Neocloud's Signal: Nebius Earnings Validate the GPU-as-a-Service Thesis, But the Risk Lies in the Unit Economics

On August 13, Neocloud stocks surged. IREN climbed over 5%. Nebius and Coreweave each rose more than 3%. The catalyst: Nebius's Q2 earnings. Revenue hit $582.3 million, a 454% year-over-year increase. Adjusted EBITDA turned positive at $236.2 million. The market responded with a 34% single-day jump—the largest since September last year. These numbers are not just growth; they are a verification event. The thesis that AI cloud infrastructure can scale profitably is now backed by data. But data is not truth. It is a signal that requires decoding. As a smart contract architect who has spent years analyzing protocol economics, I look beyond the top-line euphoria. The real story is in the unit economics, the capital efficiency, and the hidden assumptions behind the revenue. The Neocloud sector sits at the intersection of AI and blockchain infrastructure. It comprises companies that provide GPU compute for AI training and inference, often repurposing assets from cryptocurrency mining. IREN, formerly Iris Energy, is a Bitcoin miner that has pivoted to AI cloud using its low-cost power assets. Nebius, spun off from Yandex's European division, positions itself as an AI-native cloud provider with a full toolchain. Coreweave is a pure-play GPU cloud, deeply integrated with NVIDIA. The sector has attracted massive capital because AI demand is growing exponentially. But the infrastructure is capital-intensive. The key metric is not revenue growth alone, but the return on invested capital. The market is pricing these stocks as if they are the next AWS. But AWS had a decade of infrastructure investment and a moat. Neocloud companies are racing to build capacity before demand materializes. This is a high-stakes game. Let me dissect the numbers. Nebius reported Q2 revenue of $582.3 million, up from $105.1 million a year earlier. That is a 454% increase. Adjusted EBITDA came in at $236.2 million, a stark contrast to the negative EBITDA in the prior year. Net loss narrowed to $33.2 million, a 64% improvement. The implied EBITDA margin is 40.6%. On the surface, this is a textbook turnaround. The company has crossed the inflection point where revenue growth outpaces operating costs. But I have seen this pattern before. During the 2020 DeFi summer, I audited a yield aggregator that posted 500% revenue growth in a quarter. The team was celebrating. Three months later, a flash loan attack drained the liquidity pool. The growth was driven by a single whale that was using the protocol for arbitrage. Revenue was not recurring. Nebius does not disclose customer concentration in the press release. I have seen the same opacity in dozens of DeFi audits. The SEC filings will reveal the truth. If the top five customers account for more than 40% of revenue, the risk is real. Now, let's examine the cost structure. EBITDA excludes depreciation, interest, and taxes. In a capital-intensive business like GPU cloud, depreciation is the largest cost. GPUs have a lifespan of three to five years. A single NVIDIA H100 costs $30,000. If Nebius deploys 100,000 GPUs, that is $3 billion in capital expenditure. The annual depreciation at 33% is $1 billion. To generate $236 million in quarterly EBITDA, the annual EBITDA is $944 million. That leaves a $56 million deficit after depreciation. But the company still has a net loss. The path to free cash flow is narrow. The market is betting on continued revenue growth to cover the depreciation. But if revenue growth slows, the depreciation burden will crush earnings. I have seen this exact dynamic in the Bitcoin mining industry. In 2022, when Bitcoin prices fell, miners with high CapEx were forced to sell hardware at a loss. The same fate awaits Neocloud companies if demand softens. Let's compare IREN. IREN is a Bitcoin miner that has pivoted to AI cloud. Its core advantage is low-cost power. IREN operates hydro-powered mining sites in Canada and the US. The electricity cost for Bitcoin mining is around $0.03 per kWh, which is also ideal for GPU compute. The company has announced a 100 MW GPU data center in Texas. But the scale is small compared to Nebius. IREN's AI cloud revenue is likely less than $50 million in Q2. The stock rose 5% on the news, but that is pure sentiment. The real value of IREN is its optionality. If the AI cloud business succeeds, it could double the valuation. But if it fails, the company is still a Bitcoin miner with a volatile revenue stream. The market is pricing in the call option but ignoring the risk. I have seen this in the NFT space. Projects with a mining pivot often trade at a premium before the pivot is validated. The validation comes from customer contracts, not price action. Without disclosed contracts, the 5% move is noise. Coreweave is the most aggressive player. It is built on leverage. The company has raised billions in debt, using GPUs as collateral. In 2023, it secured a $2.3 billion debt facility from Magnetar Capital. The thesis is that GPU demand will outstrip supply, so the rental income will cover the debt service. But the debt is floating-rate. If interest rates stay high, the interest expense eats into margins. Coreweave's IPO in March 2025 valued it at $230 billion. That is a 50x multiple on trailing revenue? The company has not disclosed its Q2 2025 numbers yet. But the market is already pricing in perfection. The risk is that the debt structure is a time bomb. If GPU utilization drops below 70%, the company cannot cover its debt payments. The same risk exists in the crypto lending market. In 2022, Celsius Network used customer deposits to fund yield-generating strategies. When the market turned, the leverage collapsed. Coreweave is