Tech • IA • Crypto
You Pay, They Cash In! Who Really Captures the Value?
9.4/10TL;DR
Artificial intelligence costs are exploding in companies while industry giants capture most of the value, against a backdrop of geopolitical and financial reshuffling.
KEY POINTS
Runaway costs, uncertain ROI
Token spending doubles every 45 days in some companies, while productivity gains plateau around 5%. According to several studies, 16% of companies report negative returns on investment and 73% consider results below expectations. A phase of disillusionment is setting in, with nearly 70% of operators ready to reduce AI usage if results do not follow.
Widespread adoption nonetheless
At Uber, 99% of engineers use AI tools and more than 70% of pull requests come from automated agents. Despite initial cost overruns, these technologies are now seen as unbeatable in terms of value for money. AI is becoming an essential production tool, even if its short-term effectiveness remains hard to measure.
Revenue explosion for leaders
Anthropic reports $47 billion in annualized revenue in May 2026, compared to $10 billion for all of 2025, with a projection of $100 billion by the end of 2026. The enterprise AI market has grown from $1.7 billion to $37 billion in three years. A duopoly is emerging between OpenAI and Anthropic, concentrating most of the value.
A still unbalanced market
Models are often oversized relative to actual needs, leading to costly overconsumption. Companies lack tools to precisely control their usage. Moreover, part of the costs is artificially reduced through indirect subsidies via venture capital, distorting economic signals.
The decline of open source
Open source in production drops from 19% to 11%. It remains used for testing, but companies are shifting to proprietary models in production, seen as more reliable and easier to integrate. The real cost of open source, especially GPU infrastructure, is becoming a barrier compared to subsidized commercial offerings.
China tightens control
Beijing is considering restricting foreign access to models from Alibaba, ByteDance, and Zhipu. After widely distributing open models, China now seeks to retain value within its borders. This strategy brings AI closer to a sovereignty issue comparable to nuclear weapons.
A new geopolitical and economic battle
AI models capture not only revenue but also strategic data through user interactions. Every query feeds learning and strengthens the advantage of leaders. This dynamic is turning AI into critical infrastructure, at the heart of power dynamics between states and companies.
Toward financialization of consumer AI
In the United States, the idea of a $1,000 investment account for every newborn aims to democratize gains from the tech boom. Inspired by models like Australia, this approach reflects a strategy opposite to that of France, where redistribution prevails over capital-based logic.
CONCLUSION
Between soaring costs, profit concentration, and strategic state retrenchment, artificial intelligence is emerging as a major economic and geopolitical force, reshaping the global tech landscape.