OpenAI just launched a $100/month premium tier. Higher rate limits, faster responses, priority access to their most advanced models. They are also hitting $25 billion in annual recurring revenue and taking early steps toward an IPO.
On the surface, this looks like a company firing on all cylinders. Revenue growing, product expanding, IPO on the horizon.
But here is the number that should be getting more attention: Anthropic just passed OpenAI in revenue at $30 billion ARR. While spending roughly 4x less on model training.
Read that again. Anthropic is making more money while spending dramatically less to build the product. That is not a rounding error. That is a fundamentally different business.
The $100 tier tells you where OpenAI’s strategy is headed
The new premium tier is interesting not because of what it offers, but because of what it signals about OpenAI’s revenue architecture.
They now have three consumer price points: free, $20/month Plus, and $100/month Premium. The strategy is classic price segmentation. The free tier drives volume and brand awareness. The $20 tier converts casual users into paying customers. The $100 tier extracts maximum value from power users who will pay for speed and priority.
This is a proven playbook. Spotify, Notion, Figma, they all do variations of it. Nothing wrong with the approach.
The problem is the cost structure underneath. OpenAI’s inference costs are enormous. GPT-5.4 is expensive to run. Every query from a $100/month user costs real money, and power users query a lot. The margin on these subscriptions depends entirely on how much compute each user consumes, and power users consume the most compute while also being the ones who pay the most.
The bet is that the $100 tier generates enough margin to subsidize the free tier and fund continued model development. It is a reasonable bet if your models are efficient. It is a dangerous bet if your models are expensive to run. And OpenAI’s models, by all available evidence, are among the most expensive in the industry to serve.
The Anthropic gap is the real story
The SaaStr analysis cuts to the core of what is happening in the AI industry right now.
Anthropic: $30 billion ARR. Training budget estimated at roughly a quarter of what OpenAI spends.
OpenAI: $25 billion ARR. Projected losses of $40 billion by 2027. Valued at $852 billion after a $122 billion round.
The revenue gap alone is notable. Anthropic overtaking OpenAI in revenue was not supposed to happen this soon, if ever. But the efficiency gap is the part that should concern OpenAI investors.
Spending 4x less on training while generating more revenue means one of two things. Either Anthropic has found architectural efficiencies that OpenAI has not, or OpenAI is overspending on training runs that are not translating into proportional revenue. Probably some of both.
From an investor’s perspective, this matters enormously. OpenAI’s $852 billion valuation is built on the assumption that they will be the dominant AI company. Dominance requires either the best product or the best economics, ideally both. Right now, Anthropic is arguably winning on economics and competing effectively on product quality.
The IPO timing question
OpenAI is taking early IPO steps. This makes sense for a company that has raised $122 billion and needs to give early investors a liquidity event. You cannot stay private forever at that scale.
But the timing creates an interesting dynamic. An IPO requires financial disclosures. Detailed financial disclosures. The kind that show exactly how much money you are losing on each product line, what your unit economics look like, and where the margin is actually coming from.
Right now, OpenAI controls the narrative. They release revenue numbers selectively and frame them favorably. An IPO changes that. Public markets demand transparency, and transparency will reveal whether the business model actually works at scale or whether the revenue growth is being purchased with unsustainable spending.
My prediction: the IPO will happen, and the initial reception will be enthusiastic because the revenue numbers are genuinely impressive. But within two or three quarters, the cost disclosures will raise hard questions about the path to profitability. The same pattern that played out with Uber, WeWork (before it imploded), and dozens of other high-growth, high-burn companies.
Revenue is not profit. And in AI, the gap between revenue and profit is wider than in almost any other industry because the cost of compute scales with usage.
What the $100 tier really means
Step back and think about what it means when a company launches a premium tier at 5x the price of its standard tier.
It means the standard tier is not generating enough margin. You do not create a $100 product when your $20 product is printing money. You create it because you need the revenue density. You need a subset of users paying enough to make the economics work.
This is not a criticism. It is just economics. The question is whether enough users will pay $100/month for faster ChatGPT to meaningfully change the financial picture. And the answer probably depends on what “faster” and “priority access” actually mean in practice.
If the $100 tier offers noticeably better performance, if the responses are genuinely faster and the rate limits are generous enough that power users never hit them, then it could work. There is a real market for professional-grade AI tools at that price point. Developers, researchers, and content professionals already spend more than that on other software tools.
If the difference is marginal, if it is mostly a badge and a slightly higher rate limit, then it becomes a hard sell. People are generally willing to pay a premium for meaningful improvement, not for incremental bumps.
The bigger picture
The AI industry just entered a new phase. For the past three years, it was a one-horse race. OpenAI was synonymous with AI in the public mind. ChatGPT was the product. Everyone else was playing catch-up.
That era is over. Anthropic has more revenue. Google has distribution advantages that neither can match. Meta is building competitive open-source models. The market has gone from monopoly to oligopoly, and in an oligopoly, efficiency wins over scale.
OpenAI still has enormous advantages: brand recognition, developer ecosystem, enterprise relationships. But brand recognition does not fix unit economics. And enterprise relationships do not survive if a competitor offers comparable quality at better margins, because eventually those savings get passed on as lower prices.
The $100/month tier, the $25B ARR milestone, the IPO exploration: these are signs of a company that is growing fast and spending faster. The question for 2026 is not whether OpenAI can generate revenue. They obviously can. The question is whether they can generate profit before the capital markets lose patience.
Anthropic is proving you can build a frontier AI company without burning money at the same rate. That is a more dangerous competitive threat than any model benchmark.