a regulated company, but the math is the same. The market is ignoring the tail risk. The broader sector is subject to what I call "hashrate inflation." Every Neocloud company is ordering GPUs. NVIDIA's supply is constrained, but the total GPU capacity is increasing. The combined CapEx of IREN, Nebius, Coreweave, and others is likely to exceed $50 billion in 2025. If AI demand grows at 50% per year, but GPU capacity grows at 100%, the unit price of compute will fall. This is a classic commodity cycle. The revenue of each company is a function of price times quantity. If quantity increases but price falls, the revenue growth may be an illusion. The market is treating GPU compute as a growth asset, but it is a commodity. The long-term equilibrium is a low-margin business. The only way to sustain high margins is through differentiation. Nebius has an AI-driven platform. Coreweave has a close relationship with NVIDIA. But both are still selling raw compute. The software layer is thin. I have seen this in the cloud computing industry. AWS, Azure, and GCP have deep moats because of their platform services. Neocloud companies are building the infrastructure layer, which is the most vulnerable to commoditization. Now, let's apply the forensic lens. Every audit report is a promise, not a guarantee. The same applies to earnings reports. The Q2 numbers for Nebius are unaudited. The full-year audit may reveal adjustments. I have seen cases where companies capitalize operating expenses to inflate EBITDA. For example, if Nebius is capitalizing the salaries of its cloud engineers as part of CapEx, the EBITDA margin is artificially high. The footnotes will tell the story. Investors should look at the ratio of CapEx to depreciation. If CapEx is significantly higher than depreciation, the company is incurring costs that are not yet hitting the income statement. This is a red flag. In the crypto world, we call this "hidden liquidity risk." The same principle applies here. Another blind spot is the regulatory risk. Nebius has a Russian heritage. The founder, Arkady Volozh, was sanctioned by the EU in 2022. The sanctions were lifted in 2024, but the geopolitical risk remains. If the US imposes secondary sanctions on entities that do business with Russia, Nebius could be cut off from NVIDIA's supply chain. The company's data centers are in Europe, but the parent company is registered in the Netherlands. The US has extraterritorial jurisdiction. The market is ignoring this risk because it is a tail event. But tail events happen. I have seen them in the crypto space. The Terra collapse was a tail event. The FTX collapse was a tail event. The market is pricing in a smooth growth trajectory, but the volatility is real. Let me share a personal experience. In 2021, I analyzed the NFT standard ERC-721A. The protocol claimed a 40% gas reduction. I spent months verifying the math. The reduction was real, but only for batch minting. For single minting, the gas cost was higher. The team had cherry-picked the metric. The same happens in earnings reports. Nebius highlights the 454% revenue growth and the EBITDA turnaround. But it does not mention the customer concentration, the CapEx intensity, or the debt maturity profile. The yield is a function of risk, not just time. The market is discounting the risk because the story is compelling. But the story is always compelling before the crash. Liquidity is just trust with a price tag. The Neocloud sector is built on trust: trust that NVIDIA will deliver GPUs, trust that AI demand will grow, trust that the capital will be available. This trust is priced into the stocks. But trust can evaporate. The moment a major customer cancels a contract, the trust chain breaks. The sector will experience a re-rating. The question is not if, but when. The contrarian view is that the sector is overvalued on a risk-adjusted basis. The market is pricing in a 10-year growth trajectory in a 2-year stock price. The history of technology cycles shows that the early winners are often disrupted by later entrants. The Neocloud companies are the infrastructure layer. The real value will accrue to the application layer. The box sellers rarely win. What does this mean for an investor? The sector is a high-beta play on AI. The returns will be volatile. The safe play is to wait for the next correction. The risk takers will buy the dip. But the dip may not come until the next earnings miss. I have seen this pattern in the DeFi sector. The tokens that performed best in the bull run were the ones with the most aggressive tokenomics. The same applies here. The companies with the highest leverage will be the most volatile. Coreweave could double or fall by 50% in a quarter. That is not an investment; it is a gamble. In conclusion, the Neocloud sector is a technological marvel but a financial minefield. The Nebius earnings are a positive signal, but they are not a guarantee. The true test will come in the next two quarters as capital expenditures ramp up and customer contracts are revealed. Investors should watch two metrics: capital efficiency (revenue per dollar of CapEx) and customer concentration. The companies that can grow without diluting unit economics will win. The rest are just riding the wave. In the words of an auditor: "The code is the law, but the balance sheet is the reality." The code of the blockchain is immutable, but the balance sheet of a company is mutable. The auditor's report is a promise. The market is buying promises. I am buying time. The next 12 months will separate the players with real customer contracts from those with speculative capacity. The metric that matters is not revenue growth, but capital efficiency. The yield is a function of risk. The risk is not priced in. That is the signal.