The Profitability Dilemma in the Large Model API Market: Revenue Soars, Yet Losses Reach Tens of Billions of Dollars

Recently, OpenAI and Anthropic have both initiated IPO plans, with market valuations approaching the trillion-dollar mark, reflecting investors' strong optimism toward the AI industry. However, Sun Mingchun, Senior Economic Advisor at Tencent Research Institute, and Cheng Wanching, Business Analytics Manager, point out that high revenue growth does not equate to profitability. The large model industry is currently trapped in an economic paradox: 'revenue without profit'.

The 'Monopolistic Competition' Structure Limits Pricing Power

Microeconomic analysis reveals that the large model API market currently exhibits characteristics of 'monopolistic competition' rather than the expected monopoly or oligopoly. Although R&D costs are extremely high, abundant investment capital, the widespread adoption of open-source models (which lower learning barriers), and frequent talent mobility have made market entry more feasible than anticipated.

Statistics show that over 500 institutions worldwide are engaged in large model development. On the OpenRouter platform alone, more than 70 vendors offer over 400 different models.

Under this structure, vendors lack stable pricing power. The emergence of model aggregation gateways (AI Gateways) further reduces users' switching and search costs, making them highly sensitive to 'value for money'. Once a model with similar performance but lower price appears, traffic rapidly shifts, making it difficult for vendors to maintain supernormal profits.

High Costs and Operational Pressures

Even industry leaders face severe financial pressure. For example, OpenAI is projected to exceed $20 billion in annualized revenue by 2025, but internal documents suggest it may still incur a $14 billion loss in 2026. Anthropic has seen growth in operating profit, but after accounting for high equity compensation costs and continuous R&D investment, net profit may remain negative.

Due to the rapid pace of technological iteration, many vendors must invest heavily in next-generation models before recouping costs from their first-generation models, resulting in prolonged periods of losses.

From 'Selling Tokens' to Building Differentiated Moats

Tencent Research Institute argues that a simple 'token-selling' business model cannot achieve long-term profitability. The market may eventually evolve into an oligopoly, but even then, if vendors continue price competition rather than quantity competition, the initial massive R&D investments may never be recovered.

To break this impasse, large model vendors must establish differentiated moats. The research recommends the following strategies:

'AI+' Model: Embed AI capabilities into existing products or services to enhance customer stickiness and increase the value of current business lines.

Contract-Based Customization: Deeply integrate with enterprise-specific data and workflows to increase switching costs and gain greater pricing power.

Ecosystem Development: Build competitive advantages in industry adaptation, enterprise workflows, and application ecosystems to reduce users' price sensitivity.

In conclusion, Tencent Research Institute emphasizes that while the technological value of large models is undeniable, vendors unable to build moats beyond model capabilities will face serious sustainability challenges amid fierce market competition.