Mary Meeker's team releases "Artificial Intelligence Trends Report" (PDF file) - AI Tutorial Resources

Mary Meeker's team released "artificialintelligentThe report, "Development Trends," provides a comprehensive analysis.AIThe explosive growth and global impact of technology. The report points out that...AIUser growth (such as Chat)GPT(reaching 800 million users within 17 months) and capital investment (the six largest US tech companies by 2025)AIWith spending exceeding $200 billion, far surpassing any previous technology, its performance metrics (such as MMLU test accuracy of 89.8%) are approaching human levels. The US and China dominate global competition, with the US excelling in large language models (such as...).GPTChina leads in industrial robots (accounting for 50% of the global market) and domestic models (such as Qwen 2.5).fastTo catch up.AIIt has penetrated deep into finance (such as Bank of America).AIErica (assistant), medical (FDA approved nearly 700 items)AIEquipment), education (such as Arizona State University)AIThe report covers areas such as courses. Within the next five years, human-level text generation and personalized medicine may be achieved, but risks such as autonomous weapons and job losses must be considered. The report emphasizes that...AIThe compound growth of algorithms (algorithms × computing power × data) is reshaping the economic, social, and geopolitical landscape, calling for a balance between innovation and ethical security.

Get Mary Meeker's《Trends – Artificial Intelligence》  Original PDF fileScan the QR code to follow and reply:20250601

Is the rate of change faster than before? Yes, it certainly is.

• The speed of technological change is unprecedented.Comparison of the Internet era (1990s) andAIThe speed of technology adoption in the 2020sAIIts user growth and technology iteration speed far exceed that of any previous technology. ChatGPTIt reached 800 million users in 17 months, while the internet and the iPhone took 7 years and 3 years respectively to reach the same scale.
• Technological compound growth effect:AIPerformance improvements are driven by a combination of algorithmic improvements, increased computing power, and expanded data volume, creating a "compound growth" effect. MMLU test accuracy jumped from 30% in 2019 to 89.8% in 2024 (approaching human-level performance).
• Developer ecosystem explodesNVIDIA's developer count has grown by 100% in four years (3 million in 2021 → 6 million in 2025), reflecting...AIThe pace of technology implementation is accelerating.
• Capital investment surgeThe top six US tech companies (Apple, NVIDIA, etc.) by 2025AIRelated capital expenditures exceed $200 billion, far exceeding the investment during the dot-com bubble.
• Global simultaneous popularization:AITechnology (such as Chat)GPTSince its release, it has achieved simultaneous growth in multiple regions around the world, whereas the early Internet spread only in Europe and the United States.

AIUser, usage, and capital expenditure growth = Unprecedented

• Consumers/UsersAIThe speed is unprecedented.:
  • ChatGPTUser growthWithin 17 months, the number of users reached 800 million (P.52), a growth rate 5.5 times that of the Internet.
  • Global popularityChatGPTLaunched across multiple regions, with a significant proportion of users outside North America.fastIt has climbed to over 50% (it took the internet 20 years to reach the same percentage).
  • NVIDIA Developer EcosystemThe number of developers will increase by 100% within 4 years (3 million in 2021 → 6 million in 2025).
  • Google ecosystem:GeminiThe number of developers increased fivefold to 7 million within a year.
  • Six major technology companiesApple, NVIDIA, and others in 2025AIRelated capital expenditures exceed $200 billion.
  • Investment direction:AIInfrastructure (such as data centers),Large Modeltrain,open sourceEcology.
  • Bank of America Case:AIErica, the assistant, has accumulated 2 billion interactions, saving 40% in customer service costs.
  • JP Morgan Practice:AISave thousands of hours of manpower and optimize risk management.
  • FDA approvalAIequipmentThe number of items increased from 0 in 1995 to nearly 700 in 2023.
  • Government policiesNVIDIA partners with 30+ countries on "sovereignty"AI"plan.

  AIHigh/increasing model computation costs + decreasing cost per inference = performance convergence + increased developer usage

  • Training costs continue to rise:Large ModelTraining costs,GPT-4 Training costs exceed $100 million (P.129), with an average annual growth rate of 200% (2019-2025).
  • Cost of a single inferencefastdecline:GPT-3 toGPT-4o reduces the cost of a single inference attempt by 90%. Anthropic Claude The reasoning cost is 3.5.GPT-1/5 of 4.
  • Performance convergenceMainstream models (GPT-4.5、Claude Both 3.7 and Qwen 2.5 achieved an accuracy of over 85% in the MMLU test.
  • Developer ecosystem explodesNVIDIA's developer base is projected to grow by 100% in four years (3 million in 2021 → 6 million in 2025). Google GeminiThe number of developers increased fivefold to 7 million within a year.
  • open sourceModels drive popularizationMeta's Llama 3 downloads surpass 50 million. The performance of Alibaba's Qwen 2.5 series models in China is close to...GPT-4o.

