OpenAI needs a lot of power for its operation, currently they have committed to rental of servers with a combined capacity of 36 Gigawatts, primarily from Microsoft and Amazon.
In a recent report from the bank HSBC took a closer look at the economics of it all, which- Financial Times has read, and come to some interesting conclusions.
OpenAI has committed to spending $1.8 trillion , resulting in a $620 billion bill annually, despite the analysis predicting that OpenAI will only be able to utilise 1/3 of the power at their disposal.
The problems arise despite HSBC are utilising an extreme best case scenario, including that 44% of the entire world population is using their product - although not including China, providing a user base of 3 billion people, almost 4x the number of users today - in five years, and having 10% subscriptions, twice the current number, and not least, profiting 2% on digital advertising, currently very close to 0%.
This very positive approach to income still does no come even close to break even, with the projection being an annual loss of almost $18 billion in 2030.
This leaves fundraising on a gigantic scale as the only viable way to stay afloat, and this is despite AI revenue estimated to hit $129 billion for consumer sales, and $386 billion from enterprises in 2030. But with huge operating costs, and xAI and Anthropic gaining marked shares, even when excluding Google completely from the calculations.
The main thing is that OpenAI's server rental costs will be $792 billion over the next five years, and then explode to $1.4 trillion from 2030-2033. This is matched by projections by OpenAI made public three weeks ago. In total, at least $207 billion are missing from the spread sheets, no matter how optimistic a frame work can be made.
(Logo credit: OpenAI)