Oracle and the hard truths about software

RUNNING A GIANT software business used to be fun. Sure, coming up with a great product was a grind. But once you had one that customers could not live without—be it Microsoft Office, Amazon.com, Google search, Facebook or an Oracle database—you could lie back and watch your unit costs fall, your profit margins rise and cash flow in. Debt was optional. Capital spending was an afterthought.

Artificial intelligence is now turning this cosy business model on its head. Capital expenditure has soared as America’s hyperscalers—Alphabet, Amazon, Meta and Microsoft—and, more recently, Oracle have poured money into data centres crammed full of pricey AI chips. The combined capital expenditure of the five tech companies reached $313bn in the 12 months to the end of June, up from $146bn in the same period three years before. Amazon’s free cashflow (the proceeds left after subtracting operating and capital expenditures) was cut in half over the past year. In October Meta borrowed $30bn in the bond market.

Nowhere has the transition been as unsettling as at Oracle. It was founded 48 years ago to sell “relational” databases, which let businesses cross-reference their digital information with ease. For the first 45 of those years it did mostly that, first in client-owned server farms, then in the computing cloud. In the decade to early 2022 its capex rarely exceeded 5% of sales. Gross margins hovered around 80%. Free cashflow stood at a reliable $12bn or so a year.

The company that reported its latest quarterly results on December 10th looks almost unrecognisable. And that is not just because in September Oracle’s long-time chief executive, Safra Catz, stepped down and was replaced by a duo of successors, Clay Magouyrk and Mike Sicilia.

Revenue in the three months to the end of November rose by 14%, year on year, to $16bn. But capex ate up two-thirds of those sales. Free cashflow turned deeply negative. Net debt grew by $11bn, to $88bn. Gross profit is nowadays closer to 70% of sales rather than 80%. Return on capital, which reached 13.5% months before ChatGPT was released in November 2022, has fallen below 10%. Were it not for the steadfast presence of Larry Ellison, its preternaturally youthful 81-year-old founder, chairman and chief technologist, you might have thought you had dialled into the wrong earnings call.

To put those changing finances in perspective, consider how Oracle stacks up against the 1,000 biggest non-financial firms listed on the rich world’s top bourses along six measures of corporate performance: operating and free-cashflow margins; operating profit to interest expense; capex to revenue; total debt to operating profit (before depreciation and amortisation); and return on capital. Then run a regression to find the closest match to Oracle’s profile across these six variables.

Ten years ago the answer was Amgen, an American biotech giant. The remaining top ten doppelgangers included other innovators (Pfizer and AbbVie in life sciences and Broadcom in chipmaking), reliable services firms (Equifax in credit scoring, Roper Technologies in business software and Verisk Analytics in consulting) and a few industrial businesses. Today Oracle is a spitting image of Antofagasta, a London-listed Chilean copper miner, followed by nine American utilities.

Investors are not sure what to make of it all. Many applauded Oracle’s giant AI gamble three months ago, when it bagged a $300bn, five-year contract to supply computing power to OpenAI, the maker of ChatGPT and poster child of the AI boom. Oracle’s market value shot up by $255bn overnight. Days later it briefly flirted with $1trn. Since then it has shed all those gains, and then some. The disappointing pace of AI adoption has put a question mark on the ability of OpenAI and Oracle’s other model-building clients to honour their contractual commitments.

This, in turn, has left bond investors wondering if Oracle can honour its swelling commitments to them. Total debt to operating profit (before depreciation and amortisation) is an edgy 4.2, well above the safe range of zero to three. Throw in data-centre leases and Oracle’s total liabilities will nearly treble from just over $100bn in its latest financial year to $300bn three years from now, reckons Morgan Stanley, an investment bank. The yield on its bonds, which moves inversely to their price, has spiked in the past two months even as it has declined for the hyperscalers. It costs three times as much to insure against the risk of Oracle defaulting as it did in July, and three times more than for Microsoft.

Things could get dicier. In July Moody’s and Standard & Poor’s, two credit-rating agencies, put Oracle’s debt, which is currently two notches above junk status, on watch for a possible downgrade. Morgan Stanley recommends dumping Oracle’s bonds. Short-sellers are circling its stock. Mr Ellison’s son, David, may discover daddy’s ability to bankroll a $108bn hostile bid for Warner Bros Discovery to be somewhat diminished now that his pocketbook has thinned from a peak of nearly $400bn to $280bn.

Oracle’s predicament is more acute than the hyperscalers’. Its business is smaller, its pockets shallower and its lot hitched more tightly to the fate of a single customer, OpenAI, which accounts for over half its $500bn in pledged revenue. Its larger rivals look nothing like a utility.

But the hyperscalers’ bosses, starting with Meta’s newly debt-curious Mark Zuckerberg, ignore the lessons of Oracle’s transformation at their peril. Mr Ellison deserves a big “thank you” for giving them a complimentary crash course on what to do when you find yourself in charge of a capital-intensive, debt-hungry business—and what not to. ■