- Article: AI Trends by Bond by Bond Capital
The Bond Capital report, authored by Mary Meeker and her team, posits that the artificial intelligence revolution is occurring at a pace and scale that is entirely unprecedented, dwarfing previous technological shifts like the internet and mobile. The core thesis is that AI is a “compounder” technology, building on decades of internet infrastructure, data accumulation, and computing power to create a moment of explosive, simultaneous global change.
The report identifies a “whack-a-mole” data environment where every metric—user growth, compute spending, model creation, and performance—is accelerating dramatically. This is creating a landscape of intense and complex competition among tech incumbents, emerging AI-native companies, and sovereign nations, particularly the USA and China, which are locked in a new “space race.”
Key dynamics highlighted are the unprecedented growth in AI users and capital expenditure (CapEx), a fundamental economic split between skyrocketing training costs and plummeting inference costs, and the resulting uncertainty around monetization and profitability. While AI offers immense benefits and productivity gains, its development is characterized by high cash burn, immense valuations, and a rapidly shifting competitive field where moats are hard to build and defend.
Key Themes and Observations
The report is structured around eight core themes, supported by extensive data and charts.
1. Change is Happening Faster Than Ever
Technology has been compounding for centuries, but the current AI-driven phase is accelerating at an unparalleled rate.
- Data & Compute Growth: The amount of data (+260% annual growth) and compute (FLOPs, +360% annual growth) used to train AI models has grown exponentially over the last 15 years.
- Model Proliferation: The number of large-scale AI models (using >10²³ FLOPs) has increased by +167% annually since 2020.
- Performance Milestones: In 2024, AI systems surpassed the human baseline (89.8%) on the MMLU benchmark for general knowledge. By early 2025, advanced models like GPT-4.5 were being mistaken for humans by testers over 70% of the time in Turing-style tests.
2. AI User + Usage + CapEx Growth = Unprecedented
This section details the three pillars of AI’s explosive growth.
- User Growth: ChatGPT reached 800 million weekly active users in just 17 months, a ramp 5.5x faster than Google Search to reach comparable query volumes. This adoption is materially faster and cheaper (often free) than any previous foundational technology, including the Ford Model T, iPhone, or Netflix.
- Global Adoption: Unlike the internet, which started in the USA and diffused globally over decades, ChatGPT achieved 90% of its user base outside North America within three years, demonstrating simultaneous, around-the-world adoption.
- Capital Expenditure (CapEx): The “Big Six” US tech companies (Apple, NVIDIA, Microsoft, Alphabet, Amazon, Meta) are projected to spend $212 billion in CapEx in 2024, a +63% year-over-year increase. This spending, now representing 15% of their revenue (up from 8% a decade ago), is overwhelmingly directed towards AI infrastructure like GPUs and data centers.
3. AI Model Economics: Training Costs Rise, Inference Costs Fall
A central tension in the AI economy is the divergence between the cost to build models and the cost to run them.
- Training Costs: The cost to train a frontier AI model has grown ~2,400x in eight years, with models now costing hundreds of millions and soon, potentially billions, of dollars.
- Inference Costs: The cost to run a model (inference) is plummeting. The price for customers to process 1 million tokens has fallen 99.7% in two years. NVIDIA’s Blackwell GPU is 105,000x more energy-efficient per token than its 2014 Kepler GPU.
- Impact: This makes AI more accessible and fuels an explosion in developer usage, but it also commoditizes the output, making it difficult for model providers to maintain pricing power.
4. AI Usage + Cost + Loss Growth = A High-Stakes Gamble
The combination of rapid user growth and high costs creates a precarious business environment.
- High Burn, High Valuation: Leading private AI companies like OpenAI and Anthropic have raised nearly 11 billion in annualized revenue. This has led to extremely high valuation-to-revenue multiples (e.g., OpenAI at 33x, Perplexity at 75x).
- Consumer Benefit: This dynamic is currently “good news for consumers,” who get access to powerful, often-free tools subsidized by venture capital and corporate balance sheets. The long-term profitability for the companies themselves remains “TBD.”
- Historical Parallels: The report draws parallels to the cash burn phases of Amazon, Tesla, and Uber, which eventually built sustainable, data-driven network effects but endured years of massive losses.
5. AI Monetization Threats: Competition, Open Source & China
The path to profitability is threatened on multiple fronts.
- Rising Competition: The number of new AI model releases (multimodal, language, vision, etc.) is exploding. In 2024-2025, there was a +1,150% rise in large-scale multimodal models released.
- Open-Source Momentum: Open-source models (e.g., Meta’s Llama, DeepSeek) are rapidly closing the performance gap with closed-source leaders. This provides developers with low-cost, customizable alternatives, reducing vendor lock-in.
- China’s Rise: China has become a dominant force, particularly in the open-source community and in specific applications like industrial robotics. This geopolitical competition is a primary theme of the report.
6. AI & the Physical World: Fast and Data-Driven Ramps
AI’s impact extends beyond screens into the physical world.
- Autonomous Vehicles: Tesla’s fleet has logged a cumulative ~4 billion fully self-driven miles, growing ~100x in just 33 months. In San Francisco, Waymo has captured 27% of the rideshare market in its operating zone.
- Defense & Industry: Companies like Anduril (defense), KoBold Metals (AI-driven mineral exploration), and Carbon Robotics (AI-powered weed removal) are embedding AI into physical hardware and operations, creating new data and efficiency loops.
7. Global Internet User Ramps: Powered by AI from the Get-Go
AI is set to fundamentally change how new users experience the internet.
- The Next 2.6 Billion Users: The report suggests that the next wave of internet users, many in developing nations, will have an AI-native, voice-first experience, potentially skipping traditional apps and browsers.
- Enabling Infrastructure: This is made possible by low-cost satellite internet from providers like SpaceX’s Starlink, which now has over 5 million subscribers and provides coverage to vast, previously unconnected areas.
8. AI & Work Evolution: Real and Rapid Transformation
AI is reshaping the labor market faster than previous technologies.
- Cognitive Automation: AI is moving from automating physical tasks to cognitive ones—reasoning, creating, and problem-solving.
- Job Market Impact: Since 2018, AI-related IT job postings have surged +448%, while non-AI IT postings have fallen by 9%. Companies like Shopify and Duolingo have declared themselves “AI-first,” making AI proficiency a baseline expectation for all employees.
- Productivity Gains: Early data shows significant productivity improvements. One study cited shows a +14% increase in hourly chats handled by customer support agents using AI.
The Geopolitical Race: USA vs. China
This is arguably the most critical undercurrent of the report.
- The New “Space Race”: The competition is framed not just as economic but as a battle for technological and ideological influence.
- USA’s Lead: The U.S. currently leads in public market capitalization (83% of the top 30 global companies are US-based), frontier model development (OpenAI, Anthropic), and total CapEx spend.
- China’s Rapid Ascent: China is aggressively closing the gap. It now installs more industrial robots than the rest of the world combined. Its domestic AI champions (e.g., DeepSeek, Baidu, Alibaba) are rapidly improving, achieving comparable performance to US models at a lower training cost. Chinese companies now lead the world in the number of open-source models released.
- The Battleground: The fight is for control over the entire AI stack, from semiconductor manufacturing (where Taiwan’s TSMC is pivotal) to data, models, and platforms. The report emphasizes that AI leadership could beget geopolitical leadership.