business. But it means that the company’s ability to generate free cash flow depends on how much they reinvest.
Net Loss: The True Cost of Growth
The adjusted net loss of $33.2 million is a more honest picture. It strips out the most volatile items but still includes SBC and interest. The GAAP net loss from continuing operations of $190.4 million is the raw truth. This company lost $190 million in one quarter while generating $582 million in revenue. That is a -32.7% net margin. Growth is expensive. The market is worried that this loss will persist as capex ramps up.
Cash Flow: The Critical Missing Data
The article did not provide operating cash flow or free cash flow. This is a crucial omission. For a company investing billions in GPUs, free cash flow is the ultimate measure of sustainability. Based on the EBITDA and depreciation, we can estimate operating cash flow roughly equals EBITDA minus changes in working capital. If capex is, say, $1 billion per quarter (which is plausible given the growth), then free cash flow is deeply negative. This means Nebius is dependent on external financing – debt or equity – to fund growth. The stock drop may reflect fear of dilution.
The Stock Price Reaction: Reading the Market’s Mind
A 9.5% drop on a strong earnings beat is unusual. It suggests the market was expecting even more. The whisper number was likely higher than the reported $582 million. Or the market is looking at the net loss and the capex trajectory and concluding that the risk-reward is unfavorable. In a bull market for AI, investors are forgiving of losses. But they are not forgiving of slowing growth or rising capital intensity.
Contrarian: What the Bulls Got Right
Before I sound too bearish, let me give the bulls their due. The market is often short-sighted. Nebius has three structural advantages that are not fully priced in.
First, the team from Yandex is a genuine competitive moat. Building a neocloud is not just about buying GPUs. It is about building a scheduler that maximizes GPU utilization, a network that minimizes latency, and a billing system that handles complex usage patterns. Yandex engineers have been doing this at scale for over a decade. That is hard to replicate.
Second, the European market is underserved. US-based neoclouds like CoreWeave are focused on domestic demand. European AI startups face higher latency and regulatory hurdles if they use US clouds. Nebius is positioned to be the European champion. The regulatory tailwind from GDPR and data sovereignty is real.
Third, the sequential growth of 46% QoQ is not a fluke. It indicates that the company has solved the supply chain bottleneck. They are getting GPUs and deploying them fast. If they can maintain this momentum for another two quarters, the annualized revenue run rate will exceed $4 billion. At that scale, the EBITDA margin could expand to 50% or more, making the company profitable on a net basis.
Takeaway: The Accountability Call
Nebius is a bet on the AI infrastructure buildout. The Q2 numbers confirm the thesis, but they also expose the cost. The market is correct to ask: how much capital will this require? And at what point does the return on that capital justify the risk?
As an auditor, I look at the numbers and see a company that is executing well but is walking a tightrope. The revenue growth is real, but it is fueled by past capex. The EBITDA margin is high, but it masks the depreciation tsunami. The net loss is large, but it is typical for this stage. The stock drop is a signal that the market wants more clarity on the path to cash flow positivity.
The truth hides in the assembly, not the press release. The assembly of Nebius’s financials reveals a company that is growing faster than its peers but is also burning cash faster. The next two quarters will be critical. If they can show operating leverage – meaning revenue grows faster than capex – the stock will recover. If not, the market will continue to punish the stock.
Silence is the only honest consensus mechanism. For now, the market is silent, waiting for the next data point. I am watching the capex-to-revenue ratio. That is the number that will tell the true story.
Every exploit is a story poorly told. Nebius’s story is still being written. The plot twist depends on whether the company can convert its technical advantage into financial discipline. The code is written. The execution is underway. The market will judge.
