"The Internet Queen" on the AI Era: Decoding Mary Meeker's *AI Trends Report* | Yunqi Capital Techπ

Mary Meeker, the legendary "Queen of the Internet" and founder of Bond Capital, recently released her AI Trends Report, declaring AI "the biggest revolution since mobile internet." From the unprecedented growth velocity of products like ChatGPT, to the "scissors gap" between AI training and inference costs, to the evolving landscape of US-China tech competition and the three major challenges facing AI commercialization, the report maps out a new landscape of AI technology, industry, capital, and geopolitics through 10 core insights.

As AI enters the second half of industrialization and global competition, how should those riding this wave grasp the industry's logic and technological rhythm? This episode of "Yunqi Tech π" breaks it down for you.

Source: WeChat account "i商周" | Original title: "Ten Key Takeaways to Understand the 'Queen of the Internet's' First AI Trends Report"

Mary Meeker, widely known as the "Queen of the Internet," recently released a 340-page Trends – Artificial Intelligence report that has once again captured the global tech community's attention.

Meeker's fame stems from the Internet Trends Report she began publishing in 1995, widely regarded as the "bible" of the internet industry. She is currently founder and general partner at the venture capital firm Bond. Before founding Bond, she led growth investing at Kleiner Perkins from 2010 to 2019, backing companies including Facebook, Spotify, Ring, and Block (then Square).

The AI Trends Report continues Meeker's signature sweeping perspective, spanning everything from the invention of the printing press to Roomba vacuum cleaners, attempting to paint a panoramic picture of the AI era. Meeker describes the pace of AI development as "unprecedented," calling this wave of technological change the "biggest revolution" since mobile internet. In her view, AI represents not merely the next chapter of technological evolution, but a core force that will reshape the global economy and geopolitical landscape. Here are the report's 10 core insights:

01 The Unprecedented Speed of AI's Rise

At the outset, Meeker emphasizes that the transformative speed of this AI wave is unprecedented — she uses the word "unprecedented" 51 times throughout the report.

The report reveals staggering growth figures. OpenAI's ChatGPT reached 800 million users in just 17 months, outpacing any technology in human history. In terms of global expansion, ChatGPT achieved 90% of its users from outside North America within three years — a milestone that took the internet 23 years to reach. ChatGPT's annual search volume hit 365 billion, 5.5 times Google's peak, which took 11 years to achieve. This growth rate is unmatched by any technology product in human history.

02 The AI Scissors Gap: Soaring Training Costs, Plummeting Inference Prices

The report highlights one of the most fascinating contradictions in AI model development: the cost of training cutting-edge AI models is extraordinarily high and climbing, while the per-unit cost of running these models (inference) is falling rapidly. Estimates suggest that training costs for the most advanced AI models have risen approximately 2,400-fold over the past eight years. Anthropic CEO Dario Amodei predicts these training expenses could jump from the current $100 million baseline to $1 billion in 2025, and even reach an astonishing $10 billion between 2025 and 2027. Meanwhile, inference costs have plummeted 99.7% in just two years, making once-prohibitively complex AI tools economically viable and igniting a new wave of innovation.

03 High Growth, High Valuation, and High Burn

The report notes that alongside explosive user growth, capital expenditures have ballooned dramatically. The combined capex of America's "Big Six" tech companies reached $212 billion in 2024, up 63% year-over-year. As a share of total revenue, these expenditures climbed from 8% a decade ago to 15% in 2024. Amazon AWS alone directed 49% of its 2024 revenue toward AI/ML infrastructure. Meeker points out that the three leading American AI startups — OpenAI, Anthropic, and xAI — now command a combined valuation approaching $400 billion against projected annual revenue of $12 billion. As of May 13, 2025, total funding raised by major private AI model companies including OpenAI, Anthropic, xAI, and Perplexity reached approximately $95 billion. As generative AI enters productization and scaled deployment, whether this "burn now, profit later" business model can sustain itself hangs like a sword of Damocles over every AI startup. The report draws parallels between the current frenzy of AI investment and high-risk environment and previous major technology transformation cycles — the 19th-century railway boom, or the dot-com bubble at the turn of the millennium. These cycles typically featured initial irrational exuberance, massive capital inflows, and brutal competition, ultimately winnowing out true industry leaders.

