The announcement of a $500 billion data center joint venture this week has set the industry abuzz.

The enormous investment, set to come over four years, could reshape the entire sector, and have a profound impact on supply chains and market dynamics.

Stargate
– Stargate/MGM

If it's real

If the pledge of $100 billion upfront and another $400 billion in the coming years is real, and if it is spread equally at $125 billion a year, that would be more than the combined global capex of Meta, Microsoft, and Google. Stargate is just US-focused and likely won't have a huge office budget.

At the same time, the hyperscalers are themselves ramping up spending. Microsoft has said that it will spend some $80 billion this year on AI data centers, up from a $53 billion general capex spend in 2024.

All of this will put enormous pressure on the industry to keep up, although there is a caveat - not all of this spending is truly distinct. Much of Stargate's initial spend is funneled through Oracle's Abilene, Texas, site, which itself is actually a Crusoe facility backed by Blue Owl.

Its new spending could include acquiring existing sites and projects that have already been factored into our understanding of data center spending.

Still, the sudden appearance of what is essentially a new hyperscaler of record proportions will disrupt the market in a way that will reach deep into the supply chain.

Whether that supply chain will actually be able to keep up is unclear. Chip makers have already struggled to provide enough GPUs, while getting a generator requires year-plus preorders.

Stargate may lean on Oracle for some of its supply chain contacts and nouse - although it will be somewhat of a competitor, and Oracle has never worked at this scale. Expect supply crunches and workforce challenges.

This puts the venture at an immediate disadvantage to existing hyperscalers, meaning the likelihood of two outcomes - first, Stargate will again lean heavily on the existing market, such as Crusoe and CoreWeave, to take advantage of their current order books.

And, second, there is a good possibility that the company decides to forego some of the redundancy and design decisions that traditional operators adhere to. Training runs, which Stargate will focus on, do not care about latency, nor do they necessarily require perfect uptime.

AI training save states mean that, should a data center have an outage, it can resume work with minimal loss beyond the lost time itself. This is unlike live services, where downtime can mean expensive losses for customers (and operators).

This has led many in the industry to question whether we need as many generators for training systems, or whether we can strip as much redundancy out as possible to save on time and cost.

Then there's the question of power. Stargate's announcement of focusing on 'AI infrastructure' is vague - it's not clear whether that could include power projects, but it will at the least probably include power costs of operation (which are not included in the hyperscale capex comparison above).

Grid constraints have already plagued the first innings of the AI boom, and are set to get harder in the years to come. Stargate may be able to outbid competitors, as well as buy out existing projects, but it is still hamstrung by the same broader power crunch.

Stargate could throw money at the problem, but traditional power projects are painfully slow, even with Trump's expected bonfire of environmental regulations. With Altman invested in both Helion and Oklo, the company should be expected to throw money at exotic and unproven nuclear solutions.

This may seem like a risky bet, and it is - but it is also a smokescreen. Expect much of Stargate's infrastructure to rely on fossil fuels, including gas and coal, with the promise of a greener future around the corner that may or may not come.

Exactly how much power Stargate expects to consume is unknown, but we can do some very, very rough back-of-the-napkin math. OpenAI claimed that they expect a $100 billion data center to consume 5GW, so that would come to 25GW overall.

Again, there's no guarantee that Stargate plans to build facilities of that scale, nor is it even clear they could find enough power. To put it in perspective, it is more capacity than twice the nation's entire colocation data center capacity of mid-2024.

The Department of Energy already believes that data center power use could jump from 4.4 percent of the nation's power in 2023 to as much as 12 percent by 2028, and that does not include the Stargate boom.

Should the company manage to secure close to this power, it will have two major knock-on effects. It will crowd out other operators and increase the prices of potential sites, and it will act as a focal point for wider data center awareness.

After operating for decades in the shadows, Stargate risks linking the sector to what may end up being controversially large, polluting, and job-taking projects. When people envision AI today, they often conjure up images of The Terminator or HAL 9000. In a few years, as sectors are replaced by AI, they may be picturing data centers.

Beyond power and supply chains, other technical challenges exist. No one has built a data center at the scale envisioned, and new problems are bound to arise. Already, cosmic radiation is impacting training runs in clusters of 100,000 GPUs. Should that move into the million-GPU range, OpenAI will stumble upon entirely new issues.

