This article compiles the annual AI trends report by Mary Meeker, the “Queen of the Internet,” under Bond Capital, which vividly depicts the panoramic view of how AI technology is reshaping the world at an astonishing speed.
(Background summary: Podcast highlights: AI and robots cover the cryptocurrency boom, the next era of micro-entrepreneurship is the hottest)
Mary Meeker, renowned as the “Queen of the Internet,” has made Bond Capital famous worldwide for its annual Internet trends reports. Recently, BOND released its latest report titled Trends – Artificial Intelligence, which extensively illustrates how AI technology is reshaping the world at an astonishing speed over 340 pages. This article is the second in a series summarizing the key points of the report, with the first article available here.
AI Users + Usage + Capital Expenditure Growth = Unprecedented
Unprecedented consumer/user AI adoption:
Phenomenal user growth: Using ChatGPT as an industry benchmark, its user growth rate is nothing short of phenomenal. In just 17 months, its weekly active user count skyrocketed by 8 times, reaching an astonishing 800 million. This figure not only demonstrates the attractiveness of AI technology but also reflects its rapid penetration into the mass market.
Global synchronous diffusion: Unlike previous technologies that typically spread gradually from their origin, the current wave of AI is characterized by global synchronous explosion. In terms of the proportion of users outside North America, the internet took 23 years to reach a 90% penetration rate, while ChatGPT achieved a similar level in just 3 years, highlighting the globalization and rapid dissemination capabilities of AI technology.
Surpassing products of the internet era: When compared to star products of the internet era, ChatGPT’s user acquisition speed is astonishing. It took ChatGPT only 0.2 years to reach the milestone of 100 million users, far outpacing the growth miracles of TikTok, Instagram, and Facebook.
Disruptive acquisition costs: Compared to foundational technological products in history like the Ford Model T, TiVo, and iPhone, ChatGPT not only leads in reaching millions of users/customers but also has an almost zero cost of acquiring initial users. This “freemium” or low-barrier strategy has greatly accelerated the popularization of AI technology.
Accelerated household penetration: In the United States, the time taken for new technology products to reach a 50% household penetration rate has halved with each generation. AI technology is expected to continue or even accelerate this trend, with the report projecting it will reach this milestone in about 3 years, significantly impacting all aspects of social life.
The flourishing ecosystem of NVIDIA:
As a core provider of AI computing power, NVIDIA’s ecosystem has experienced explosive growth over the past four years. The number of developers, AI startups, and applications using GPUs within its ecosystem has all achieved growth exceeding 100%. This indicates that the software and application ecosystem surrounding AI hardware is rapidly maturing.
Tech giants prioritize AI at the strategic top level:
Focus of earnings call conferences: From NVIDIA, C3.ai, Baidu to Google, Meta, IBM, Microsoft, Salesforce, and Intel, nearly all leading tech companies have made “AI” a core topic in their earnings call conferences, with a significant increase in mention frequency. This reflects the high recognition of the strategic position of AI by capital markets and corporate management.
Shared vision among executives: Leaders of major tech giants, such as Andy Jassy of Amazon and Sundar Pichai of Google, have publicly expressed their firm belief in the disruptive potential of AI, viewing it as a core strategy for their companies’ future development. Their statements not only depict the broad prospects of AI technology but also reveal the ambitions and layouts of industry leaders in the AI field.
Focus of S&P 500 companies: In the broader corporate world, the influence of AI is also rapidly expanding. The proportion of S&P 500 companies mentioning “AI” in their quarterly earnings call has surged from negligible levels in 2015 to 50% in the fourth quarter of 2024. This indicates that AI has transformed from a niche technological concept into a significant factor influencing mainstream corporate decision-making.
Shift in global corporate strategic focus: For global companies, the goals for utilizing generative AI over the next two years are more concentrated on areas that can directly lead to revenue growth, such as improving product/output efficiency, optimizing customer service, and enhancing sales efficiency, rather than merely cutting costs. This reflects the expectation of companies regarding AI’s value creation capability.
Active exploration among CMOs: In the marketing field, the application of AI is also gaining momentum. As many as 75% of global Chief Marketing Officers (CMOs) report actively using or testing AI tools to enhance marketing effectiveness and consumer insight capabilities.
Practical cases in finance, healthcare, and other industries: Cases such as Bank of America’s Erica virtual assistant, JPMorgan’s end-to-end AI modernization, Kaiser Permanente’s multimodal AI scribe, and Yum! Brands’ AI-driven restaurant management platform vividly showcase the practical application and value creation of AI in traditional industries like finance, healthcare, and dining.
Active integration of AI in education, government, and research:
Cross-disciplinary AI collaboration and application: From Arizona State University’s AI acceleration program to the deep collaboration between Oxford University and OpenAI, and the NextGenAI alliance formed by top institutions like MIT and Harvard, as well as ChatGPT Gov tailored for U.S. federal agencies, the integration of AI technology into critical areas like education, government, and research is accelerating. U.S. national laboratories have begun leveraging AI to drive breakthroughs in frontier scientific fields like nuclear energy and cybersecurity.
