HSBC’s US software and services team updated its OpenAI model to include the outfit’s enormous cloud compute rentals, which now include a $250bn deal with Microsoft and a $38bn arrangement with Amazon. These late October and early November signings add another 4 gigawatts of compute capacity, bringing OpenAI’s contracted total to 36 gigawatts.
The bank says the various cloud agreements together could reach a cumulative value of $1.8tn and translate to a yearly data centre rental bill of around $620bn, even though only a third of that power will be switched on by the end of this decade.
To test whether OpenAI can pay the tab, HSBC first had to build a revenue model. It plugged user numbers into an S curve that hits three billion by 2030 which it says equals 44 per cent of the world’s adult population outside China. That’s a sizeable leap from an estimated 800 million users last month.
The analysts think advertising and agentic AI might add revenue later, but for now, OpenAI is mainly persuading users to subscribe. HSBC claims subscription services will become as common as Microsoft 365 and forecasts that by 2030, around ten per cent of OpenAI’s users will pay, compared with roughly five per cent today.
The bank also assumes LLM companies will grab two per cent of the digital advertising market, which is slightly more than nothing at present. The result is wild revenue growth paired with equally wild cost inflation, which leaves OpenAI subsidising its users well into the next decade.
HSBC then lists the assumptions behind this modelling. It expects consumer AI revenue to reach $129bn by 2030, with $87bn coming from search and $24bn from advertising. It assumes OpenAI’s consumer share falls from 71 per cent this year to 56 per cent by 2030, while Anthropic and xAI sit in single digits, and a hazy “others” bucket gets 22 per cent. Google is conveniently left out.
Enterprise AI revenue is pitched at $386bn by 2030, with OpenAI’s market share slumping from about half today to 37 per cent by then. Everyone else is assumed to muddle along at roughly the same levels.
Even with these bullish forecasts, HSBC says it is nowhere near enough. It calculates that OpenAI’s rental costs will total $792bn between now and 2030, rising to $1.4tn by 2033. This more or less lines up with OpenAI’s eight year guidance that chief executive Sam Altman is apparently tired of being asked about.
The bank predicts cumulative free cash flow of around $282bn by 2030. It adds Nvidia’s promised cash injections and OpenAI’s disposal of AMD shares, which together could deliver another $26bn along with $24bn in undrawn debt and equity facilities and $17.5bn of liquidity at the midpoint of 2025.
Once all this is stacked against expected costs, HSBC finds a $207bn funding hole and adds a $10bn buffer to keep things looking steady.
These estimates may still be on the cautious side, although guessing how is anyone’s game. HSBC says every additional 500 million users adds around $36bn to cumulative revenue by 2030, while boosting paid conversion to 20 per cent could add $194bn. It flexes assumptions about LLM spending and compute pricing in similar ways and does not factor in the possibility that OpenAI might stumble upon artificial general intelligence.
If revenues fall short and investors get twitchy OpenAI might have to make grim choices. Debt markets have wobbled after Oracle’s recent antics, Microsoft’s backing has flickered, and SoftBank is the next largest shareholder.
The least awful option might be to lean on friends and wriggle out of some data centre commitments, either before or when the usual four- to five-year periods expire. HSBC says that, given the tangled ties among AI labs, clouds and chip designers, the bigger players might show some flexibility, as less capacity is preferable to a liquidity crunch.
HSBC’s software team is extremely bullish about AI’s long-term value and insists the technology will reach every production process and vertical. They argue that even a few small productivity boosts across a global economy worth more than $110tn could dwarf today’s frightening capital spending.
When it is framed like that, HSBC seems to suggest a $207bn shortfall might be a small price to keep the whole show running.
