Some figures from the report worth knowing:
The cost of inference has dropped by 99% over the year. Demand for computing on OpenRouter has increased 25 times since December 2024. When intelligence becomes almost free, the market explodes not just in conversation but in real usage.
Investments in data centers in 2025 are around $500 billion. By 2030, ARK expects $1.4 trillion. This is not a boom of models, but a giant cycle of infrastructure restructuring.
AI agents have learned to reliably perform tasks 5 times longer over the year - from 6 to 31 minutes of human work. The cost of coding has dropped by 91% in 8 months: from $3.50 to $0.32 per million tokens.
The path to purchase collapses from an hour to 90 seconds. By 2030, AI agents could conduct $8+ trillion in online purchases and capture 25% of global online sales. The share of AI search will grow from 10% to 65% of traffic.
Cursor has surpassed $1 billion ARR. Harvey, OpenEvidence, Sierra - around $100 million. The market is no longer paying for "access to the model," but for ready-made vertical solutions.
In pharma, artificial intelligence can reduce the cost of bringing a drug to market by 4 times, from $2.4 billion to $0.7 billion. By 2030, about a third of FDA-approved diagnostic systems will be AI-powered.
Geopolitics: Chinese models lag behind American ones by about 6 months but are limited by hardware - TSMC produces 38 times more computing power than SMIC. Leadership in artificial intelligence is no longer about algorithms, but access to chips, factories, and energy.
Artificial intelligence is no longer an option. It is the new foundational layer of business.
