Core Insights - The article reflects on the predictions made for the AI industry in 2025, noting that most judgments about industry dynamics and technological paths have proven accurate, although there was an overestimation of technological advancements and infrastructure maturity [2] - The emergence of positive signals such as World Model, multimodal capabilities, and robotics indicates that the AI field will continue to surprise, but high expectations have been priced in, leading to increasing market anticipation [2] Group 1: OpenAI and Microsoft Dynamics - In 2025, OpenAI transitioned to a profitable organization, and Microsoft invested in Anthropic, altering the landscape of models and cloud services [6] - Microsoft built an internal LLM team through the acquisition of Inflection AI and ended its exclusive relationship with OpenAI, leading to a multi-cloud model where all models are supported across various cloud platforms [7] Group 2: Google's Positioning - Google, initially seen as lagging in LLM training, has become the "most advanced follower" with significant resources, including TPU and distribution channels, allowing it to regain its competitive edge [8] - The launch of Gemini 3 in Q4 2025 marked a significant comeback for Google, sparking discussions about AI competition and demonstrating its advantages in AI infrastructure and talent [9] Group 3: Agent and OS Development - The competition among model vendors resembles the historical Windows/DOS battle, focusing on developer mindshare and ecosystem control, with Anthropic showing a strong commitment to building an OS [10] - The trend of transforming chatbots into advanced agents capable of complex tasks is evident, with significant investments in OS-level capabilities [11] Group 4: Coding Agents and Automation - The rise of coding agents, exemplified by Claude Code, signifies a shift in AI's role from simple assistance to generating and modifying entire projects, with substantial growth in ARR [13] - The focus on task automation highlights the importance of long-horizon task success rates as a measure of agent capabilities, with agents evolving to handle more complex tasks [17] Group 5: Context Layer and Infrastructure - The context layer is identified as a critical infrastructure capability for agents, with companies like Palantir benefiting from context engineering to enhance agent performance [22][23] - The demand for context-driven solutions is driving competition among AI and data companies, emphasizing the need for effective context layer construction [22] Group 6: Hardware and Inference Trends - The shift in focus from pre-training to reinforcement learning (RL) scaling indicates a significant change in the AI training paradigm, with post-training becoming equally important [26][27] - NVIDIA maintains its leadership in the computing market, with its market cap surpassing $5 trillion, while other companies like AMD are struggling to keep pace [25] Group 7: M&A Activity and Market Dynamics - The AI sector is experiencing active M&A activity, with larger companies acquiring AI-native applications and smaller firms, driven by the need to stay competitive [46][47] - The trend of "acqui-hire" is emerging as a strategy to quickly build high-level teams in response to the AI arms race [49][50] Group 8: Energy and Nuclear Power - The ongoing energy crisis is leading to a resurgence in nuclear power, with companies benefiting from stable power sources seeing significant valuation increases [51][52] - The demand for reliable energy sources is becoming a critical asset in the AI infrastructure landscape [52] Group 9: AI in Scientific Research - The rapid development of AI in scientific fields is leading to the emergence of specialized foundation models across various disciplines, with significant advancements expected [54][55] Group 10: Market Performance and Predictions - The U.S. stock market experienced fluctuations in 2025, with a notable recovery driven by AI investments, particularly in SaaS companies [61][63] - The narrative around AI is shifting from hype to a focus on practical applications and profitability, with companies needing to demonstrate real-world value [63]