2026年人工智能状况报告

Introduction This report is intended to summarize the advancementsin AI and its media perception in the last 12 months(January 2025 - January 2026): the progress in research,specific AI news in the media and emerging trends(catalysts) that will impact the near future. Ourunderstanding of the AI advancements and the trendsallows us to make trend predictions we believe willhappen in the next year (last 9 slides). We will reviewaccordingly in one year’s time. 2026 Table of Contents 01. Introduction ●26 Catalyst 5: The Crypto-AI Ecosystem●27 Catalyst 6: Supply Chain, Sovereignty & Workforce●28 Catalyst 7: VC Investment Landscape●29 Catalyst 8: Labor Shortages●30 Catalyst 9: Public Spending & Debt Ratios●31 Catalyst 10: Virtualization and automation of Living●32 Catalyst 11: Freelancer (gig) Economy ●03 Author●06 Notes on the Report & Methodology●07 How to Read This Report●08 Review Last Year Catalyst Trends●10 Review Last Year Predictions 02. 2025 Retrospective 06. 2027 Trend Predictions ●08 Review Last Year Catalyst Trends●10 Review Last Year Predictions ●33 Prediction 1: AI Capital, Energy & Economy●35 Prediction 2: AI Hardware & Chips●38 Prediction 3: Governance, Defense & Cybersecurity●41 Prediction 4: Industry AI Adoption●43 Prediction 5: Consumer AI Adoption●45 Prediction 6: Model Advancements (Text & Multimodal)●47 Prediction 7: Agents, Coding & Platforms●49 Prediction 8: Audio, Image & Video Models●51 Prediction 9: Science & Robotics●53 Bonus Prediction: The Rise of "Pynglish" 03. Research 2025 Review ●15 Best AI advancements 04. Secular Trends (The Long View) ●17 Trend 1: Human-Machine Hybridization●18 Trend 2: Global Aging & Demographics●19 Trend 3: Obesity & Health Economics●20 Trend 4: The Rise of Asian Economies●21 Trend 5: World Polarization & Geopolitics 05. Catalyst Trends (The Immediate Signals) 07. Conclusion ●22 Catalyst 1: AI Governance & Regulation●23 Catalyst 2: The New Data Landscape (Human vs. Synthetic)●24 Catalyst 3: Cloud Infrastructure & Data Centers●25 Catalyst 4: Energy Supply & Grid Constraints ●52 Contact & About AI Technologies Author Andrea IsoniChief AI Officer Physicist with a PhD from Imperial College, FormerFounders Factory (FF), Chief AI Officer at AITechnologies. Experienced in leading edge AI algorithmsand how to translate them into business applications(including pioneering Human Machine Interfacesystems). Advisor to Waed Saudi Aramco and AI writer('Machine Learning for the Web' published by Packt inEnglish, Chinese and Korean and more than 10,000copies). Committee member of ISO and IEEE for AIstandards and International AI speaker (London, NewYork, Dubai, Saudi Arabia). AI Newsletter ‘Thoughts aboutAI by a Human’ (3.1K+ subscribers). 2026 2026 Who read the State of AI (audience) The State of AI Report is designed for a diverse audience of strategicdecision-makers, technologists, and investors who require a comprehensive yetaccessible overview of the current AI landscape. Its primary readers include C-suiteexecutives, policy makers, and researchers who rely on the report to make sense ofthe complex web of current ("contingent") events, breakthrough research, andmarket dynamics without getting lost in the weeds. The report analyzes the forcesat play—both the long-standing "secular" trends driving fundamental change andthe emerging "catalysts" sparking immediate disruption—to help leaders anticipatehow AI will shape the near future of society, industries, and consumer adoption.Each scenario presented is governed by a specific logic framework: Secular Trend +Catalyst→Contingent Event + Prediction. Crucially, the analysis is not intended tobe an exhaustive encyclopedia or an overly technical manual; rather, it provides justenough detail to ensure readers understand the mechanics and implications ofeach scenario discussed. Ultimately, the report’s predictions (whether right or not)serve as a guide, giving readers a ballpark understanding of how AI is likely toimpact life and business in the year ahead. For more details on this methodology,see "How to Read the Report" on page 7. Disclaimer: the content of this report is for education/information only, notfinancial, legal, or tax advice. 2026 Notes on the Report * This year’s data is considered until 30 January 2026: so from January2025 to January 2026 included. * AI Leaderboard used: https://huggingface.co/spaces/ArtificialAnalysis, https://lmarena.ai/?leaderboard,https://opendfm.github.io/MULTI-Benchmark/#leaderboard, https://vlaleaderboard.com/https://artificialanalysis.ai/leaderboards/models, https://www.wolfram.com/llm-benchmarking-project * Definitions (only very technical ones. It is expected the reader knows what GPU means etc.):-pre-training: a process where a model is trained on a large, general dataset to learn general features.-mid-training: additional training with synthetic data, optimizing data mixes, domain-specific data andlong-context training stages. -post-training: set of