Problems With Oracle’s AI Growth?

Problems With Oracle’s AI Growth?

https://www.thegoodinvestors.sg/problems-with-oracles-ai-growth/

Chong Ser Jing

Posted on September 11, 2025

Categories Investing Analysis

Oracle’s management is projecting a 14x increase in AI revenue over the next five years. But the picture is not as rosy as it seems.

Oracle Corporation’s (NYSE: ORCL) management stunned stock market participants earlier this week during the company’s conference call for the release of its first-quarter earnings for FY2026 (fiscal year ending 31 May 2026). Management announced stupendous future growth for Oracle’s Cloud Infrastructure business, driven by an enormous increase in RPO (remaining performance obligations) because of AI-related demand:

“We have signed significant cloud contracts with the who’s who of AI, including OpenAI, xAI, Meta, NVIDIA, AMD and many others. At the end of Q1, remaining performance obligations, or RPO, now to [US]$455 billion. This is up 359% from last year and up [US]$317 billion from the end of Q4. Our cloud RPO grew nearly 500% on top of 83% growth last year…

…The enormity of this RPO growth enables us to make a large upward revision to the Cloud Infrastructure portion of our financial plan. We now expect Oracle Cloud Infrastructure will grow 77% to [US]$18 billion this fiscal year and then increase to [US]$32 billion, [US]$73 billion, [US]$114 billion and [US]$144 billion over the following 4 years. Much of this revenue is already booked in our [US]$455 billion RPO number, and we are off to a fantastic start this year.”

For context, Oracle ended FY2025 with total revenue of US$57.4 billion, and Cloud Infrastructure revenue of merely US$10.2 billion. The newly expected windfall for Cloud Infrastructure drove Oracle’s stock price 36% higher the day after its FY2026 first-quarter earnings.

But when I looked at the details of Oracle’s RPO and financials, I found potentially serious problems with the company’s AI-growth story.

Problem 1: Risky customer?

During the earnings conference call, Oracle’s management did not name the customers responsible for the massive increase in the company’s RPO. But a subsequent article from the Wall Street Journal revealed that OpenAI had recently signed a US$300 billion, five-year deal with Oracle – in other words, nearly 95% of Oracle’s sequential US$317 billion increase in RPO in the first quarter of FY2026 came from just OpenAI.

Intense customer-concentration alone can be a headache for any company. But when the customer is itself burning lots of cash, it can be a thunderclap headache. OpenAI’s leaders expect the company to earn around US$13 billion in revenue this year, but its deal with Oracle works out to an annual average spend of US$60 billion. Moreover, The Information reported earlier this month that OpenAI’s leaders are now forecasting significantly higher cash burn over the next few years than recently expected:

“OpenAI projected its cash burn this year through 2029 will rise even higher than previously thought, to a total of [US]$115 billion. That’s about [US]$80 billion higher than the company previously expected…

…The company projected it will burn more than [US]$8 billion this year, or roughly [US]$1.5 billion higher than its prior projection from earlier this year. Cash burn will more than double to more than $17 billion next year—[US]$10 billion higher than what the company earlier projected.

And in 2027 and 2028, the company projects to burn roughly [US]$35 billion and [US]$45 billion, respectively. In the prior projection, the company said its 2028 cash burn would be [US]$11 billion, meaning the new estimate is more than four times higher.”

OpenAI’s spending plans with Oracle will have to depend on the largesse of would-be investors and lenders, so there’s no guarantee that OpenAI will have access to funding in the future. In the meantime, Oracle will have to procure the AI hardware (mostly AI chips) ahead of time. This brings me to the second potential problem.

Problem 2: Risky finances?

Purchasing AI hardware requires capital. Lots of capital. And Oracle’s not in the best financial shape for this.

As of 31 August 2025, Oracle had US$11.0 in cash and marketable securities, but a staggering US$91.3 billion in debt, giving a high net-debt position of US$80.3 billion. If Oracle’s operating lease liabilities are included, the net-debt position rises further to US$94.4 billion. Oracle’s trailing operating cash flow and net income are US$21.5 billion and US$12.4 billion, respectively. Using the lower net-debt figure gives Oracle net-debt-to-operating-cash-flow and net-debt-to-net-income ratios of 3.7 and 6.5. These ratios suggest Oracle is unable to increase its debt significantly without risking its financial health. To be clear, the ratios are high not because Oracle’s trailing operating cash flow and net income are temporarily compressed; Table 1 below shows Oracle’s operating cash flows and net incomes for FY2021-FY2025.

