OpenAI and Anthropic both missed their gross profit targets, with non-paying users and computing power becoming the main burdens.

The profit paths of the two leading AI companies are facing severe challenges.

According to The Information, both OpenAI and Anthropic have failed to meet their respective gross margin targets, with the unexpectedly high growth in inference costs being the main drag. This financial pressure is particularly pronounced as their user bases continue to expand, raising doubts externally about whether the two companies can achieve their gross margin target of over 60% by the end of this decade.

Data shows OpenAI's gross margin fell from 40% to 33% last year, below its own forecast of 46%. For Anthropic, the gross margin improved significantly from -94% in 2024, but its 2025 gross margin is expected to be 40%, still 10 percentage points lower than the previous target. Both companies are not currently facing funding difficulties, but investors are increasingly scrutinizing the sustainability of their business models.

Inference Costs Out of Control, Exceeding Forecasts

Inference costs, which are fees paid to cloud providers to power AI models responding to user requests, are the core reason for the pressure on these companies’ gross margins.

According to The Information, OpenAI's inference costs grew about fourfold year-on-year last year, reaching $8.4 billion, higher than its forecast of $6.6 billion last summer. The company explained to potential investors that, due to demand exceeding expectations, it had to purchase on-demand server resources from cloud service providers at higher prices. Cloud service providers usually charge a premium for on-demand server rentals compared to pre-booked prices, further driving up costs.

For Anthropic, inference costs are expected to increase more than threefold to $2.7 billion in 2025, also above previous forecasts. Notably, this cost growth has occurred while cloud computing rental prices are generally falling and both companies continue to claim improvements in model efficiency, making the deterioration in gross margin even more striking.

Free Users and High-Compute Products, Double Drag on Profits

OpenAI's gross margin pressure also comes from its massive group of non-paying users. According to The Information, OpenAI had around 910 million weekly active users, of which only about 5% are paying subscribers. Last year, nearly half (around $3.9 billion) of its total inference costs supported non-paying users, while costs corresponding to paying users were $4.5 billion.

Product structure is also a major factor. OpenAI's video generation tool Sora, launched last year, consumes much more server computing power than text-based queries; its inference model also requires more computing power to generate answers than traditional large language models. Additionally, before usage restrictions were introduced, the company allowed users to freely experience high-compute features, including the widely popular GPT-4o model for generating Studio Ghibli-style images—which, according to an insider, consumed a large amount of computational resources in a short period.

Efficiency Improvements Among Paying Users, Long-Term Goals Remain Uncertain

Despite overall gross margin pressures, OpenAI has made significant efficiency improvements in serving paying users. According to The Information, its compute profit margin for paying users (the share of revenue after deducting AI model operating costs) reached about 70% last October, up from around 52% at the end of last year and about 35% in January 2024.

For revenue structure optimization, OpenAI plans to boost monetization efficiency for non-paying users through advertising, e-commerce, and subscription expansion. In January, the company launched an ad-supported ChatGPT subscription service globally, priced at around $5–8 per month.

From a long-term perspective, OpenAI expects about 66% of its estimated $14.1 billion inference cost this year will serve paying users; by 2030, this proportion will rise to about 94%, with total inference costs expected to reach $85 billion and a gross margin target around 67%. However, how to achieve this goal amidst continuously rising costs remains the core challenge facing both OpenAI and Anthropic.

Risk Warning and DisclaimerThe market carries risks, and investments should be made cautiously. This article does not constitute personal investment advice and does not take into account specific users’ individual investment objectives, financial situations, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article are suitable for their specific circumstances. Investments made based hereon are at your own risk.