The Hard Reality of OpenAI’s Business Model
By Siddharth Jindal ·
In 2015, OpenAI was founded as a research lab with a mission to ensure AI benefits all of humanity. Today, it’s one of the most talked-about for-profit business stories in tech, generating billions in revenue due to rapid enterprise adoption.
The AI giant’s annualised revenue surged past $20 billion in 2025, up from $2 billion in 2023, as the company rapidly expanded its compute capacity, according to OpenAI’s chief financial officer, Sarah Friar.
While ChatGPT is its most visible revenue stream, with the company reportedly boasting a paying subscriber base of 35 million, OpenAI CEO Sam Altman recently busted the myth that the company generates revenue primarily through ChatGPT subscriptions.“We have added more than $1B of ARR in the last month just from our API business. People think of us mostly as ChatGPT, but the API team is doing amazing work!” Altman wrote on X.
OpenAI’s API platform is available on GPT-5, mini, and nano models, supporting text, image, audio, and visual inputs.
Analysts note that OpenAI’s revenue mix is more nuanced. Arun Chandrasekaran, a ‘distinguished vice president’ and analyst at Gartner, tells AIM that subscriptions and enterprise contracts such as ChatGPT Plus, Pro, Business, and Enterprise provide a stable revenue base, while API usage and spending by large customers tend to fluctuate month to month.
“Model API consumption also has the lowest revenue predictability, as the barrier to exit remains relatively low with powerful, comparable models continuing to emerge,” he says.
This has pushed the company to explore revenue streams beyond subscriptions and APIs. OpenAI is now planning to introduce advertising in ChatGPT—a model that has long underpinned Google’s ad business.
Looking ahead to 2026, Friar said OpenAI’s financial focus will be on practical adoption, particularly in health, science, and enterprise use cases where improved intelligence can directly translate into measurable outcomes. She also signalled future revenue models, including licensing, IP-based agreements, and outcome-based pricing, as AI expands into areas including drug discovery, energy systems, and financial modelling.
Meanwhile, OpenAI’s arch-rival, Anthropic, has raised its revenue forecasts for the coming years by about 20%. According to a report by The Information, the company now projects sales to quadruple to as much as $18 billion in 2026, before climbing to around $55 billion in 2027.
OpenAI’s Enterprise Push Takes Shape
OpenAI is also exploring shifting the way it charges enterprise customers.
Under this approach, OpenAI would seek a share of the economic value generated when customers use its tools to produce outcomes such as new drugs or materials. This would move the company beyond subscription and API-based pricing towards revenue models tied directly to results.
Chandrasekaran says platform and licensing partnerships could make this shift more sustainable. By embedding AI into vertical-specific platforms such as legal services or pharmaceutical research, OpenAI can align its revenue more closely with customer success. Over time, he adds, partnerships with hardware or device providers could further reduce reliance on token-based billing.
As part of this push, OpenAI is stepping up its focus on enterprise and government customers. The company has reorganised its leadership to strengthen its enterprise strategy, appointing Barret Zoph to lead enterprise AI sales. Today, OpenAI serves more than one million business customers globally.
In consulting, OpenAI is also placing its engineers and researchers within customer teams to customise AI systems for specific business needs, following a model used by companies such as Palantir, Accenture, and ServiceNow.
Jacob Joseph, VP – Data Science at CleverTap, argues that the long-term opportunity lies in core enterprise infrastructure. “When an AI is baked into a company’s CRM or customer support stack, it stops being a ‘cool tool’ and becomes operational infrastructure, something you don’t cancel when budgets get tight,” he says.
Notably, ServiceNow recently entered into a multi-year partnership with OpenAI to embed its AI models, including GPT-5.2, directly into the ServiceNow platform. This integration is intended to automate complex business tasks and improve workflow efficiency.
What About ChatGPT Ads?
Joseph says OpenAI is not eyeing the ad market because it has realised that the “subscription-only” model has a ceiling.
“ChatGPT is currently handling roughly two billion prompts a day. That is a staggering amount of compute to burn through while hoping users continue paying for a $20 Plus subscription. Advertising becomes an obvious escape hatch to monetise the masses without putting up a ‘pay to play’ sign,” he says.
