Oracle co-CEOs defend massive data-center expansion, plan to offer AI ecosystem

“We’re really in a unique situation to deliver what we call applied AI," said Mike Sicilia, the former president of Oracle Industries who was appointed as a CEO of the company last month. That includes infrastructure, analytics and applications, he said in an interview.

Clay Magouyrk, former president of Oracle’s cloud infrastructure business, was also appointed as a CEO. They take over just as concerns about an AI bubble are spreading quickly.

Oracle shares soared by more than 40% last month after the cloud company revealed that it had added $317 billion in future contract revenue during its latest quarter that ended in Aug. 31. Much of that new revenue came from a five-year, $300 billion deal with OpenAI, The Wall Street Journal reported.

Investors and tech analysts in the last month have expressed concerns about the extent to which the new build-out depends on OpenAI, which won’t generate a profit until 2029, according to the ChatGPT maker’s chief executive, Sam Altman. Credit-ratings firm Moody’s last month highlighted risks to Oracle’s balance sheet due to how much its future AI data centers rely on OpenAI.

Earlier this month, Oracle shares fell as much as 7.1%—though they later rebounded—following a report suggesting that its margins on renting specialized Nvidia chips were razor thin.

Sicilia and Magouyrk are preparing to make their case to investors this week, including on Thursday’s investor day, for how Oracle’s data-center expansion will provide a clear path to profitability.

The ecosystem and inference argument

The idea of selling an ecosystem of AI solutions to large enterprises is the long-term Oracle playbook, said Balaji Abbabatulla, an analyst at market research and IT consulting firm Gartner.

“They’re not going to be able to show clear returns if they don’t go for those large and multibillion-dollar deals," Abbabatulla said, “and the only way to get there is to go beyond the components, put them together into the solutions."

That means linking together Oracle’s AI infrastructure capabilities with its applications like enterprise resource planning and human resources software, plus its databases, to sell a package that makes Oracle the tech vendor of choice for large corporate clients.

Questions also remain over where the training of AI models is headed. While most AI infrastructure powers the building of these models, the tide will soon turn to AI inference, or the actual usage of the models where users send prompts and receive outputs from them.

Providing the computing for inference, rather than just AI training, is a durable business, analysts say. But it’s unclear if Oracle can succeed at both.

Magouyrk said Oracle has a huge opportunity to help clients “do their inferencing right alongside their data with the best models." The company’s new AI Data Platform will allow for that, he said, predicting that customers’ AI usage will increase as much as one thousand times once they start using the platform.

Shawnna DelHierro, CIO of SoundHound AI, said the AI voice company is using Oracle’s cloud infrastructure for both training its AI models and inference. The company, which also uses some of Oracle’s back-office software, handles over one billion queries a month using Oracle’s cloud, DelHierro said. SoundHound initially chose Oracle as a cloud vendor because the company was looking for a “true partner" and a platform that could provide zero latency, she added.

At the moment, most enterprises say they don’t see the immediate benefit of their AI investments even as they continue to spend precious tech dollars on the technology.

Pressured margins and debt

Part of Oracle’s challenge is facing a belief among investors that renting out access to advanced AI chips and building data centers are a low-margin business. That’s especially true for Oracle, which has staked part of its reputation on being a lower-priced cloud competitor.

Magouyrk said margins are “the wrong way to look at the business." “I understand the economics of each marginal unit and how this works out as it scales," he said, “and that actually ends up being a very profitable business."

Derrick Wood, an analyst at TD Cowen, said margins are going to be lower at the beginning of an AI infrastructure build out.

“You have to go build the infrastructure before you can turn on all the revenue meters," he said. “But as the consumption meters start going on, you start to recoup a lot more of your capital expense and start to see gross margins significantly improve."

Oracle has also taken on debt to finance its AI infrastructure build-out, selling $18 billion of investment-grade bonds in late September. The question of financing has troubled investors, who are hoping Oracle can explain exactly how much capital it needs for data center projects like its Stargate plan with OpenAI.

Focusing just on debt paints a very scary picture, Magouyrk said, but looking at the combination of new deals, projected revenue and cash flow presents a rosier one.

In terms of new deals, Magouyrk said Oracle strikes many contracts worth hundreds of millions, but they often go unnoticed. Rather than putting all of its eggs in one customer basket—or rather than solely relying on OpenAI as its primary AI customer—Oracle has a more varied business than many realize, he added.

“Pretty much all of the big model providers use our cloud in one form or another, so I can’t be more diversified than the entire market is. You can’t get more than 100% of the Pokémon," Magouyrk said.