Rise of sovereign AI policies: As the strategic significance of AI technology becomes increasingly prominent, governments worldwide are paying more attention and beginning to formulate and adopt sovereign AI policies aimed at gaining control over AI development. NVIDIA’s Sovereign AI partner program has launched globally.
Accelerated approval and application of AI in healthcare: The number of AI medical devices approved by the U.S. Food and Drug Administration (FDA) is rapidly increasing, reflecting the immense potential of AI technology in enhancing medical diagnosis and treatment levels. At the same time, the FDA has announced ambitious plans to promote the use of AI within all its centers by June 2025.
AI empowers accelerated drug development: Companies like Insilico Medicine and Cradle are leading the application of AI in drug development, successfully shortening traditional medical research timelines by 30% to 80%. Cradle’s GenAI platform has astonishingly increased the speed of preclinical research by 1.5 to 12 times, significantly accelerating the process of new drug discovery.
Unprecedented growth in AI usage—expansion in depth and breadth:
AI popularity across age divides: In the United States, the use of AI tools like ChatGPT is rapidly increasing among adults of all age groups. Notably, the highest usage rate is among the 18-29 age group. OpenAI founder Sam Altman has even observed that “young people see it (ChatGPT) as a life advisor,” revealing the increasingly important role of AI across different generational demographics.
Surge in mobile application engagement: Active users in the U.S. have seen the daily time spent on the ChatGPT mobile application grow an astonishing 202% in just 21 months. Meanwhile, user session durations and the number of daily sessions per user have also exhibited significant growth trends.
High user retention rates highlight AI stickiness: The desktop user retention rate for ChatGPT has risen from about 50% to 80% in 27 months, far surpassing the retention rate of Google Search during the same period. This fully demonstrates the user stickiness and irreplaceability of AI tools.
AI helps improve work efficiency: Over 72% of employed adults in the U.S. using AI chatbots report that these tools significantly help them complete work tasks faster and better.
Widespread application of AI in education: A survey by OpenAI targeting U.S. students aged 18-24 reveals that ChatGPT has become an important supplementary tool for their learning and research, primarily used for writing papers, brainstorming, and conducting academic research.
Emergence of “deep research” capabilities: Leading AI companies like Google Gemini, OpenAI ChatGPT, and xAI Grok are launching product features equipped with “deep research” capabilities aimed at automating and enhancing knowledge acquisition and analysis in specialized fields.
Evolving AI agents—from response to execution: AI technology is undergoing a transformation from simple chat responses to executing more complex tasks.
Gradually Evolving into a New Type of Service Provider
AI agents are capable of reasoning, taking actions, and completing complex multi-step tasks on behalf of users, such as scheduling meetings and submitting reports. The market’s attention to “AI agents” has surged, with related Google search volumes increasing by over 1000% in 16 months. Meanwhile, industry leaders like Salesforce, Anthropic, OpenAI, and Amazon are accelerating the launch of their respective AI agent products.
General Artificial Intelligence (AGI) — The Next Frontier of AI
The BOND report further explores the immense potential of General Artificial Intelligence (AGI) and indicates that experts have significantly advanced their expectations regarding the timeline for achieving AGI. AGI is no longer seen as a distant hypothetical endpoint, but rather as an increasingly clear and attainable technological threshold. Once realized, AGI will fundamentally redefine the boundaries of software and hardware capabilities.
Unprecedented Growth in Capital Expenditure (CapEx) — Laying the Foundation for AI’s Future
The historical evolution of capital expenditure by technology companies shows a fluctuating upward trend over the past two decades, driven by increased demand for data storage, distribution, and computing. Early investments were primarily focused on building internet infrastructure, while the current focus has shifted to enhancing computing power for data-intensive AI workloads, which includes substantial investment in dedicated chips (such as GPUs and TPUs), liquid cooling technologies, and cutting-edge data center designs.
Massive Investments from Major Tech Companies
Taking the “Big Six” tech companies in the United States as an example, their capital expenditures have grown at an annual rate of 21% over the past decade, aligning closely with the rapid increase in global data generation (which has grown at an annual rate of 28%). These substantial investments are also directly reflected in the strong growth of cloud service revenues, with global hyperscale cloud service providers achieving an annual compound growth rate of up to 37% over the past decade.
The Profound Impact of AI’s Rise on Capital Expenditure
The scale of training datasets for AI models is expanding at an astonishing annual rate of 250%. To meet the growing demand for AI computing power, the capital expenditures of the “Big Six” tech companies have increased by 63% year-on-year, with the growth rate still accelerating. Notably, the proportion of their capital expenditures relative to total revenue has risen from 8% a decade ago to 15% today. For industry leader Amazon AWS, the ratio of capital expenditure to revenue reached an astonishing 49% in 2024, significantly higher than 4% in 2018 and 27% in 2013, underscoring the enormous funding requirements for AI/ML infrastructure development.