On the other hand, Google Search, he explains, is built on explicit intent, where users type a query and receive a list of links. Conversational AI, however, is shifting that behaviour. Around 55% of users now treat chatbots as the starting point for their curiosity, bypassing the traditional search bar altogether.
That behavioural change has direct implications for monetisation. On advertising in ChatGPT, OpenAI CFO Sarah Friar says that ads, if executed correctly, can align with OpenAI’s mission without eroding trust.
Analysts see real potential here. Evercore ISI’s Mark Mahaney suggests in a note to investors that OpenAI could generate over $25 billion in annual ad revenue by 2030, rivalling some of the largest digital ad platforms.
OpenAI is reportedly targeting around $60 per 1,000 impressions for early ads placed below ChatGPT responses. That rate is roughly three times the typical Meta pricing, positioning ChatGPT as a premium channel for high-intent users.
For comparison, Google generates the bulk of its revenue from advertising, with ad sales reaching about $264.6 billion in 2024, according to Statista. This accounts for more than 75% of parent company Alphabet’s total revenue.
Sanchit Vir Gogia, CEO of Greyhound Research, tells AIM that OpenAI is walking a tightrope. It is introducing ads to offset massive compute costs and to signal future monetisation capability ahead of a potential IPO. He adds that OpenAI is essentially trying to replicate what Google achieved with AdWords in the early 2000s, but within a conversational interface with no natural break between content and commerce.
“If even a small fraction of users start to suspect that ChatGPT’s answers are influenced by advertising, the platform’s utility for decision-making collapses,” he adds.
Cash Burn And Infrastructure
OpenAI expects to burn more than $100 billion through 2029, with losses continuing to outweigh profits in the near term. The company has committed to massive compute deals totalling over $1 trillion through 2035, diversifying beyond Microsoft to fuel its AI infrastructure needs with cloud services and GPUs.
Confirmed agreements include a $250-billion incremental contract with Microsoft Azure for expanded cloud capacity, a $300-billion five-year pact with Oracle tied to the Stargate supercomputer project, and a $38-billion seven-year deal with Amazon AWS for NVIDIA GPUs and CPUs that is poised to scale up by late 2026.
“Training a frontier model is a massive, episodic flex of capital. But inference (actually running the thing) is the daily utility bill that never stops growing. Every time you ask a model to ‘think’ longer or generate a video, a server somewhere starts sweating,” says Joseph.
Echoing a similar sentiment, Gogia says the dominant driver of OpenAI’s cash burn is not training new models, but serving existing ones at scale.
“OpenAI’s infrastructure requirements are enormous. Running billions of queries per week means sustained demand for GPU compute, especially when many users expect premium experiences with GPT-4 or multimodal capabilities. The company has scaled up to roughly 1.9 GW in compute capacity, and most of that is tied to inference rather than training.”
Challenges Ahead
Gogia says the competitive pressure facing OpenAI is twofold. On one side, hyperscalers are embedding LLMs into broader cloud and productivity bundles, often subsidising access through volume pricing or including AI features as part of larger enterprise contracts. This, he notes, makes OpenAI’s standalone pricing harder to defend, particularly when procurement teams evaluate costs in isolation rather than against the quality of outcomes delivered.
On the other side, open-source models such as Llama, Mistral, and others are now good enough for many non-critical enterprise tasks. For organisations with strong internal engineering capabilities, these models offer lower costs and greater deployment flexibility.
OpenAI’s challenge is compounded by the fact that many enterprises are now designing for a multi-model future. Buyers are benchmarking outputs across providers, routing simpler workloads to cheaper models, and reserving OpenAI’s systems for higher-complexity use cases. This effectively places a ceiling on pricing, as even marginal increases invite closer scrutiny and potential shifts in volume.
Joseph points to competition from Big Tech as another structural hurdle. “Unlike a startup that needs to justify every dollar spent, Google has the luxury of being a massive, cash-flow-positive machine,” he says. “They can subsidise AI aggressively, offer competitive pricing or even free tiers, and use old-world ad profits to fund new-world AI dominance.”
OpenAI’s response, he adds, has been to split the stack. The company is competing on price with smaller, cheaper models and advertising, while simultaneously positioning its most advanced systems as premium offerings. “They’re effectively charging for luxury intelligence,” Joseph observes.