Open sesame

Oracle spent $6.3bn on capital equipment in the six months to 30 November. It had $10.9bn in cash and $88.6bn of borrowings on its balance sheet.^{4Oracle (SEC filing)Oracle Corporation Form 10-Q for the quarter ended 30 November 202410 Dec 2024} On 21 January its chairman, Larry Ellison, stood in the White House beside Masayoshi Son of SoftBank and Sam Altman of OpenAI to announce a venture that will start with $100bn, about 16 times that half-year’s capital spending, and is meant to reach $500bn within four years.^{1The RegisterOpenAI wants to blow through $500B on AI infrastructure for itself, with help from pals22 Jan 2025,2Al JazeeraTrump announces $500bn 'Stargate' venture to build up AI infrastructure22 Jan 2025} The money is to buy data centres in America. It is the sort of figure that makes a balance sheet look like a rounding error, and the arithmetic is worth doing before the applause.

The structure, as announced, is straightforward. A new company, Stargate, will be funded by SoftBank, OpenAI, Oracle and MGX, an investment firm. SoftBank has financial responsibility, OpenAI has operational responsibility and Mr Son is chairman.^{1The RegisterOpenAI wants to blow through $500B on AI infrastructure for itself, with help from pals22 Jan 2025,3TV TechnologyTrump Unveils $500 Billion AI Project22 Jan 2025} Arm, Microsoft and Nvidia are named alongside Oracle and OpenAI as technology partners.^{1The RegisterOpenAI wants to blow through $500B on AI infrastructure for itself, with help from pals22 Jan 2025,2Al JazeeraTrump announces $500bn 'Stargate' venture to build up AI infrastructure22 Jan 2025} It is, the president said, the largest AI infrastructure project in history “by far”.^{2Al JazeeraTrump announces $500bn 'Stargate' venture to build up AI infrastructure22 Jan 2025} Mr Son called it the “beginning of our golden age”.^{2Al JazeeraTrump announces $500bn 'Stargate' venture to build up AI infrastructure22 Jan 2025} Mr Ellison offered something rarer in a launch, a fact: “The data centers are actually under construction”, the first of them in Abilene, Texas.^{1The RegisterOpenAI wants to blow through $500B on AI infrastructure for itself, with help from pals22 Jan 2025}

Big number, small base. $bn

	$bn
Oracle capex, six months	6.3
Oracle revenue, six months	27.4
Oracle borrowings	88.6
Stargate, first tranche	100
Stargate, four-year plan	500

Source: Oracle Form 10-Q for the quarter ended 30 November 2024; Stargate announcement as reported by The Register¹⁴

We think the right response to Stargate is neither a cheer nor a sneer. It is a ledger. The plan deserves attention because the demand for computing power that it presumes is real enough for the biggest firms in the business to commit to it. It deserves scepticism because a commitment, a pledge and a cheque are three different things, and so far the announcement supplies only the first.

Consider the doubters. Elon Musk, who has a quarrel with Mr Altman’s firm, said on X that “They don’t actually have the money”.^{3TV TechnologyTrump Unveils $500 Billion AI Project22 Jan 2025} Mr Altman replied that he was wrong and invited him to see the first site.^{3TV TechnologyTrump Unveils $500 Billion AI Project22 Jan 2025} Mr Musk’s motives are not disinterested, which is not the same as saying he is mistaken. The headline figure is a ceiling over four years, not a sum on deposit. The first $100bn is the only tranche with a timetable, and the announcement as reported does not say who puts up which part of it. Oracle already carries $88.6bn of borrowings,^{4Oracle (SEC filing)Oracle Corporation Form 10-Q for the quarter ended 30 November 202410 Dec 2024} and its capital spending is a small fraction of the sums being discussed. Whether it can pay its share is a fair question. It has not been answered.

The best case for Stargate is the one its backers make. America wants to lead in AI, the models are hungry for computing power and the existing supply is spoken for. A company that commits to building at scale before it has customers is doing what railway and telephone firms did in earlier centuries, and some of them went bust while the track remained. A consortium also solves a coordination problem, because chips, buildings, power and money must arrive in roughly the same year. We accept all of that. We do not accept that it needs a presidential podium.

Here the politics matters more than the maths. The venture is, by every account, private money.^{3TV TechnologyTrump Unveils $500 Billion AI Project22 Jan 2025} That is how it should stay. Washington has real work to do for anyone who wants to build data centres: faster permits for transmission lines and generation, clearer rules for connecting to the grid, a predictable approach to the export of chips. Those are public goods, and they should be offered to every builder, not to the one that was photographed. A government that blesses a single consortium invites the next firm to ask for a similar blessing, and then a subsidy to go with it. Industrial policy tends to begin with a ribbon-cutting and end with a bill.

What should readers look for? The first answer is plain: concrete. Abilene is under construction, so it should be possible to count the buildings, the megawatts and the chips by the end of this year. The second is the funding. If SoftBank, which carries the financial responsibility,^{1The RegisterOpenAI wants to blow through $500B on AI infrastructure for itself, with help from pals22 Jan 2025} shows how it will pay its share without piling on debt, the doubters will have a harder time. If it does not, the number will quietly shrink, as such numbers do.

Mr Altman and his partners have asked the country to believe in a very large gate. They are entitled to ask. Sensible people will wait to see it open.