In late April 2026, an investor in both OpenAI and Anthropic was asked the question hanging over the industry: how can AI companies losing billions ever turn a profit? His answer was straightforward. They will raise prices. Some customers will leave, the ones who stay will more than make up the difference.
The plan is already in motion, and the numbers behind it tell two very different stories.
In March 2026, OpenAI announced it was making about $2 billion a month (roughly $25 billion a year), but internal documents leaked at the same time showed the company expects to lose $14 billion in 2026 alone, and to keep losing money every year until 2030. Anthropic is on a different track, by April 2026 it was on course to make $30 billion a year, up from $9 billion only four months earlier, and expects to be making more than it spends by 2027. For context, it took Salesforce about 20 years to reach that $30 billion figure; Anthropic did it in under three years from a standing start.
Another way to look at this: Counterpoint Research found Anthropic earns an average of $16.20 a month from each person who uses Claude. OpenAI earns $2.20. Google earns $1.10.
Cheap, cheerful and IPO ready
The revenue gap is not the biggest gap though. OpenAI closed a $122 billion funding round in March 2026 at an $852 billion valuation — but a Wall Street Journal report in April revealed it had missed its targets for revenue and user growth, including its internal goal of one billion weekly ChatGPT users. Anthropic closed its Series G at $380 billion in February; by May, secondary markets were implying valuations above $1 trillion ahead of a possible October IPO. Both are racing toward the largest pure-play AI listings in history. What justifies those valuations is not next year's profit. It is the adoption story being told now - and every subscription, every API call, every enterprise seat helps tell it.
The losses are part of a numbers game. Make the technology cheap, get people using it at scale, and four things follow - whether or not they were the original intent.
First, get in front of as many people as possible, as fast as possible. When it launched, ChatGPT reached 100 million weekly users in two months — for context, TikTok took nine months, and Instagram 30 months. By early 2026, ChatGPT had crossed 900 million weekly users. This is not just signing up customers. It is building a habit at a scale no one else can catch up to. Once people instinctively reach for ChatGPT the way they reach for Google, no new entrant can compete, no matter how good their technology is.
Second, collect data. Every conversation a person has with an AI teaches the AI something — what people ask, what they reject, what they correct, what they come back and refine. The cheaper the access, the more conversations happen, and the more the AI learns. OpenAI has already announced that its adverts will be matched to the topic of users' current conversations, their past chats, and how they have interacted with previous ads. That same setup also captures something deeper: how people think, where they hesitate, how they make up their minds. AI is the first consumer technology in history where using it for free is the job. Free users are not a cost to the company, they become a workforce doing unpaid training.
Third, get AI woven into company workflows. This is happening fast and at scale. Anthropic has more than 1,000 customers paying over $1 million a year for Claude - and that number doubled from 500 in less than two months. A JetBrains survey in January 2026 found 74% of professional developers worldwide use AI coding tools, with most teams running three or four at once. Once a company has rebuilt how it does its work around an AI provider, switching is not a licence-fee question. It means retraining staff, rewiring how data flows, and rebuilding internal rules. Cloud computing took 15 years to build that kind of lock-in. AI is doing it in 18 months.
Fourth, set the price later, when companies cannot say no. McKinsey put it plainly in its April 2026 analysis: companies that move first into AI "lock in lower cost positions" and make it "harder for competitors to catch up." The strategy is straightforward: give it away cheaply, become essential, then charge what you need to charge. This is the same playbook that every large platform of the last 20 years has run. What is different with AI is how fast it is happening.
The dependency bill
Writer's 2026 Enterprise AI Adoption Survey found that 97% of organisations had deployed AI agents in the past year - and only 29% reported significant ROI. Fifty-four per cent of senior executives went further: they admitted that AI adoption was "tearing their company apart."
That is the visible cost. The invisible one is what is happening to the people inside those companies. We wrote about this at length in Borrowed Thinking — the way frequent AI use produces what researchers call "cognitive debt," a measurable decline in the brain's willingness to do the work itself. The neuroscience is there. What the Writer data now shows is that the same dynamic is playing out at the level of the organisation. Companies have not just installed AI. They have installed an insidious, accumulating dependency on it.
A company can switch AI providers, but it cannot switch back to a workforce that has stopped thinking for itself.
The consulting firm Teneo runs a big annual survey of CEOs and the investors who back them. Their Vision 2026 edition found that 53% of investors expect AI to start paying back within six months. Only 16% of the CEOs running the largest companies think that is possible, most CEOs expect it to take well over a year, and most are still increasing their AI spending. A separate Deloitte survey of nearly 2,000 senior executives across Europe and the Middle East found that only 6% had seen any payback from AI within a year.
We are seeing four ways this gap could close based on current trends:
Prices will go up. Anthropic has already moved away from flat subscriptions toward tokens use, and OpenAI is heading the same way.
Adverts will spread. They have already arrived in ChatGPT's cheapest tiers, and OpenAI's internal forecasts expect advertising revenue of $2.5 billion in 2026 rising to $100 billion by 2030.
More layoffs. More companies will sack people and call it AI productivity, even where the AI has not actually delivered any. Per Writer's survey, 69% of companies are doing layoffs due to AI, but 39% don't have a formal strategy in place to drive revenue from these tools.
Providers preference. For the companies most heavily committed, the option of "let us just switch to a different provider" will turn out to be no option at all, because their work, their data, and the skills of their staff have all been rebuilt around the one they have.
The real price we pay
What started as the dreams of a handful of Silicon Valley founders — can artificial intelligence be created and become smarter than human intelligence? — has become the main focus of CEOs across the globe: scrambling to lead the AI transition and fearing losing their own job if they fail. Within four years of ChatGPT being released, the results of that work are the daily companion of children doing their homework, professionals writing contracts, and CEOs making strategic decisions. The technology arrived faster than any consumer technology in modern history. But the conversation about what it is doing to the people using it has barely begun.
The impact of AI is more important than social media's. Social media made money from our attention - what we looked at, what we clicked on. AI is being built to make money from our thinking - what we ask, what we worry about, what we write down and then change our mind about. The dependency it creates is not on a website we can close. It is on the habit of letting a machine do part of our thinking for us. The children growing up with it are growing inside a business model the companies running it have not yet finalised.
The AI in 2026 reaches into hundreds of millions of homes in a way no competitor can build. That position gives its providers the right to charge whatever the market will bear, once the customer has no real choice but to pay.
But the price is not the same for everyone, especially for enterprises. The Writer survey points to a minority - roughly one in three companies - who are getting AI right. They have not done it by going faster or spending more. They have done it by building contracts that assume the price will move, by treating AI as something that augments their people rather than replaces them, and by protecting the skills, the judgement, and the institutional knowledge that no model can be trusted to hold on its own.
They are paying the first bill - the subscription, the API tokens, the rising costs that come with every platform of the last 20 years. But they will not pay the second one.
The second bill is the one the rest of the market is underwriting: a workforce that has lost the habit of thinking without help, and the long, expensive work of getting it back.
The first price is unavoidable. The second one is still a choice.