Although not yet publicly traded, the shadow of OpenAI — and its still-unprofitable business despite ChatGPT’s smash success — has shaken markets throughout the second half of 2025. Talk of an artificial intelligence bubble has not been silenced, despite Nvidia posting yet another explosive quarter in November.
The question remains: how will OpenAI balance ChatGPT’s seemingly endless appetite for computing power, fueled by data centers, with a business model that takes it from red to black? This was the same question that OpenAI CEO Sam Altman answered with a single exasperated word in a recent podcast appearance: “Enough.”
Also read:
Continues after advertising
Investment bank HSBC, although it clarifies that it still believes that AI is a “megacycle” and that its forecasts “indicate a leading position for OpenAI in terms of revenue”, nevertheless calculates that the company faces an extraordinary financial mountain if it is to fulfill its ambitions.
HSBC Global Investment Research projects that OpenAI will still not be profitable in 2030, even if its consumer base by then grows to encompass about 44% of the world’s adult population (up from 10% in 2025).
Furthermore, at least another US$207 billion (R$1.1 trillion) in computing will be needed to keep up with its growth plans. This harsh assessment reflects rising infrastructure costs, intensified competition, and an AI market that is growing in demand and use of capital to a degree beyond any technological trend in history.
HSBC’s semiconductor analysis team, led by Nicolas Cote-Colisson, arrived at the number by updating its forecasts for OpenAI for the first time since mid-October, considering recent multi-year commitments to cloud computing, including a $250 billion deal with Microsoft and a $38 billion deal with Amazon.
More importantly, HSBC notes, these deals came without any new capital injection, and are the latest in a series of capacity expansions that now have OpenAI targeting 36 gigawatts of AI computing power by the end of the decade.
Assuming that one gigawatt can power approximately 750,000 homes, electricity on that scale would represent the needs of a state slightly smaller than Texas and slightly larger than Florida.
Continues after advertising
The Financial Times’ Alphaville blog, which had previously reported on HSBC’s forecast, described OpenAI as “a bottomless money pit with a website on top.”
HSBC projects that OpenAI’s cumulative free cash flow through 2030 will still be negative, leaving a $207 billion funding gap that will need to be filled with more debt, equity or more aggressive forms of revenue generation.
Analysts model OpenAI’s cloud and AI infrastructure costs at $792 billion between the end of 2025 and 2030, with total computing commitments reaching $1.4 trillion by 2033 (HSBC notes that Altman has laid out a plan for $1.4 trillion in computing over the next eight years). With data center rentals alone, the company will have a bill of US$620 billion.
Continues after advertising
Despite this, projected revenues—while growing rapidly, surpassing $213 billion by 2030—simply would not be enough to close the gap. (The bank’s revenue projections are based on the assumption of a higher proportion of paid subscribers in the medium term and the assumption that language model providers, or LLMs, will capture share of the digital advertising market.)
The bank points to several options to raise the necessary money, including significantly increasing the proportion of paid subscribers (going from 10% to 20% could add US$194 billion in revenue); capture a larger share of digital advertising spend; or extract extraordinary efficiencies from computing operations. But even in optimistic conversion and monetization scenarios, the company would still need new capital after 2030.
OpenAI’s survival is closely tied to its backers and the AI ecosystem. Microsoft and Amazon are not only cloud providers but also major investors, and cloud players like Oracle, Nvidia, and Advanced Micro Devices (AMD) have a lot to gain — or lose — depending on the fate of OpenAI.
Continues after advertising
But the risks are considerable: unproven revenue models; possible saturation of the AI subscription market; threat of regulatory oversight; and the colossal scale of the capital injections required.
HSBC suggests that OpenAI could raise more loans to finance its computing needs, but this would be “potentially the most challenging path under current market conditions” as Oracle and Meta have recently raised a “significant amount” to finance AI-related capital expenditure, “raising market concerns about overall AI financing.”
The bank notes that this is an exception, as most so-called hyperscalers (cloud computing giants) have been financing themselves with free cash flow, as recently noted by JPMorgan’s Michael Cembalest.
Continues after advertising
HSBC also saw a “strong increase” in credit swaps (default insurance) of Oracle in recent days, about which Morgan Stanley’s Lisa Shalett had already expressed concern in a previous interview with Fortune.
HSBC, like many other banks analyzing the AI revolution, has again resorted to Nobel laureate Robert Solow’s famous quote that “you can see the computer age everywhere except in productivity statistics,” noting wryly that “weak productivity gains, driven by low total factor (labor and capital) productivity, are an unfortunate feature of today’s developed economies.”
Indeed, the bank notes that some are still not convinced of a significant return from even the 30-year-old internet revolution, citing Federal Reserve Chairman John Williams’ 2017 comment that “the productivity enabled by modern technologies like the internet has so far influenced only our leisure consumption — and it has not yet reached offices or factories.”
Bank of America head of U.S. quantitative and equity strategy Savita Subramanian told Fortune in August that she sees a “tidal shift” in productivity emerging in the 2020s economy, in ways that have no fundamental relationship to AI.
Through a combination of factors, including post-pandemic wage inflation, she said companies have been driven “to do more with fewer people,” replacing people with processes in a scalable and meaningful way.
One point she cautiously highlighted was a shift in focus away from asset-light models (businesses with fewer physical assets) para asset-heavy (more physical goods)as many of the most innovative technology companies have discovered a nearly insatiable thirst for a type of hardware that carries a lot of risk: data centers.
A few months later, Harvard economist Jason Furman did a rough calculation and found that without data centers, GDP growth would have been just 0.1% in the first half of 2025.
OpenAI appears to be asking the market a question: How long can growth be sustained by the expectation of future returns — and a productivity revolution — from AI that is by no means guaranteed to arrive?
2025 Fortune Media IP Limited