The Core Driving Force of GPU Performance Improvement
NVIDIA’s GPU performance has increased 225 times over the past eight years, and its installed GPU computing capacity has grown more than 100 times in just about six years. This has made NVIDIA a major beneficiary of this round of capital expenditure among tech companies, with its data center business revenue accounting for 25% of global data center capital expenditure, and this proportion continues to rise.
Concurrent Growth in R&D Investment
Alongside capital expenditures, research and development (R&D) investment has also increased. The ratio of R&D spending to revenue for the “Big Six” U.S. tech companies has risen from 9% a decade ago to 13% today. Fortunately, these industry giants have ample cash reserves (with significant increases in free cash flow and cash on their balance sheets) to support continued investments in AI and related capital expenditures.
Computing Expenditure as the Main Driver of Capital Expenditure
The training costs of AI models remain high and are rapidly increasing, despite a decline in inference costs. This “high and low” cost dynamic puts continuous pressure on budgets for cloud service providers, chip manufacturers, and enterprise IT departments.
Data Centers — The Key Infrastructure and Beneficiaries of the AI Wave
The enormous demand driven by AI has propelled data center-related expenditures to historic highs. The pace of data center construction is accelerating; for instance, the annual private construction investment value of data centers in the United States has achieved an average annual growth rate of 49% over the past two years. The capacity of newly constructed data centers far exceeds the fill capacity of existing facilities. A striking example is the Colossus data center by xAI, which was built in just 122 days, with its computing power rapidly growing from zero to 200,000 GPUs within seven months.
The Enormous Power Consumption of Data Centers
Data centers are veritable “power consumption giants.” The International Energy Agency (IEA) has explicitly stated, “Without energy, there is no AI.” Since 2017, the power consumption of data centers has grown at an annual rate of 12%, far exceeding the overall global power consumption growth rate. The United States accounts for 45% of global data center power consumption. Globally, data center power consumption has tripled over the past 19 years, with the U.S. leading in regional power consumption.
Rising Training Costs of AI Models + Declining Inference Costs = Convergence of Performance + Increased Developer Usage
Pages 129-152 of the report delve into a core and intriguing contradiction in the economics of AI models: the coexistence of persistently high training costs and rapidly declining inference costs.
The Subtle Balance of Cost Dynamics
Training the current state-of-the-art large language models (LLMs) has become one of the most expensive and capital-intensive investments in human history, often costing billions of dollars. Ironically, this race to build the most powerful general models may be accelerating the commoditization process of the industry and leading to diminishing returns. Meanwhile, the costs of applying and utilizing these models (i.e., the “inference” process) are dropping rapidly.
High Training Costs of AI Models
The estimated training costs for cutting-edge AI models have increased approximately 2400 times over the past eight years. This figure highlights the high barriers to entry in the field of AI model development.
Continuously Declining Inference Costs/Token
Significant improvements in hardware efficiency (for example, NVIDIA’s latest Blackwell GPU has reduced energy consumption for generating each token by a remarkable 105,000 times compared to the 2014 Kepler GPU) and breakthroughs in model algorithm efficiency have jointly driven the steep decline in inference costs.
Token as the Basic Measurement Unit in AI Inference Process
The reduction in costs directly affects the economics of AI applications. For instance, using NVIDIA GPUs, the energy required to generate each LLM token has decreased by 105,000 times over the past decade. The prices of customer-facing AI model inference services (calculated per million tokens) have dropped by as much as 99.7% in just two years.
Compared to historical key technologies like electricity and computer memory, the speed of improvement in AI’s cost efficiency appears even more rapid. This trend of “cost decline + performance improvement → increased adoption” is an eternal theme in the history of technological development, now replaying at a faster pace in the field of AI.
Convergence of Performance
As technology matures and competition intensifies, the performance scores of top AI models on benchmark platforms such as LMSYS chatbot arenas are rapidly converging. This means that the gap in core capabilities between models from different providers is narrowing.
Surge in Developer Usage
The significant reduction in inference costs has made AI experimentation and productization feasible and economical for nearly all developers. Simultaneously, the convergence of model performance has altered developers’ considerations when choosing models. They no longer need to pay a premium for pursuing absolute top-tier performance, especially when models can be effectively fine-tuned for specific application scenarios. The explosion of foundational models provides developers with unprecedented flexibility and options, which in turn drives the growth flywheel of developer-led AI infrastructure. As the report states: The AI Developer Next Door.
According to survey data from Stack Overflow, the adoption rate of AI tools among developers has rapidly risen from 44% in 2023 to 63% in 2024.
The number of AI-related developer repositories on the open-source community GitHub has grown by approximately 175% in just 16 months. Google processes Token volumes that have increased by 50 times year-on-year, while Microsoft Azure AI Foundry platform processes Token volumes that have also increased by 5 times year-on-year. The applications of AI developers are extremely diverse and extensive, covering all aspects of software development, from code generation and automated testing to project management.
The Road to Profitability Remains Long
Despite the explosive growth in developer usage, the persistently high training costs and continuously declining prices for inference services suggest that the path to profitability for AI model providers may still be long and filled with challenges.
》》Trends – Artificial Intelligence Full Report