Year	Oracle operating cash flow (US $ billion)	Oracle net income (US $ billion)
FY2021	15.9	13.7
FY2022	9.5	6.7
FY2023	17.2	8.5
FY2024	18.7	10.5
FY2025	20.8	12.4

Oracle’s management was asked during the FY2026 first-quarter earnings conference call about the capital expenditures needed to fulfill the company’s RPO. Management was coy and suggested that Cloud Infrastructure’s projected growth would happen in an asset-light way:

“As I mentioned in the prepared remarks, and as I’ve said very clearly beforehand, we do not own the property. We do not own the buildings. What we do own and what we engineer is the equipment. And that’s equipment that is optimized for the Oracle Cloud. It has extremely special networking capabilities. It has technical capabilities from Larry and his team that allows us to run these workloads much, much faster. And as a result, it’s much cheaper than our competitors. and depending on the workload.

Now because of that, what we do is we put in that equipment only when it’s time and usually very quickly, assuming that our customer accepts it, we’re already generating revenue right away. The faster they accept the system and that it meets their needs, the faster they start using it, the sooner we have revenue. This is, in some ways, I don’t want to call it asset-light from the finance world, but it’s asset pretty light.”

I disagree with management’s “asset pretty light” characterisation. Earlier, I mentioned that Cloud Infrastructure’s revenue was expected to increase from US$10.2 billion in FY2025 to US$18 billion in FY2026. During the earnings conference call, management projected US$35 billion in capital expenditure in FY2026, up 65% from US$21.2 billion in FY2025. I think it’s reasonable to assume that most of the US$35 billion in expected capital expenditure for FY2026 will be for the Cloud Infrastructure business, so we’re looking at a capital-expenditure-to-revenue-ratio of nearly 2 (US$35 billion over US$18 billion). That’s hardly “asset pretty light”.

Exacerbating the problem for Oracle is that its operating cash flow in FY2025 was just US$20.8 billion, meaning it had negative free cash flow during the year. Unless Oracle’s operating cash flow increases by nearly 70% in FY2026, the company will have to raise capital externally for its projected capital expenditures. I already mentioned that Oracle’s heavy net-debt position is an obstacle to any large future increases in debt. This said, issuing shares could work, given Oracle’s current market capitalisation of US$922 billion. Oracle’s high price-to-earnings (P/E) ratio of 76 also makes issuing shares a palatable option. Nonetheless, there could still be material dilution given the potentially significant capital expenditures needed to support Oracle’s RPO. Coming back to the possibility of Oracle’s operating cash flow increasing by nearly 70% in FY2026, I think it’s very, very unlikely because of the lower margin of Cloud Infrastructure, which brings me to the third potential problem.

Problem 3: Margin pressure?

Cloud Infrastructure has been Oracle’s fastest-growing business in the past few years. Table 2 shows the changes in Cloud Infrastructure revenue and Oracle’s total revenue for FY2023-FY2025.

Year	Cloud Infrastructure revenue (US $ billion)	Oracle total revenue (US $ billion)
FY2023	4.5	50.0
FY2024	6.9	53.0
FY2025	10.3	57.4

Cloud Infrastructure revenue is likely all reported under Oracle’s Cloud services and license support segment. What has happened over the same period shown in Table 2 is that the Cloud services and license support segment’s operating expense has grown much faster than its revenue, as illustrated in Table 3, suggesting that Cloud Infrastructure is a lower-margin business for Oracle.

Year	Cloud Infrastructure and license support revenue growth	Cloud Infrastructure and license support operating expenses growth
FY2023	17%	49%
FY2024	12%	21%
FY2025	12%	23%

This brings into question how much Oracle’s net income and cash flow can benefit from the rapid projected-growth in Cloud Infrastructure revenue. If Cloud Infrastructure’s revenue indeed grows as management expects, there’s no doubt that Oracle’s net income will grow – but to what extent remains to be seen. It’s worth noting that with Oracle’s shares carrying a P/E ratio of 76 at the moment, the market is expecting stellar net income growth.

Conclusion

Larry Ellison, Oracle’s founder, chairman, and chief technology officer, once said:

“Why do we do these things? George Mallory said the reason he wanted to climb Everest was because it’s there. I don’t think so. I think Mallory was wrong. It’s not because it’s there. It’s because we’re there, and we wonder if we can do it…

…So how do I get off this merry-go-round? How do I stop when I’m winning? It’s hard for me to quit when I’m losing, and it’s hard for me to quit when I’m winning. It’s just hard for me to quit. I’m addicted to competing.”

I wouldn’t count out any business leader with such a ferocious competitive spirit. But there are potential problems with Oracle’s AI-growth story, namely, (1) high revenue-concentration from a risky customer in OpenAI, (2) having a debt-laden balance sheet while having to invest heavily in AI chips, and (3) margin-compression from lower-margin AI-related services. I wonder how this will all work out.