According to Woofun AI, AI cloud infrastructure giant Nebius (NBIS.US) disclosed its results for the second quarter ending June 30 on August 12. The financial report showed an extremely tense set of divergent data: revenue for the quarter recorded US$582.3 million, while cash expenses for the purchase of fixed assets and intangible assets reached US$5.657.4 billion.
The huge gap between this revenue schedule and the early deployment of GPUs and data centers revealed the truth about the financial mismatch behind the expansion of computing power. Although the company disclosed 4 iconic AI cloud contracts with an average total value of over 1 billion US dollars during the same period, contract signing, cash inflow, equipment landing, and revenue recognition all occurred asynchronously. Deeply dismantling the misalignment of these elements in the timeline is the key to understanding its financial health.
Judging from the composition of revenue and profit caliber, the business is growing rapidly, but the quality of profits needs to be carefully assessed. The Group's revenue jumped from US$227.7 million in the fourth quarter of 2025 to US$582.3 million this quarter, of which the AI cloud business contributed US$574.9 million, accounting for almost all revenue share.
It is worth noting that the $3 billion AI cloud ARR (annualized recurring revenue) disclosed by the company is not locked in revenue for the next 12 months, but is based on the annualized operating rate obtained by multiplying AI cloud revenue by 12 at the end of the quarter. According to calculations, ARR increased 56.3% month-on-month in the second quarter, slightly faster than the Group's overall revenue growth rate of 45.9%.
This difference is due to the fact that ARR only reflects the instantaneous speed of the AI cloud business at the end of the quarter, while revenue covers the entire group's confirmed quarterly total. The juxtaposition of the two is aimed at revealing the time difference between the pace of demand and revenue recognition. In terms of profit, non-GAAP adjusted EBITDA rose to US$236.2 million in the second quarter, but GAAP net loss from continuing operations still reached US$190.4 million. After the adjustment, EBITDA excludes non-cash items such as depreciation, amortization, and equity incentives. Although it can reflect operating performance, it cannot directly equate to bagged cash profits, and revenue growth does not fully cover construction costs.
According to data compiled by Woofun AI, the deep dismantling of the cash flow structure further revealed a serious mismatch between deferred income and capital expenditure. In the first quarter of 2026, the change in deferred revenue was still higher than capital expenditure, but after entering the second quarter, capital expenditure rose sharply, far exceeding the change in deferred revenue. The cumulative change in deferred revenue for the first half of the year was US$4.395 billion, while capital expenditure for the same period reached US$8.130.3 billion. What needs to be clarified is that the change in deferred revenue for the second quarter is estimated from the first half of the year. It is neither current revenue nor equivalent to all customer advance payments received during the quarter. Changes in deferred revenue affect operating cash flow, and the balance at the end of the period is balance sheet stock; the two cannot simply be added together as a deposit.
Although operating cash flow remained positive in the second quarter, based on the absolute value of cash outflow, capital expenditure was already about 9.7 times the current quarter's revenue. This model is similar to first investing in electricity, cabinets, and servers to build a shopping mall, and then waiting for customers to settle at the pace of use. Although customer cash can adjust the time distribution of funds, it is impossible to prove that every equipment investment has been covered by the customer's payment.
Prepayment coverage and recovery cycle estimates from the perspective of contract economics reveal the true resilience of the capital chain. The 4 iconic AI cloud contracts signed in the same quarter had an average TCV (total contract value) of more than 1 billion US dollars, but this only represents the size of the contract rather than confirmed revenue. More extensive data shows that around 70% of all transactions in the second quarter included customer advances, and these prepayments only covered 50% to 60% of the associated capital expenditure.
This ratio does not apply to all landmark contracts, let alone extrapolate to full capital expenditure. Nebius estimates that the payback period for capital expenses and operating costs corresponding to newly signed transactions is 1 year and 10 months, while the historical range is 2 to 3 years. The estimate relies on projected costs and future capacity that has been contracted but not built. The company disclosed a contract value of $20 million to $25 million per megawatt year, based on revenue recognition caliber and excluding advance payments. Capacity corresponding to most second-quarter transactions will be in place by the end of 2026, mainly contributing 2027 revenue, thus forming an asymmetric relay where advance payments first mitigate capital, equipment landing, and delayed revenue confirmation.
Clarification of forward-looking indicators helps avoid misinterpretation of growth figures. US$582.3 million is confirmed quarterly revenue up to the end of June, and US$3 billion is the annualized operating rate of the AI cloud business at the end of the quarter. The two describe the business that has already occurred, but the observation window is different. Customer commitments of more than $40 billion reflect subsequent performance and revenue streams, and according to the company's shareholders' letters, advance payments are expected to exceed $9 billion in 2026. This is management's forward-looking expectations rather than existing cash on the balance sheet.
Furthermore, 5GW is the contracted power target that is expected to be achieved at the end of 2026. It is supported by signed land and power commitments. It is not the GPU capacity currently online, nor can it be directly converted into the computing power that can be delivered during the season. Nebius' second-quarter earnings report indicates that the market should not simply add contracts, prepayments, equipment, and revenue to exaggerate growth. These factors are entering the same expansion chain at different speeds, and the time difference is the core variable.