04 Three Challenges to AI Commercialization: China's Rise Commands Attention

Despite AI's dazzling future, its path to commercialization faces three major challenges: cutthroat market competition, the disruptive impact of rapidly improving and low-cost open-source models, and China's swift emergence as a new AI superpower. Large multimodal model releases surged 1,150% over the past two years, while language model releases grew 420%. In a single week — May 19, 2025 — industry heavyweights including Google, Microsoft, Anthropic, and OpenAI unleashed a barrage of new products and features.

Open-source models are gaining remarkable momentum. Meta's Llama series downloads grew 3.4-fold in just eight months, reaching 1.2 billion by April 2025. More critically, open-source model performance is rapidly catching up to and even surpassing proprietary alternatives. DeepSeek's R1 model, developed by the Chinese company, scored 93% on the challenging MATH Level 5 benchmark, very close to OpenAI o3-mini's 95%. Statistics from 2017 to 2024 show China has pulled even with the US in the number of "large-scale" AI systems released. Domestic tech giants including DeepSeek, Alibaba (with its Qwen models), and Baidu (with its Ernie models) are rapidly improving model performance and dominating the Chinese domestic market. Additionally, China's installed base of industrial robots now exceeds the rest of the world combined. Notably, Chinese public optimism about AI's benefits — 83% of respondents believe benefits outweigh drawbacks — significantly outpaces the US, where only 39% hold this view.

05 Accelerating Intelligence in the Physical World

AI's influence is extending beyond the purely digital realm, penetrating and reshaping the physical world as "physical intelligent agents" emerge. Autonomous vehicles stand out as a leading example. Tesla's FSD (Full Self-Driving) cumulative miles driven grew roughly 100-fold in just 33 months, exceeding 3.5 billion miles by March 2025. Waymo's robotaxi service, meanwhile, captured 27% of San Francisco's ride-hailing market within 20 months of launch as of April 2025, starting from zero.

This intelligence wave is spreading across multiple industries. Applied Intuition, focused on intelligent vehicle software, now serves 18 of the world's top automotive OEMs and has expanded into trucks, construction equipment, and defense. In defense tech, Anduril — centered on AI-driven autonomous systems — has doubled its annual revenue for two consecutive years.

06 Where Is the Next Billion?

Approximately 2.6 billion people — nearly one-third of the world's population — remain offline. The report offers a striking insight: when these potential new internet users come online, their first experience of the digital world will likely differ fundamentally from previous generations. Rather than exploring through traditional browsers and search engines, they may interact directly through AI-driven, native-language, conversational, multimodal intelligent agent interfaces. The global proliferation of AI applications lends weight to this prediction. ChatGPT's mobile app accumulated 530 million monthly active users in just 23 months, with India (14%), the US (9%), and Indonesia (6%) as top source markets. Similarly, DeepSeek's mobile app attracted 54 million MAUs in four months, concentrated in China (34%) and Russia (9%). Enabling this trend is the spread of low-cost satellite internet technology. SpaceX's Starlink has been particularly notable, with user numbers growing at a compound annual rate of 202% over the past three-plus years, surpassing 5 million total users and providing unprecedented connectivity — especially in remote and underserved regions.

07 AI's Impact on Labor: Not Replacement, but Redefinition

Addressing widespread fears of "AI replacing human jobs," the report offers a more nuanced view: cognitive automation is emerging, with AI systems capable of reasoning, creating, and problem-solving, transforming how work gets done. The report states that AI demonstrates core competence in tasks relying on large volumes of structured historical data with rule-based decision-making and judgment, which will inevitably automate certain existing jobs. However, looking at technological history, major advances have typically eliminated old roles while spawning entirely new occupations and fields. Going forward, human roles may shift toward supervising, guiding, and training AI systems, and collaborating with them to accomplish more complex and creative tasks.

