张小珺Jùn｜商业访谈录

Investors should prioritize emerging AI Agent Platforms and Ecosystem Gateways, which offer the highest-upside growth potential over the next 2 to 5 years as the market shifts from infrastructure toward agent discovery, trust, and execution layers.

Maintain defensive allocations in established cloud infrastructure leaders like Alphabet (GOOGL), Microsoft (MSFT), and Alibaba (BABA) for reliable cash flows, but remain alert to long-term disruption risks facing their core search, enterprise software, and e-commerce interfaces.

Exercise caution with private foundation model providers like OpenAI, Anthropic, and xAI, as baseline intelligence will likely commoditize into a regulated utility with compressed profit margins over time.

Limit exposure to legacy consumer platform incumbents like Meta Platforms (META) and Tencent Holdings (TCEHY), which face structural headwinds as digital interaction evolves from traditional social networks to agent-driven workflows.

Keep Embodied AI and Humanoid Robotics on a long-term watchlist rather than actively buying now, delaying major capital deployment until hardware production scales down in cost and standardized operational models mature.

Sustained multi-trillion parameter model scaling reinforces strong demand for NVIDIA Corporation (NVDA) and its next-generation Blackwell architecture, solidifying the chipmaker's long-term competitive moat. Investors should also look at Xiaomi Corporation (1810.HK), which is well-positioned to commercialize consumer-facing AI by achieving ultra-fast inference speeds across its vast smartphone and smart device ecosystem. Downstream AI software builders present an attractive margin-expansion opportunity as low-cost structural innovations from DeepSeek continue to drive API pricing lower. Meanwhile, monitor the progress of frontier models like Moonshot AI (Kimi), as their technical breakthroughs serve as a key leading indicator for high-margin Enterprise AI adoption.

Prioritize AI hardware infrastructure like NVIDIA (NVDA), as persistent compute bottlenecks and massive model-training demand make high-end accelerators the highest-conviction core holding.

Capture near-term cash flow through AI Application Companies, but actively manage platform risk as major model developers increasingly integrate specialized features natively.

Approach pure-play AI Foundation Model Companies with caution due to heavy capital burn and rapid price erosion driven by capable open-source alternatives like DeepSeek.

Begin building long-term growth allocations in Embodied AI and Humanoid Robotics supply chains—particularly sensor and actuator manufacturers—positioning for mainstream physical deployment on a 5- to 10-year investment horizon.

Capitalize on the artificial intelligence and robotics mega-trend by investing in NVIDIA (NVDA), which serves as the essential infrastructure backbone for the entire sector.

Instead of betting on individual robot manufacturers, use NVDA as a core ecosystem play to gain diversified exposure to the emerging Physical AI market.

Expect management's long-term marathon mindset and continuous scaling of foundational models like Cosmos 3 to drive sustained, multi-year demand for the company's hardware.

Understand that NVDA actively reduces adoption friction by open-sourcing its technology, cementing its proprietary software and GPU ecosystem as the undisputed industry standard.

Maintain awareness of the heavy capital expenditures required for next-generation world models, but view any market dips driven by these costs as attractive long-term accumulation opportunities.

Investors should pivot from basic AI application wrappers to emerging leaders in Auto Research and foundational automated model design. Watch for early-stage infrastructure providers that enable natural language "vibe training" and automated custom model generation. Monitor high-potential "New Labs" originating from elite academic institutions like MIT and Stanford that are bridging deep research with commercialization. Allocate capital toward specialized verticals like AI for Science and World Models over the next 6 to 12 months as these R&D phases transition to products. Prepare for a sector-wide shift where automated, data-driven reasoning frameworks lower the costs of training private, domain-specific models.

Investors should monitor foundational open-source AI infrastructure projects like vLLM, which act as critical ecosystem gatekeepers across various hardware architectures like Nvidia and AMD.

The commercialization of vLLM through its newly formed entity, Infraact, mirrors the enterprise success of companies like Red Hat and Databricks, making it a key trend to watch for institutional AI backing.

Investors focusing on the AI deployment market should prioritize firms that successfully co-design their models with hardware constraints, as this dramatically reduces cost-per-token metrics.