  AIUsage, costs, and losses increase = Unprecedented

  • AIUsage surgeChatGPTInteraction volume, specifically daily queries, increased from 1 million in November 2022 to 1 billion in April 2025. (Bank of America)AIErica, the assistant, has processed over 2 billion interactions.
  • Differentiation between training and inference costs:
    • Training costs:GPT-4 training costs exceed $100 million, with an average annual growth rate of 200% (2019-2025).
    • Reasoning costThe cost of processing a single token starts from...GPT-3 decreased by 0.01GPT-4o of 0.001.
    • TOpenAILossThe company incurred a loss of $5.3 billion in 2024 (P.173), primarily due to investments in computing power and model training.
    • Industry Trends:worldwideAIThe average loss rate for startups is expected to rise from 40% in 2021 to 65% in 2025.
    • medical fieldFDA approvedAIThe average cost of deploying medical equipment increased by 120% (2020-2025).
    • Financial sectorJP Morgan'sAIThe annual maintenance cost of the risk control system reaches $250 million.

    AIThreat of monetization = Increased competition + open sourceMomentum + China's Rise

    • Competitive landscape, the United StatesLLMDominance challenged:USALLM(likeGPT-4.5) accounts for 62% of the global desktop user share, but in China, Qwen 2.5 and other similar models...
    • open sourceModel Impact on Business Ecosystem:
      • Key events:
        • Alibaba released 100open sourceThe Qwen 2.5 model boasts performance close to that of its Western competitors.
        • DeepSeek launches R1-Zeroopen sourceReasoning model.
        • Industrial robotsChina accounts for more than 50% of the global deployment.
        • Policy Response:
          • NVIDIA partners with 30+ countries on "sovereignty"AI"plan.
          • China's 14th Five-Year Plan willAIThe target for chip self-sufficiency is set at 70%.
          • Microsoft has integrated Copilot into the 365 suite, enhancing the subscription model.
          • Google GeminiThe developer ecosystem has grown fivefold to 7 million within a year.
          • Business risks:open sourceModel leads toLLMProfit margins are expected to fall from 45% in 2023 to 28% in 2025.
          • Geopolitical risksThe US restrictions on high-end GPU exports to China are accelerating China's independent chip development.

          AIIntegration with the physical world = fast + Data-driven

          • Industrial robot deployment acceleratesChina accounts for more than 50% of the world's industrial robot installations.
          • automaticBreakthrough in driving technologyWaymo's progress: by 2025, driverless taxis will cover 90% of the San Francisco area.
          • Medical robot applications:
            • Surgical robotsThe da Vinci system has performed over 10 million surgeries worldwide.
            • FDA approvalNearly 700 items in 2023AIOf the medical devices, 23% are robot-assisted.
            • Traffic optimizationSingapore usesAIReduce congestion by 30%.
            • Energy NetworkGoogle's DeepMind optimizes power grid efficiency by 15%.
            • Agricultural dronesXAG, a Chinese drone manufacturer, has covered more than 100 million mu of farmland.
            • Logistics robotsAmazon has deployed over 500,000 Kiva robots in its warehouses.

            Global internet user growth is driven byAIDriven by unprecedented growth

            • AIGlobal user growth curve driven byChatGPTThis will drive global internet user growth to an all-time high, with the proportion of non-North American users jumping from 35% in 2023 to 52% in 2025.
            • Emerging marketsAIuniversal:
              • CaseIndia, Indonesia, and other regions have passedAITools (such as voice assistants) skipped the PC era and went directly to mobile devices.AIUsage rate increased by 200% annually.
              • dataChat in AfricaGPTThe user share is expected to increase from 3% in 2024 to 12% in 2025.
              • Arizona State UniversityAIThe "Acceleration Program" increased course completion rates by 35%.
              • Indian government passedAIVoice assistants are providing basic education in rural areas, covering 120 million students.

              AIWork change = Real + Rapid

              • AIStructural impact on the job marketIn 2025, among global IT jobs,AIRelated positions accounted for 35%, non-AIThe annual job loss rate is 5%.
              • Skills demand transformationThe survey results show that 73% of companies require their employees to have basic knowledge.AITools (such as Copilot). Demand for programming jobs decreased by 12%, but...AIThe demand for "multi-skilled talents with expertise in various fields" has increased by 200%.
              • Industry Case Studies: Finance and Healthcare:
                • financeGoldman Sachs usedAIAnalysts completed 60% of financial statement analysis, increasing their efficiency by 3 times.
                • MedicalKaiser PermanenteAIMedical record assistants save doctors 50% of their paperwork time.
                • Government ActionThe U.S. Department of Labor will launch a "2024" program.AI"Workforce Retraining Fund".
                • Educational InnovationArizona State University opens "AIThe "Acceleration Program" increased course completion rates by 35%. Developed in partnership with businesses by Oxford University.AIEthics training module.
                • Short term (2025-2030)50% of repetitive tasks (such as data entry and basic programming) will be handled by [the relevant personnel].AItake over.
                • Long-term (2030+)"Human-machine collaboration" has become mainstream, for exampleAIManaging project processes involves humans making creative decisions.

                Get Mary Meeker's《Trends – Artificial Intelligence》  Original PDF fileScan the QR code to follow and reply:20250601

                Andrew Ng's team released "MCP: Building Rich Contexts with Anthropic". AI application"

                2024 China Marketing IndustryAIApplication Development Research Report (PDF file) - AITutorial materials