The race to build leading AI models is becoming a test of financial endurance. According to an analysis by The Information based on internal financial data, OpenAI could lose as much as $5 billion this year, while Anthropic is also expected to face losses in the billions.
The numbers point to a central challenge for the generative AI market: usage is growing, competition is intense, and the underlying systems remain expensive to train and operate. The result is a business model under pressure, even for companies at the center of the AI boom.
OpenAI's costs are scaling faster than revenue
OpenAI's operating costs could reach $8.5 billion this year, according to the source article. Revenue is expected to land between $3.5 to $4.5 billion, depending on second-half sales.
That gap explains why OpenAI could lose as much as $5 billion this year. The largest burden is the infrastructure required to train models and serve user requests through inference, the process that lets an AI system generate responses after it has been trained.
The Information reports that OpenAI's costs for model training and inference could reach $7 billion. Inference costs are expected to rise further when Apple introduces its ChatGPT integration, which could bring more usage and therefore more computing demand.
Cloud spending is a major part of the picture. OpenAI spends nearly $4 billion on renting Microsoft servers alone, despite receiving discounted computing power at $1.30 per Nvidia A100 chip per hour.
That relationship also highlights why Microsoft's AI strategy is closely tied to Azure. The source article notes that Microsoft's own AI products, including Copilot and Bing integrations, are underperforming by comparison, while demand for cloud infrastructure remains central to the economics of the market.
Training, data and staffing add to the pressure
Training costs are not limited to chips and servers. OpenAI's AI training costs, including data payments, could rise to $3 billion this year.
The company currently employs about 1,500 people and plans to grow further. According to The Information, personnel expenses could reach $1.5 billion by year-end.
Taken together, these figures show how many parts of the AI stack carry heavy costs at the same time. A leading AI company must fund research, model training, data access, inference systems and staff while also trying to turn consumer and enterprise demand into revenue.
The issue is not that generative AI has no value. The source article makes a narrower point: early doubts are forming over whether current investment levels are proportionate to the benefits companies and customers can measure today.
Anthropic faces a smaller but sharper gap
Anthropic's position appears more difficult on a smaller scale. A source familiar with the figures says Anthropic expects to spend over $2.7 billion this year, while its revenue is only a fifth to a tenth of OpenAI's.
Computing is the biggest line item. Anthropic estimated $2.5 billion for computing costs alone.
By the end of the year, Anthropic expects to generate approximately $800 million in annualized revenue, or $67 million per month. But that revenue is not fully retained by the company, because Anthropic has to share it with Amazon.
The comparison with OpenAI is important because both companies face the same broad market reality. Building and operating large AI models requires major spending before it is clear how much revenue customers will reliably generate across products and use cases.
Competition is widening across the AI market
The cost problem is not happening in a quiet market. OpenAI and Anthropic are competing with each other, but the field is broader than those two companies.
Meta is pushing open-source models. Smaller companies including Mistral and Cohere are also seeking opportunities, either regionally in Europe or in specific niches such as B2B data chat.
This matters because expensive AI development becomes harder to justify when customers have more options. Competition can make it harder for any single provider to turn technical leadership into durable pricing power.
Enterprise adoption also remains difficult to measure in some cases. Organizations may deploy chatbot systems as a general purpose technology without clear use cases for every employee. The source article points to Microsoft's Copilot and OpenAI's ChatGPT Enterprise as examples of products where it can be hard to systematically measure who is using AI, what they are using it for, and whether quality or speed is actually improving.
The next bet is broader capability
Growth may still come from new products and more capable models. OpenAI's SearchGPT is one possible growth area mentioned in the source article, though replicating ChatGPT's success is uncertain.
Other subscription products have struggled to make a large impact. Google's Gemini subscription service is cited as a competing product that has failed to make a significant impact, raising the possibility that ChatGPT might be a one-trick pony.
More versatile multimodal models could create new use cases, broader adoption and higher revenue. If efficiency improves at the same time, margins could also improve. But the source article notes that questions remain about the quality of these capabilities and the cost of producing multimodal content such as video.
The larger strategic bet is a breakthrough in scaling general reasoning capabilities. That could open new automation and business opportunities and help address current AI weaknesses, including generating bullshit.
For OpenAI CEO Sam Altman and others, that appears to be the reason large companies continue to put billions into research and development. Google CEO Sundar Pichai summarized the investment logic in his company's recent earnings call: "The risk of underinvesting is dramatically greater than the risk of overinvesting for us here."