As NVIDIA CEO Jensen Huang emphasizes: "People won't lose jobs because of AI itself, but likely because of competitors who know how to use AI effectively."

08 A New Geopolitical Chessboard: AI Supremacy as Core National Interest

Leadership in AI technology has evolved into a central focus of competition between nations, particularly between the US and China. This contest concerns not just future economic power, but directly shapes the architecture of global influence. Meta CTO Andrew Bosworth has unambiguously likened the current AI development dynamic, especially competition with China, to a new "Space Race."

National strategies clearly reflect this urgency. China is pushing full throttle on key technology industries including AI, robotics, electrification, and advanced information technology. The US has responded through legislation like the CHIPS and Science Act (this portion is already somewhat dated, being a Biden-era initiative), seeking to revitalize domestic semiconductor manufacturing and ensure resilience and security in critical technology supply chains. In this high-stakes game, Taiwan's TSMC holds absolute dominance with 80-90% market share in the world's most advanced semiconductor manufacturing.

By comparing the lists of the world's 30 most valuable public companies and tech companies in 1995 versus 2025, the report vividly illustrates the profound shift in global tech leadership: American companies' share has risen significantly (from 53% to 83% on the overall list, from 53% to 70% on the tech list), Japanese companies have declined dramatically from their former prominence, and Chinese companies have emerged as a new force and significant presence on these rankings.

09 A Double-Edged Sword: Boundless Potential Alongside Inherent Risks

Running throughout the report is an awareness of AI's dual nature: it harbors enormous potential to bring unprecedented benefits to society, while carrying multiple and significant risks that cannot be ignored.

The authors echo Stuart Russell and Peter Norvig's classic perspective: human civilization itself is the product of human intelligence, so if we can harness machine intelligence far surpassing our own, it would undoubtedly elevate our collective ambitions as a species. AI and robotics promise to liberate humanity from drudgery, dramatically improve production efficiency of goods and services, and potentially usher in an era of universal prosperity and peace. AI's capacity to accelerate scientific discovery could bring revolutionary breakthroughs against global challenges like curing diseases, addressing climate change, and solving resource scarcity. As DeepMind CEO Demis Hassabis envisions: "First we solve AI, then we use AI to solve everything else."

Yet the path to this bright vision is strewn with potential minefields. The report emphasizes that long before we can claim to "solve AI," we will have to confront risks from its misuse. These risks are already emerging, including:

• Development of lethal autonomous weapons systems
• Potential for mass surveillance and precision information manipulation
• Entrenchment and amplification of algorithmic bias in decision-making
• Severe disruption to existing labor markets
• Failures in safety-critical applications
• Entirely new cybersecurity threats

The late physicist Stephen Hawking's warning still resonates: "Success in creating AI could be the biggest event in the history of our civilization. But it could also be the last — unless we learn how to avoid the risks."

10 Invest Cautiously, Maintain Strategic Flexibility

For investors, Meeker's advice is unusually sober: "Only invest what you can afford to lose." She emphasizes that the current AI market presents an illusion of invincibility — everything seems to go up — but it is precisely when everything rises that risks lurk where least expected. She reminds investors to focus on long-term value creation, not short-term bubbles.

Facing AI's high-stakes, high-risk, high-reward battlefield, Meeker quotes her favorite venture capital adage: "Only invest money you can afford to lose." She compares the current AI startup frenzy to past eras of Uber, Amazon, and Tesla — all long-term money-losers that ultimately became giants.

She warns investors that while this AI wave seems to lift everything, "never put all your eggs in one basket." Once market enthusiasm cools and profit prospects dim, valuation bubbles can burst at any moment.