Oracle Has A $638 Billion Backlog, And That Is Exactly What Worries Me

Oracle disclosed remaining performance obligations of $638 billion, up 363% year on year and $85 billion higher in a single quarter. The stock added around 5% on Monday. Before you decide that number is bullish, it is worth understanding what RPO actually is, because it is one of the most misread lines in software reporting.

What RPO is, and what it is not

Remaining performance obligations are contracted revenue that has not been recognised yet. It is a promise, not cash, and not profit. It tells you what customers have signed for. It tells you nothing about when the money arrives, what it costs to deliver, or whether the customer will still exist when delivery is due.

For a normal software company that distinction is minor, because delivering the next year of a subscription costs very little. For Oracle right now the distinction is the entire story, because the thing it has promised to deliver is AI compute capacity, and that has to be built first.

Who the backlog actually belongs to

Two details change how I read the $638 billion.

The first is concentration. Bank of America has noted that over half of the RPO comes from a single customer, OpenAI. A backlog where one counterparty is more than 50% is not a diversified order book, it is a bilateral bet on that counterparty's funding continuing.

The second is structure. Oracle disclosed that most of the RPO increase in the last two quarters came from large scale AI contracts where the customer either prepaid for GPU purchases or bought and supplied the GPUs to Oracle itself. That is unusual and it cuts both ways. It reduces Oracle's capital outlay on those specific deals, and it also tells you the economics are being negotiated in the customer's favour.

The cash gap

Here is the tension in plain terms. Revenue is recognised as capacity gets delivered. Capacity requires datacentres, power and chips, and those get paid for first. That is why the same company reporting a record backlog also went to the market to raise another $20 billion.

Growing backlog while raising debt to fund the delivery of that backlog is not automatically a problem. It is how infrastructure has always been financed. But it does mean the correct question is not how big the backlog is. It is what the return on invested capital looks like once you fund it, and that number is much harder to find than the $638 billion headline.

The part I do like

Oracle expanded its Alphabet partnership, putting Gemini across Oracle Cloud Infrastructure and Fusion Applications, which adds Google to a model lineup that already includes OpenAI, Meta and xAI. Shares rose over 6% on that news, and I think that reaction was better justified than the backlog reaction.

Being the neutral venue where every frontier model is available is a genuinely defensible position, and unlike the RPO figure, it does not require Oracle to fund a decade of capex to be worth something.

What I would need to see

I am not short this and I am not long it. I want two disclosures before the backlog means anything to me: the margin profile of the AI contracts specifically, and the customer concentration stated by the company rather than estimated by an analyst. Until then, $638 billion is a very large number attached to a very small number of customers.

Not investment advice. Do your own reading of the filings, because the interesting parts are never in the headline.