A recent study by **SemiAnalysis** has found that Anthropic's **Claude** subscriptions may offer about five times more value than OpenAI's **ChatGPT** subscriptions for mid-tier models, based on calculations that compare the equivalent value of API access. This assessment comes as OpenAI has recently made changes to its subscription plans, reducing the token quota for its $200-per-month plan and introducing a new $500-per-month plan that offers only a 21% increase in access to GPT-6 Astra compared to the previous plan. Meanwhile, Anthropic's subscriptions, while contributing only 10% of its revenue, use over 40% of its computing power, highlighting the high cost of running these AI services. The study compared subscription plans from several companies, including Anthropic, OpenAI, Meta, SpaceXAI, MiniMax, Moonshot, Z.ai, Cursor, and Cognition. It noted that the value of a subscription can vary depending on the model and the type of work being done, as token costs—used to measure the amount of AI processing—can differ from API prices. For example, Anthropic's $20-per-month **Claude Pro** subscription provides 2.9 billion tokens per month, compared to just 1 billion for the **ChatGPT** subscription. However, other analyses suggest that the value of AI models can depend on the criteria used. While **SemiAnalysis** primarily focuses on token generation throughput, other firms measure the cost per specific task, which might make OpenAI's GPT-6.1 Sol more affordable for particular tasks compared to Anthropic's Claude Opus 5.5. Subscription plans are heavily subsidized compared to standard API access rates, which has led to the growth of a secondary market for reselling tokens from shared accounts in various regions. Anthropic and OpenAI have taken different approaches to manage their margins. Anthropic reduces the equivalent API value as more powerful models are deployed, while OpenAI has directly cut the token limits on its $200 subscription. These strategies reflect the ongoing challenges of balancing cost and performance in the AI sector. The impact of these subscription models is more pronounced for businesses, which do not benefit from the subsidized rates and are facing rapidly rising API costs as AI becomes more widely used. A report by **McKinsey** found that 93% of companies have exceeded their allocated AI budgets. To manage these costs, many companies are switching to open-source models, which can be hosted at a lower cost. For example, **Skydive** claims to have reduced its AI costs by more than 75% by using a model router that directs tasks to high-performance open-source alternatives. Token prices have dropped significantly due to increased competition and reduced production costs, especially with the rise of Chinese open-source models and strategic price cuts. This trend benefits end users but raises questions about the profitability of industry players. OpenAI, which raised $122 billion in funding earlier this year, is projected to face a cash flow loss of $280 billion by the end of 2030, despite generating approximately $840 billion in cumulative revenue. Anthropic, preparing for an initial public offering, is expected to record a loss of $42 billion for the 2025 fiscal year and spend over $518 billion on cloud, computing, and infrastructure. Analysts suggest that the AI industry needs to generate about $6,000 billion in annual revenue by 2031 to justify current investments, but current applications and employee productivity are only covering a fraction of this amount. The financial imbalance within the AI industry, with chip manufacturers' profits funded by investors' capital rather than real revenue, and AI application developers showing negative operating margins, creates a fragile business model that relies on future profitability. If end users do not quickly generate concrete returns on investment, the massive funding could suddenly dry up, potentially leading to a collapse in the AI sector.