If and how it addresses this will also have a knock-on effect. Liquid cooling will be required for the 130kW+ racks that should be expected, but the sector is still nascent and yet to converge around standardized solutions.

Should Stargate put its weight behind a specific approach, or supplier, that will ripple through the broader industry. They would serve as kingmakers for the market, with some providers set to benefit enormously.

That is, of course, if it actually happens.

According to reports from The Information, OpenAI and SoftBank will both invest $19 billion into Stargate, while Oracle and Abu Dhabi's MGX will pump in a combined $7 billion. That comes to $45 billion - already a huge sum, but one that is not so industry-shaking.

Altman, Trump, and SoftBank's Masayoshi Son are all fantabulists of different flavors.

Altman is trapped in the hype cycle, where the disappointment of today's AI not reaching the valuation OpenAI commands can only be mitigated by promising ever grander technologies in the future - an approach that vaulted Elon Musk to the top of the world's rich list.

At OpenAI, Altman also faces the uncomfortable truth that the company's advantage against rivals like Anthropic and Google appears to be dwindling. Much of the core technology is replicable, and most of the data is open source (or copywritten, but used by them all) leaving only one clear moat - who can bring the most compute to bear. Stargate is an attempt to dramatically move ahead of rivals by sheer force of money.

Next is Masayoshi Son, once the world's richest man, who hopes to regain that crown. He made his billions by betting the farm on grand technological shifts, benefitting most from early bets in Yahoo and Alibaba. But it's important to note that he has often been wrong, nearly bankrupting SoftBank after going all in just before the dot-com crash, and wasting billions on companies like WeWork after spending no time on due diligence.

And then there's Trump, a peripheral character in this story, but one that obviously benefitted from announcing an enormous investment at the beginning of his term - whether it happens or not. Similar announcements marked his first term, including a $10 billion Foxconn project that amounted to naught.

All three cannot be trusted to stick to their investment pledge and share a 'fake it till you make it' mentality of pushing large numbers and lofty promises in the hope they become self-fulfilling.

Similarly, the market's willingness to provide the remaining $455 billion, either through debt or equity, is unknown. This represents an enormous and unprecedented sum and is far from guaranteed. It's also a project that Microsoft - which has deep pockets, a close relationship, and knows more than most about the future of data centers - decided not to be involved in.

Should billions flow in, pressure on Stargate and OpenAI is set to be immense. The company plans to invest nearly as much as it has raised for its own operations into the company, despite running at a loss and having yet to quite prove the business model for generative AI.

The company was already in a precarious position, and Stargate only ups the risks exponentially. Getting into the server game means taking on a fast depreciating asset class, all while draining resources and focus.

At the same time, OpenAI will have to keep ahead of its well-funded competitors, learn how to run a data center business, and move at a speed as yet unheard of.

All for a business model that could be a bubble.

The Mother of All Bubbles

Should it pop, again, there are several potential outcomes, somewhat dependent on how hard the market crashes and how far along Stargate is.

The terms of the agreement with Microsoft, which was OpenAI's exclusive cloud provider, put some limits on what Stargate can do. "Microsoft has approved OpenAI’s ability to build additional capacity, primarily for research and training of models," the company said in a statement.

Should Stargate and OpenAI get cash-strapped, there is a version of the future where they resell the capacity to other users, or lease it through Oracle - but only if Microsoft allows it. Alternatively, the company could fully sell off developments either for a profit or a fire sale.

The other alternative is the company overstretches just as the market contracts, risking half-built projects being left fallow with little appetite for acquisition. A broader industry rout could be expected, as hype dies and rationality returns.

Then comes the classics we see every cycle turn: Layoffs, bankruptcies, and mothballed projects. Data centers themselves will remain critical to civilization and business, but the frothy valuations and breakneck buildouts will be distant memories.

For now, though, such possibilities remain simple suppositions.

The most likely version of events is that Stargate becomes a major player, buying existing capacity alongside a few flagship projects (built with partners), but does not manage to reach the half-trillion dream.

Even if that happens, how it operates and who it works with, will help reshape the data center sector at a time where it is already remaking itself for the AI era.