At a certain point, numbers lose meaning.
With this week’s announcement that OpenAI plans to launch a $500 billion data center company, we wanted to put that number into perspective. Instead of a new AI infrastructure business called Stargate, what could OpenAI, SoftBank, Oracle, and MGX have done with the money? (That is, if they have the money.)
All the biggest data center companies
For a fraction of that price, you could have been in all top ten of the world’s largest data center deals - CyrusOne for $17.36 billion (2021, all figures inflation-adjusted), AirTrunk for $16.11 billion (2024), Switch at $11.79bn (2022), Coresite for $11.69bn (2021), QTS for $11.58bn (2021), Interxion at $10.31bn (2019), Global Switch for $9.82bn (2019), DuPont Fabros at $9.73bn (2017), Compass Datacenters for $5.66bn (2023), and Data4 at $4.07bn (2022).
Together that would amount to a mere $108.12 billion. Alternatively, on the current public market, you could buy Equinix for $90.49bn, along with Digital Realty ($60.95bn), and throw in DigitalBridge ($1.99bn) for good measure.
Now what you wouldn’t get is the actual hardware within those data centers, including any pricey GPUs. But you would get their customer contracts and stable revenue flow, the latter of which generative AI outfits such as OpenAI have yet to achieve.
The world’s fastest supercomputer, 833 times over
The 1.742 exaflops (HPL) El Capitan system at Lawrence Livermore National Laboratory currently tops the Top500 list, with some 11,039,616 CPU and GPU cores, all for an estimated $600 million.
That figure includes some of the research costs of building a new frontier system that would be spread out across the 833 supercomputers, but does not include the land El Capitan is set on, nor operational costs. So you may have to settle for a few hundred less.
All of Intel, AMD, and Arm
Nvidia may be out of reach, but why not dominate the rest of the chip game? With that money, you could pick up Intel ($95bn), AMD ($203bn), and Arm ($187bn), all for $485bn, saving you $15bn.
A decade of 2024 hyperscaler capex costs
Last year, Microsoft spent a record $53 billion on capex. This includes office spend, but primarily went to data centers and servers. This would cover nearly 10 years of such spend, although Microsoft now plans to spend $80bn on AI data centers in 2025, so you’d have to settle for just over six years of that buildout.
Again, such hyperscaler figures don’t include operating costs, which cost billions more.
Pay off Oracle and SoftBank’s debt
Two of Stargate’s backers could do with a cash infusion of their own. With this money, one could instead pay off the debts of Oracle ($86.9bn) and SoftBank (~$140bn), and have $273bn left over.
11 Twitters
At the $44bn price Elon Musk paid for it ($47.16bn, adjusted). Or you could buy 53 versions of X, based on its October valuation.
Five International Space Stations, for a decade
Each one costs around $100bn, including initial costs. Alternatively, the current one could be maintained for 167 years at current costs (although it is falling towards Earth).
China’s military, for a year or so
Estimates for how much China spends on its military are rough guesses at best, but researchers at the Stockholm International Peace Research Institute (SIPRI) put China’s military spending in 2023 at 2.096 trillion yuan, or about $296 billion.
When looked at through purchasing power parity terms with the US, China’s military expenditure is thought to be around $541 billion, according to non-profit CEPR.
Make the UK net zero
In a July 2021 report, the Office for Budget Responsibility estimated the net cost of the UK reaching net zero by 2050 to be $481.67bn (inflation adjusted, and currency converted).
One Elon Musk
The funds could just about best the current net worth of Elon Musk ($430bn), but that may change as his influence in the US continues to grow.
What would you do with the cash?