#GateEventPointsSystemLaunched #OpenAI二季度亏损扩大 OpenAI's Revenue Grew 18% Yet Was Overtaken—What Exactly Is Wrong with AI's Business Model?

#OpenAI二季度亏损扩大 OpenAI's Revenue Grew 18% Yet Was Overtaken—What Exactly Is Wrong with AI's Business Model?

OpenAI released its second-quarter data: revenue reached $6.7 billion, up 18%, yet it was overtaken by Anthropic, while losses continued to expand.
This is not merely a question of which company is ahead; the business model of the entire AI foundation model industry is undergoing a profound crisis.

The Fierce Race in AI and the Balance Between Profit and Loss

The data is brutal. OpenAI's revenue was $5.6 billion in the first quarter and $6.7 billion in the second quarter, with quarter-over-quarter growth of 18%, which looks good. But Anthropic's revenue during the same period was $8.5 billion—not only overtaking OpenAI but leading by $1.8 billion. More importantly, OpenAI's losses are expanding—this is the most dangerous signal.

Many people are still debating which model is smarter and which technology is more advanced. I want to say: don't be misled by technology's halo. When a company's losses are growing and it is overtaken by a competitor in revenue, this already points to a fundamental problem: today's AI business model may be heading in the wrong direction.

AI's “Power Plant Dilemma”: The Stronger the Technology, the Higher the Cost

Let's first look at a reality: generating a single image with OpenAI costs approximately 2.12 cents. Processing 1,000 tokens costs anywhere from 0.03 to 12 cents. How much do users pay? ChatGPT Plus costs $20 per month, which does not seem expensive.

Foundation Models Are Like High-Energy-Consuming Power Plants

But that is precisely where the problem lies. AI foundation models are like super power-hungry power plants: the more electricity they generate, the more coal they burn. The more frequently users use them, the more severely the company loses money.

My observation is that foundation model companies have fallen into a vicious cycle: to prove that their technology is leading, they must make their models larger and smarter; the larger the model, the higher the inference cost; the higher the cost, the harder it is to make money. This is a dead loop.

Why was Anthropic able to overtake OpenAI in revenue? My analysis is that it may have found a more effective customer base. Enterprise customers and developers—groups willing to pay for certainty—are the breakthrough point for AI commercialization. OpenAI, meanwhile, still relies heavily on consumer users, which happens to be the hardest market to monetize.

Lack of Product Thinking: A Foundation Model Does Not Equal a Good Product

Here I want to point out a fact many people are unwilling to admit: technological leadership does not equal product success. OpenAI has GPT-4o, Sora, and the world's most advanced foundation models. But so what? Users do not want technical specifications; they want products that can solve problems.

For example, what does a small or medium-sized business owner care about when using ChatGPT to write marketing copy? They care about the hassle of having to retune the prompt every time, the uncertainty of outputs that are sometimes good and sometimes bad, and concerns about data security. Has OpenAI solved these pain points?

I don't think so. Foundation model companies generally suffer from “technological narcissism”—believing that their models are smart enough and that users should adapt to them. This is completely misguided product thinking.

What does genuine product thinking look like? It means thinking from the user's perspective: What problems do you need to solve every day? What are you willing to pay for? What is your biggest pain point?

Judging from its financial data, OpenAI has clearly not gone far enough down this path. Revenue growth alongside expanding losses indicates that costs are out of control. Being overtaken by Anthropic indicates that it is half a step behind in commercial deployment.

Three Fatal Misconceptions in the Business Model

In my view, the business model of AI foundation models contains three fatal misconceptions.

First, excessive reliance on the API economy. Many foundation model companies think that opening up APIs for developers to use will allow them to build an ecosystem. But in reality, an API is merely a tool, not a solution. Developers want AI capabilities that are stable, affordable, and reliable—not services that become more expensive from time to time.

Second, neglecting vertical applications. Foundation models can do everything, but they do not do anything deeply enough. Healthcare, law, education, and finance—each vertical has unique needs and barriers to entry. General-purpose foundation models often struggle in these fields.

Third, underestimating enterprise needs. Enterprises do not want flashy AI demonstrations; they want AI tools that can be integrated into workflows, are secure and controllable, and deliver predictable results. This market has high barriers to entry, but also great value. Anthropic's overtaking OpenAI may well be the result of a breakthrough in this market.

My judgment is that the future winners in AI will not necessarily be the companies with the strongest technology, but those that best understand how to turn AI technology into commercial value.

The Way Forward: From a Technology Race to Value Creation

OpenAI's predicament has sounded an alarm for the entire AI industry. The speed at which the technology race is burning money has already surpassed the speed of commercial monetization.

So where is the way out?

First, move from general-purpose to vertical applications. Foundation models need to “go deeper,” penetrating specific industries and scenarios. A specialized model proficient in medical diagnosis may have greater commercial value than a general-purpose model that can do a little of everything.

Second, move from models to applications. Users do not buy models; they buy solutions. How to package AI capabilities into products that enterprises are willing to pay for will be the key to competition in the next phase.

Third, move from burning cash to generating cash flow. AI companies must find sustainable profit models instead of relying forever on investors for funding. This means strictly controlling costs, precisely identifying paying users, and establishing genuine commercial barriers.

OpenAI's second-quarter data appears on the surface to show revenue growth alongside being overtaken. At a deeper level, it reflects the collective anxiety over the business model of the entire AI industry.

The excitement surrounding technological breakthroughs has passed. It is now time to face the harsh commercial reality. AI is not magic; it needs to solve real-world problems, create real-world value, and earn real-world profits.

Over the next two years, we will see the first large-scale shakeout in the AI industry. Companies that only know how to build models but do not understand how to build products will gradually fall behind. Those that truly understand users, cultivate applications deeply, and establish sustainable business models will become the new winners.

OpenAI still has a chance, but time is running out. Every AI entrepreneur should read a clear signal from this set of data: technology is important, but business is more important. Technological breakthroughs without commercial support will ultimately be nothing more than toys in a laboratory.

The second half of AI is only just beginning$OPENAI ‌