The questions asked during Oracle’s Q3 FY26 earnings call is indicative of the legitimate concerns investors have over whether spending vast sums of money on building advanced datacentres with the latest artificial intelligence (AI) acceleration hardware will deliver a return on investment (ROI). This is a concern not only for those who invest in tech stocks and shares, but also IT and business leaders, who the hyperscalers are relying on to repay the IT infrastructure costs they are incurring.
Oracle posted cloud infrastructure revenue of $4.9bn, up 84% from last year, while its cloud applications business reported revenue of $4bn, a 13% increase.
The company is committed to spending $533bn on AI infrastructure to meet what it calls its remaining performance obligation (RPO), which is effectively the volume of compute capacity it will require to fulfil customer contracts.
Last month, the company announced that to meet the contracted demand from its largest Oracle Cloud Infrastructure customers, including AMD, Meta, Nvidia, OpenAI, TikTok, and xAI, it expects to raise $45-50bn of gross cash proceeds during the 2026 calendar year. It plans to achieve this using a combination of debt and equity financing.
But along with growing debt, Bloomberg recently reported that the company is planning thousands of job cuts as it drives forward its datacentre expansion plans to support AI workloads.
The company claims AI models for generating computer code have become so efficient that it has restructured product development teams into smaller, more agile and productive groups, enabling it to build more software in less time with fewer people.
During the earnings call, Oracle co-CEO Mike Cecilia said: “The use of AI coding tools inside Oracle is enabling smaller engineering teams to deliver more complete solutions to our customers more quickly. We are building brand-new SaaS [software as a service] products using AI, and also embedding AI agents right into our existing applications suites.”
Datacentre profitability
When asked about the profitability of Oracle’s AI datacentres, in terms of the AI acceleration hardware investment the company is making, Clay Magouyrk, who heads Oracle Cloud Infrastructure, said: “We see gross margins in the 30-40% range.”
According to Magouyrk, Oracle customers are increasingly interested in combining the best AI models with their private data in a secure and private manner. He claimed that this approach is growing in popularity over training their own large language models.
“In the early days, a lot of people thought that most customers would be doing very specific training of their own large language models,” said Magouyrk. “I think that has largely proven to not be the case. Instead, what I think is incredibly popular and growing in popularity is people taking the best models and wanting them to combine that in a private way with their private data.”
But like other cloud providers, Oracle needs to ensure its customers run AI in the company’s own cloud infrastructure. He said Oracle has introduced new funding models, such as “bring your own hardware” and upfront customer payments, which he said allows Oracle to expand infrastructure without raising additional debt or issuing equity. According to Magouyrk, this model has already resulted in $29bn in contracts.
The question Oracle and other public cloud providers need to answer is whether this growth is realistic, and whether it is sustainable or just industry hype. According to analyst Gartner, enterprises are becoming more knowledgeable and realistic about the capabilities of AI technologies, moving beyond hype to focus on business value.
Gartner’s analysis of hundreds of use cases reveals that while most agentic AI deployments (55%) are falling short on ROI due to inconsistent innovation across the agentic AI value proposition, a significant segment (40%) is achieving solid business outcomes through steady advances.
Speaking to Computer Weekly about the return on investment analysis, Gartner senior director analyst Roberta Cozza said the research shows that 5% of use cases for AI delivered “transformational return on investment”.
She said that these innovators are not only exceeding ROI expectations, but are also unlocking transformational value for enterprises, including new revenue streams and competitive advantages. “These standout competitors are poised to build lasting competitive moats,” she added.
However, according to Cozza, while they are impactful, these transformational use cases are rare. Gartner is focusing its research to understand what made these specific cases successful. What this suggests is there is uncertainty over how AI will deliver transformational change in organisations.
In December 2025, an article by investment firm BNY Mellon quoted figures from UBS and JP Morgan, showing $1.5tn of AI-related debt could strain corporate bond markets, heightening volatility. Senior investment strategist Theodore Bair Jr wrote: “Parallels to the dot-com and telecom buildouts suggest risk if infrastructure returns lag expectations.”
Is AI a tech bubble that will burst? Given that Gartner’s figures show that over half of AI deployments are not fully meeting the expectations set out in the business case to justify the cost of investing in these systems, IT and business leaders need to be realistic about how much they are prepared to spend.
