OpenAI, the company behind ChatGPT, is navigating a financial paradox that defines the current moment in generative artificial intelligence: its products are used by hundreds of millions of people, but every dollar it earns costs it more than double to obtain. In 2024, the company lost $5 billion against $3.7 billion in revenue, a spending ratio of $2.35 for every dollar earned, according to a CNBC report from September of that year.
The figures, also confirmed by The New York Times and The Information, reveal a cost structure that grows faster than revenue. Between model training, inference, and research compute, OpenAI spent approximately $5.8 billion on computing capacity in 2024 alone (Epoch AI), to which up to $1.5 billion in personnel costs were added. The result is a financial hole that, according to internal documents leaked to The Information, could reach $14 billion in projected losses by 2026 and accumulate $44 billion in losses between 2023 and 2028 before becoming profitable.
Faced with this picture, OpenAI has deployed a multi-pronged strategy to close the gap: silent degradation of the free experience, a tiered subscription architecture ranging from $8 to $200 per month, and the introduction of advertising within ChatGPT for the first time in its history.
The silent free tier
The free ChatGPT experience is no longer what it used to be. Since 2024, OpenAI has implemented a mechanism it officially calls “capacity-based fallback”: free users can send approximately 10 messages every 5 hours using the main model —currently GPT-5.3 Instant. Upon reaching that limit, the system automatically switches to a mini version of the model, significantly less capable, without notifying the user.
The change is subtle but tangible. Those who have used the free version continuously report less detailed responses and inferior reasoning quality once the threshold is crossed. OpenAI shows no warning: the only way to detect the degradation is to manually check the model indicator in the interface.
This system represents a radical shift from ChatGPT’s early days, when free access was practically unlimited. Gradually, as inference costs soared with larger models, OpenAI has been tightening the limits downward. Today, a free user who wants to maintain fluid, high-quality conversations for more than an hour will find themselves, without knowing it, talking to a significantly more limited model.
The subscription pyramid
Where there were once two plans —Free and Plus ($20/month)— OpenAI has built a five-rung ladder in just over a year. At the base, the Free plan, severely limited. Next comes ChatGPT Go, launched in January 2026 for $8 a month, an ad-supported plan that the company describes as its equivalent to the Spotify model: affordable access in exchange for ads integrated into responses.
The classic ChatGPT Plus remains at $20 a month as the entry-level ad-free plan. Above it, ChatGPT Pro —launched in December 2024 at $200/month— was split in April 2026 into two tiers: Pro at $100/month (limited access to frontier models) and Pro at $200/month (full access, including Sora video generation and the most advanced reasoning models).
According to The Information, approximately 76% of OpenAI’s revenue comes from paid subscribers, who represent roughly 20% of total users. The company reported 50 million paid subscribers in April 2026, a significant jump from approximately 10 million in December 2024. However, the Pro tier —the most profitable per user— represents less than 1% of that base, with independent estimates placing the figure below 500,000 subscribers.
Advertising: the great fallen taboo
The most significant —and controversial— move by OpenAI in 2026 has been the introduction of advertising in ChatGPT. For years, the company avoided this path, partly out of consistency with its image as a nonprofit research lab turned company. But the numbers left no room for maneuver.
In January 2026, OpenAI formally announced its plans to test ads on the Free and Go tiers of ChatGPT. The rollout began in February in the United States: free and Go plan users see ads integrated at the end of model responses. By April 2026, ads were already generating approximately $100 million in annualized revenue, according to Lapis estimates.
The company has described its advertising strategy as inspired by the Spotify model —ads on the free tier, a low-cost plan with reduced advertising, and ad-free premium subscriptions— but the analogy has limits. Unlike a music service, where ads are intrusive but predictable, advertising in a conversational assistant raises new questions: can a language model maintain its objectivity if it knows that certain responses may generate more advertising revenue? How can it be audited that an ad does not influence the system’s recommendations?
OpenAI has not publicly answered these questions. The company states that ads are “clearly separated” from the model’s responses, but has not published technical details on how that separation is implemented.
Between a record round and an IPO that hasn’t come
OpenAI’s financial pressure has not stopped investors from continuing to bet on the company. In March 2026, OpenAI closed a $122 billion funding round at an $852 billion valuation, the largest in venture capital history. SoftBank led the deal, which adds to previous commitments from the Stargate consortium —a joint venture with SoftBank and Oracle that plans to invest up to $500 billion in AI infrastructure over the coming years.
However, financial opacity persists. In April 2026, The Wall Street Journal reported that OpenAI had missed its internal revenue and user targets in the first quarter of the year, just as the company accelerates its preparations for an initial public offering (IPO) that some analysts consider the most anticipated in the tech sector since Meta’s market debut.
Documents leaked to The Information project that OpenAI will not be profitable until 2029. Meanwhile, its cash burn rate —$9 billion in 2025, projected at $17 billion in 2026 according to The Economist— tests investor patience and fuels skepticism about whether the current generative AI business model is sustainable in the long term.
What this means
OpenAI’s situation is not unique. Anthropic, Google DeepMind, and other frontier companies face similar dynamics of rising compute costs and pressure to monetize. But OpenAI, as the market leader and creator of ChatGPT, is the most visible case and the one setting the trend.
The shift toward advertising, the progressive degradation of free access, and the stratification into multiple paid tiers paint a future in which access to frontier artificial intelligence will increasingly depend on one’s ability to pay. The open question is whether these mechanisms will be enough to close the financial gap —or whether, as internal projections suggest, OpenAI will need several more years and likely new revenue streams before reaching profitability.
Primary sources: CNBC — OpenAI sees $5 billion loss | The New York Times — OpenAI Is Growing Fast and Burning Through Piles of Money | The Information — Why OpenAI Could Lose $5 Billion This Year | The Information — OpenAI Projections Imply Losses Tripling to $14 Billion in 2026 | OpenAI — Our approach to advertising | OpenAI — Introducing ChatGPT Go | OpenAI — Testing ads in ChatGPT | OpenAI — A business that scales with the value of intelligence | WSJ — OpenAI Misses Key Revenue, User Targets | The Economist — OpenAI faces a make-or-break year in 2026 | Epoch AI — Most of OpenAI’s 2024 compute went to experiments | CNBC — OpenAI challenges Anthropic with ChatGPT Pro subscription tier