# Financial Challenges for OpenAI and Anthropic: AI Model Costs Outpace Revenue
OpenAI and Anthropic are showing exponential revenue growth, but computing costs for training models are rising even faster. According to internal financial reports provided to investors, OpenAI forecasts $121 billion in computing expenses in 2028. Even with revenue doubling, the company expects $85 billion in losses — record losses for a public company.
Anthropic spends less on training, but the trend is similar: each step in model intelligence development requires ever greater investments. The companies report profitability using two metrics — with and without R&D expenses. Without computing costs, OpenAI reaches operating profit in 2026, Anthropic in an optimistic scenario.
Inference costs consume over half of revenue for both firms, though the share is declining. OpenAI also bears costs for free ChatGPT users who generate no revenue.
Differences in Monetization Models
OpenAI relies on converting free users to paid ones, while Anthropic focuses on the enterprise segment. Nearly all of Anthropic's revenue comes from business customers, ensuring stability.
Accounting methodologies differ: Anthropic includes sales through AWS and GCP partners in its revenue, OpenAI does not. This distorts direct financial comparisons.
Competitive dynamics emerged in OpenAI's response to the Claude Code release: the company redirected resources to Codex and strengthened its B2B efforts.
- Key Financial Metrics:
- OpenAI: break-even after the 2030s including training.
- Anthropic: sooner, but scenario-dependent.
- Inference: >50% of revenue for both.
- Free users: significant burden for OpenAI.
Preparation for IPO and Institutional Changes
Both companies plan IPOs in late 2026. To attract record sums, bankers accelerated Nasdaq's inclusion of new issuers in the index, opening access to larger capital pools.
Forecasts indicate continued billions in cash burn: investments are needed to maintain development pace. Break-even remains a distant prospect due to rising costs for frontier models.
Key Points:
- Training expenses outpace revenue: $121 billion vs. $85 billion losses for OpenAI in 2028.
- Two profitability metrics mask the real picture without R&D.
- Inference eats >50% of revenue, free traffic adds burden.
- Anthropic benefits from B2B focus and partner sales.
- 2026 IPO will require institutional investments to cover deficits.
Implications for the AI Industry
The situation highlights a systemic challenge: scaling laws demand exponential compute. For mid/senior developers, this means optimizing models for inference efficiency and exploring alternative paradigms that reduce compute dependency.
— Editorial Team
No comments yet.