Mary Meeker's "Trends: Artificial Intelligence" report, released in May 2025, predicts several future AI development trends.

Mary Meeker is known as the "Internet Queen" and holds significant influence in technology, the internet, and investment, making her one of the most influential investors in Silicon Valley. As a venture capitalist, Meeker has invested in numerous top internet companies, including Facebook (now Meta), Airbnb, Spotify, Slack, Square (now Block), Pinterest, Snap, Instacart, and more.

Mary Meeker began her career as a renowned Wall Street analyst, working at Morgan Stanley, where she led the IPOs of multiple well-known internet companies (including Google). She is famous for her "data + story" research method, which allows her to deeply interpret technological products and analyze business models from a financial perspective, making her thrive in the integration of technology and capital.

Below are the core contents of the report:

01 Global AI User Growth and Utilization Rate

The world is changing at an unprecedented speed; the growth rate of AI users is unprecedented; the growth rate of AI capital expenditures is unprecedented.

The report points out that the global growth rate of AI users surpasses any technological wave in history, representing an "exponential" explosion. Taking OpenAI as an example, its ChatGPT reached 530 million monthly active users within 23 months, becoming the fastest AI product to reach hundreds of millions of users. Meanwhile, China's DeepSeek reached 54 million users within four months, with 34% from China and 9% from Russia, demonstrating the diverse penetration of AI platforms globally.

Research shows that AI penetration in US enterprises increased by 21% quarter-on-quarter in Q1, and AI IT positions in the US increased by 448%, while non-AI positions decreased by 9%.

The report also emphasizes that AI technology not only remains at the digital level but also plays the role of a "digital brain" in the physical world, driving the rapid development of scenarios such as autonomous driving, national defense, and robotics. Tesla's FSD (Full-Self Driving) increased its mileage by 100 times within 33 months; Anduril's defense business doubled its revenue for two consecutive years; KoBold Metals achieved higher exploration efficiency in mining with the help of AI.

02 Changes in AI Computing Power and Inference Costs

AI training costs have increased by 2400 times from 2016 to 2024, rapidly jumping from the millions of dollars to hundreds of millions to billions of dollars. The CEO of Anthropic stated in 2024 that the training cost of current top models is approximately $100 million and could reach $1 billion by 2025.

Despite the high training costs, inference costs have been continuously declining. For example, the inference energy efficiency of NVIDIA's Blackwell GPU is 105,000 times higher than that of the Kepler architecture in 2014.

The report specifically mentions that the decline in inference costs has led to an explosive growth in the developer ecosystem. For instance, the number of AI developers in the NVIDIA ecosystem has increased sixfold in seven years, reaching 6 million; the number of AI developers in the Google ecosystem has also grown to 7 million; and the number of AI models on Hugging Face has surged 33 times in 16 months, promoting the development of the open-source model ecosystem.

This decrease in inference costs not only lowers the development threshold but also promotes the convergence of AI model performance, giving more developers the opportunity to use AI for product innovation.

03 AI Business Models and Monetization Paths

The report reveals multiple trends facing AI commercialization. On the one hand, AI is gradually evolving from "tool-based charging" to "embedded intelligence charging", with initial breakthroughs seen in vertical industries (such as FDA-approved AI medical devices, real-time pricing optimization, etc.).

On the other hand, the rise of open-source models is challenging the commercial barriers of large model platforms. The report shows that developers are no longer reliant on a single AI vendor but prefer to select the optimal model from multiple vendors (such as OpenAI, Anthropic, Meta, Google, Mistral, etc.), forming a "developer-driven infrastructure growth flywheel".

The report points out that the R&D cycle in the medical field has been shortened by 30%-80% due to AI. Data from Insilico Medicine and Cradle both indicate that AI has brought revolutionary changes to drug development.

However, the report also cautions that while inference costs are declining, the training costs of large models are rising (expected to reach $1 billion by 2025), which will further test the commercial sustainability of AI companies.

04 AI's Disruption of the Physical World

AI is gradually breaking down the barriers between the digital and physical worlds. The report mentions that KoBold Metals uses AI for mining exploration, optimizing the mineral discovery process; Carbon Robotics' AI weeding robots help farms reduce herbicide use by 100,000 gallons; Anduril's defense business has doubled its revenue for two consecutive years, demonstrating the huge potential of AI in the defense sector; Tesla's FSD feature not only reshapes the future of autonomous driving but also accumulates data on a large scale globally (billions of kilometers), further training and improving the AI system.

The report notes that the core value of AI is no longer limited to the software level but has penetrated every aspect of the physical world, becoming a new generation of "digital infrastructure".

05 Diversification of the AI Developer Ecosystem

The report emphasizes that the AI developer ecosystem is undergoing structural changes. The adoption rate of AI tools reached 63% in 2024, a significant increase from 44% in 2023, making them a standard for developers; the number of open-source repositories for AI projects on GitHub increased by 175% in 16 months.

The number of tokens processed monthly by Google's AI developer ecosystem grew from 9.7 trillion in May 2024 to 480 trillion in May 2025, a 50-fold increase annually, indicating unprecedented usage and dependence by AI developers.

This flywheel effect significantly compresses the time from concept to prototype and from prototype to commercialization, greatly promoting the large-scale implementation of AI.

06 Geopolitical Competition in AI

The report points out that AI is not only a commercial race but also a strategic competition between nations. The United States occupies the first tier of global AI with its leadership in model innovation, custom chips, and cloud computing. China is rapidly catching up in open-source ecosystems, national infrastructure, and national AI strategies, vigorously promoting AI development through cooperation between national teams and technology giants.

The Stanford AI Index report states that Chinese AI users have much higher confidence in AI products than American users, revealing regional differences in the AI consumer market.

The report cites Microsoft President Brad Smith's viewpoint: "The technology market has network effects, and the international competition in AI is likely to be won by the fastest pioneers." He calls on the US government to adopt a global strategy to support the AI industry and avoid falling behind in global competition.

07 Future AI Development Trends

AI will become the default entry point of the digital world. In the next decade, AI will no longer be a "plug-in" feature but will become the mainstream entry point for the internet and mobile devices, similar to the popularity of the internet 20 years ago.

AI globalization and localization advance simultaneously. Global platforms such as OpenAI and Anthropic are expanding rapidly, but local AI platforms like China's DeepSeek are also growing rapidly in user numbers, creating a dual-cycle pattern in the global AI ecosystem.

Diversification of AI commercialization models. AI will evolve from "tool-based charging" to "embedded intelligence charging", especially in vertical scenarios such as industry software, supply chain management, and intelligent manufacturing.

Deep integration of AI and the physical world. Autonomous driving, intelligent manufacturing, AI weeding robots, etc., will become the main directions of AI industrialization, promoting the deep integration of the digital and physical worlds.

AI security and governance issues. The report warns that the popularization of AI brings a series of issues such as data security, model bias, and national security. Competition between China and the US and the rise of open-source ecosystems have increased the complexity of AI governance.

END