#甲骨文Q1业绩超预期盘后涨超5% The narrative that “AI computing power is a bubble” has been personally shattered by Oracle’s $664 billion order backlog—when customers are lining up for years, the ending has already been written.

Oracle’s order backlog has reached $664 billion—this figure is the strongest confirmation yet of demand for AI computing power. Tonight, Oracle released an earnings report that kept me staring at it for a long time. Not because the numbers were especially beautiful—but because one figure made me realize that demand for AI computing power may last far longer than most of us estimate.
That figure is RPO—remaining performance obligations. $664 billion. This is the value of contracts Oracle has already signed but has not yet fulfilled.
In other words: $664 billion of business is currently queued up for Oracle to handle. Let me explain this earnings report from start to finish, because many people may have heard of Oracle but have not followed it closely.
First, let’s lay out the key figures from today. Oracle reported fiscal first-quarter 2027 revenue of $19.35 billion, up 30% year over year. Cloud revenue grew 62% to $11.6 billion, including cloud infrastructure revenue growth of 121% to $7.4 billion, and cloud applications revenue growth of 10% to $4.2 billion. Oracle reported adjusted earnings per share of $1.92, up 30% year over year. Analysts had forecast $1.74. Revenue of $19.35 billion, up 30% year over year—that is already a very good figure. But what truly sent the market into a frenzy was not that, but the 121% growth in cloud infrastructure. Cloud infrastructure revenue growth accelerated from 52% in fiscal Q4 2025 to 121% this quarter, marking the ninth consecutive quarter of acceleration. This unprecedented growth reflects strong demand for AI and GPU computing capacity. Oracle reported that utilization of its AI infrastructure GPUs reached 97.9%, while GPU renewal and resale contracts carried an average premium of 20%. Nine consecutive quarters of accelerating growth—not nine consecutive quarters of growth, which would already be difficult, but nine consecutive quarters in which the growth rate increased. From 52% to 121%, it ran faster every quarter than the one before. This is extremely rare in business history; almost no company can maintain this pace. GPU utilization of 97.9% means that virtually none of Oracle’s AI computing capacity is sitting idle—all of it is being fully used by customers. If your capacity utilization is at 100%,
you have only one next step: expand capacity. Then comes the most important figure—RPO.
Remaining performance obligations (RPO) increased by $209 billion year over year to $664 billion.
Remaining performance obligations (RPO) increased by $209 billion year over year to $664 billion. Oracle’s remaining performance obligations reached $664 billion, including more than $30 billion in new AI cloud contracts. What does $664 billion mean? It is equivalent to roughly three years of Oracle’s total revenue.
In other words: even if Oracle signs no new contracts starting today, it already has three years of revenue locked in simply by fulfilling its existing contracts. Moreover, more than $30 billion of that $664 billion consists of AI cloud contracts added in the latest quarter—signed this quarter, representing fresh incremental business rather than accumulated historical backlog. AI customers are locking in computing capacity in advance through increasingly large, long-term contracts. They are not conducting trials; they are locking in capacity at scale, in advance, and for the long term. Tonight’s stock-price reaction also contained a very interesting little story. After the presentation was released, the stock rebounded to $159.58 in after-hours trading, up 4.34% from the intraday closing price of $152.94, after having fallen 5.38% earlier during regular trading. It first fell 5.38% intraday and then rose 4.34% after hours—at its peak, it was up more than 7%—and this reversal was the most interesting scene in today’s market.
Why did it fall first? Oracle expects net capital expenditures of approximately $70 billion for the current fiscal year, ending next May, primarily for infrastructure expansion, and the company also completed a $20 billion equity offering. Capital expenditures of $70 billion exceeded expectations, and the market’s initial reaction was: this burns too much cash, and near-term free cash flow is negative. Oracle reported record operating cash flow of $23 billion, but due to continued investment in expanding cloud infrastructure, free cash flow was negative $5 billion. Negative $5 billion means Oracle is expanding much faster than its cash inflows, requiring it to rely on stock issuance ($20 billion) to cover the funding gap.
But the market thought it over and decided not to care. The $664 billion RPO is there, the 97.9% GPU utilization is there, and the 121% growth rate for nine consecutive quarters is there. You are spending $70 billion to build data centers because you have $664 billion in orders waiting for you; spending money to build capacity is therefore a completely rational business decision. So the stock reversed after hours and came back up.
This earnings report matters more to the entire AI sector than to Oracle itself.
Over the past few months, the market has had one persistent concern: with oil above $100, interest rates rising, and the September curse—could AI demand begin to contract in such a poor macro environment? Could companies reduce their purchases of computing capacity because money is tight? Oracle’s earnings report tonight gave a very clear answer with the data: no. This quarter, Oracle added data center capacity for more than 300k GPUs, three times the previous quarter, and that capacity began operating at a 97.9% utilization rate as soon as it came online. This is not demand cooling; there is no sign of demand cooling at all. It is even accelerating further. Another detail is very important: growth in cloud infrastructure revenue was supported by new large-scale computing projects and strong demand for computing and database services. Demand for database services is also growing—which means AI demand is not limited to the GPU computing layer; demand for infrastructure layers such as databases, storage, and networking is expanding in parallel. This is a signal of “full-stack demand expansion,” not a flare-up at just one point.
What does this earnings report mean for ordinary people?
First, if you hold Nvidia—Oracle added 300k GPUs, but where did those GPUs come from? Most came from Nvidia. Oracle’s expansion is a direct source of orders for Nvidia. Tonight’s Oracle earnings report is a positive indication for Nvidia’s Q3 revenue growth.
Second, if you are considering holding Oracle—the company’s business logic is undergoing a fundamental transformation: from “selling traditional enterprise software” to “providing AI computing infrastructure.” An RPO of $664 billion means extremely high revenue visibility for the next few years, and that certainty is a highly valuable asset in a high-interest-rate environment. But note that $70 billion in capital expenditures + negative $5 billion in free cash flow + $20 billion in equity financing means it is currently a machine that is “expanding at high speed and burning cash rapidly,” so its short-term financial pressure is significant.
Third, for the entire AI cloud computing sector—Oracle’s data tells you that demand for AI infrastructure will not disappear because oil is expensive or interest rates are high. Technology giants are using long-term contracts and advance bookings to ensure they do not fall behind in the AI computing arms race. This is a structural assessment of demand, not a short-term emotional fluctuation.
Fourth, the one figure most worth remembering: 97.9% GPU utilization. This means virtually all of Oracle’s existing computing capacity is being fully used.
Under these conditions, it is still expanding at a rate of 300k GPUs per quarter, yet demand continues to exceed supply. This is the true state of the AI computing shortage—not “there is enough computing capacity,” but a positive cycle of “build as much as is used, queue as much as is built.” Oracle cloud computing +121%, RPO of $664 billion, GPU utilization of 97.9%—this is the strongest evidence that demand for AI computing power has not cooled and is accelerating. The stock first fell 5% intraday and then rose 7% after hours—the market initially feared capital expenditures, but after thinking it through, it saw the certainty behind the $664 billion.
Do you think Oracle’s transformation from an “old-line software company” into an “AI computing platform” is the most underestimated story of the year? Share your view in the comments.$ORCL