Top Insights From Mary Meeker's AI Trends Report
'ChatGPT reached 800 million users in 17 months' and more insights from Mary Meeker's report.
Top Insights From Mary Meeker's AI Trends Report
Mary Meeker has revealed her “Trends — Artificial Intelligence” report, frequently using the word “unprecedented” to describe everything about AI. Here are the top insights from Meeker’s report.
1. ChatGPT’s Growth—Unprecedented
ChatGPT reached 800 million users in 17 months (compared to Netflix’s 100 million in over 10 years). While Instagram took 2.5 years and TikTok 9 months to reach 100 million users, ChatGPT took only 2 months. Also, while the internet took 23 long years for 90% of users to be outside North America, ChatGPT did this in its third year itself.
2. China’s AI Wave
China released three large-scale open-source AI models in 2025, leading the world, and DeepSeek took only a few months to take 21% of the global LLM user share. DeepSeek R1 scored 93% vs. o3-mini’s 95% score at far lower training costs, and Alibaba’s Qwen2.5-Max surpassed other leading models like GPT-4o and Claude 3.5 on key benchmarks.
3. India’s Fast-Paced AI Adoption
India is a major market for AI businesses. Compared to nations like the US (8.9%) and Germany (3%), India is ChatGPT’s second-biggest market and accounts for the largest percentage of its mobile app users (13.5%). For DeepSeek, India ranks third in terms of users (6.9%).
4. Massive Capex On AI And Data Centres
The capital expenditure of the big six US-based tech companies on AI is unprecedented: $212 billion, with a 63% YoY growth. Fuelled by AI-led demand, data centre spending has reached all-time highs. Global IT company data centre capex hit $455 billion in 2024 and continues to rise.
5. AI Model Training Costs Rising, Inference Costs Falling
Large language model training costs are reaching billions of dollars, with AI model training compute costs rising 2,400x over eight years. At the same time, inference costs (the cost of applying AI models) fell 99.7% over the past two years. Cost efficiency gains are being realised faster from AI than they did from older technologies.
6. The Explosion Of Open Source Models
Open-source models are gaining due to their lower costs: Meta’s Llama model had 1.2 billion downloads in 10 weeks, posting 3.4x growth in eight months; there’s been 33x growth on Hugging Face since March 2022; and over 100,000 derivative models are based on Meta Llama itself.
7. Physical Application of AI Growing Faster Than Software
Physical world AI produces datasets that are impossible for software-only AI to duplicate. Its applications include autonomous vehicles and AI-powered auto intelligence, robotics, critical infrastructure, defence, agriculture and more.