Look for opportunities tied to high-performance open-source model ecosystems, particularly those inspired by the explosive global adoption of DeepSeek-R1.

The most actionable trade from these insights is the robot “brain” software layer, where value is shifting away from hardware makers. Private leader Ant Lingbo is betting on a pure robot-native foundation model, with its next-generation release expected in 2026 to power over 20 robot types—watch for IPO or partnership news as a catalyst. Public market exposure can be gained through robotics and AI-themed ETFs (e.g., BOTZ, ROBO) or by owning sensor and data-supply-chain leaders that benefit from the race to scale real-world robot data. The data bottleneck means companies with cost-effective, high-quality robot data collection capabilities could outperform as data volumes reach the hundred-thousand-hour milestone. Monitor Chinese robotics component stocks for near-term momentum, as China’s hardware and data collection advantages may accelerate the local brain model ecosystem.

Investors should prioritize companies focusing on robotic manipulation (hand-eye coordination) over locomotion, as solving complex tasks like assembly and folding offers faster commercial viability than walking. NVIDIA (NVDA) remains the essential infrastructure play, providing the massive compute and "World Models" required for the industry's shift toward data-driven, end-to-end learning. While Tesla (TSLA) leads in humanoid ambition, the high manufacturing efficiency of Chinese hardware firms like Unitree makes them dominant players in the global robotics supply chain. Look for high-margin opportunities in "General Purpose Brain" software (VLA models) rather than hardware, which is rapidly becoming a commoditized sector. To mitigate safety and regulatory risks, favor developers of lightweight or "soft" robots over heavy industrial humanoids for the massive home-service market.

Investors should prepare for a potential SpaceX IPO in 2024 or 2025, as the company transitions from a launch provider to a dominant data and infrastructure platform via Starlink. Look for entry points into private markets or secondary funds targeting xAI, which is positioned to lead in "Physical AI" and "World Models" with a major breakout expected by late 2025. High-conviction opportunities exist in "Downstream" applications, specifically companies focusing on Microgravity Manufacturing for pharmaceuticals and high-end materials that cannot be produced on Earth. Monitor the emerging Space Data Center sector, which leverages infinite solar energy and the vacuum of space to bypass Earth's power and land constraints for AI compute. Finally, diversify into the "SpaceX Mafia" ecosystem by backing U.S. hardware startups applying first-principles manufacturing to Mini Nuclear Power and Advanced Battery technology.

Investors should consider Anker Innovations (300866.SZ) as it transitions from simple power banks to high-margin "Hard Mode" categories like AI-integrated security and high-end audio. The company’s "Third Type" strategy of managing 40–60 medium-sized product categories provides a diversified revenue stream that mitigates the risk of single-product obsolescence. Look for growth in their 7-Series flagship tier and eufy security brand, which are leveraging Edge AI and Vision-Language-Action (VLA) models to create proactive, local data processing moats. Anker is a primary play for those seeking exposure to Embodied AI and household robotics, as the company pivots toward high-end "companion" technology. Monitor the firm's ability to maintain its Amazon dominance while expanding its premium "5-Series" footprint in the Chinese domestic market.

Investors should monitor XPeng Motors (XPEV) as it pivots from a hardware manufacturer to an AI-driven group, aiming for a software-heavy value proposition that could significantly expand long-term margins. A key catalyst to watch is the rollout of Level 4 autonomous driving capabilities, which the company expects to achieve within the next 18 months through its new integrated AI architecture. The upcoming GX flagship SUV serves as a critical near-term litmus test for whether this technical "gamble" and organizational restructuring can produce a superior consumer product. For long-term growth, XPeng is positioning its Iron humanoid robot as a secondary valuation pillar, leveraging the same "Physical AI" brain used in its vehicles. Given the CEO's prediction that only five major automotive groups will survive the AI transition in China, investors should prioritize XPEV for its massive annual 1 billion RMB investment in data infrastructure and scale.

Investors should prioritize Anthropic over OpenAI for long-term exposure, as Claude 3.5 Sonnet is currently winning the critical developer "mindshare" and building a superior product environment. Look for entry points into the "Agentic" shift by backing Horizontal AI tools like Cursor or Perplexity, which focus on user workflow and "harnessing" models rather than just raw processing power. Monitor infrastructure plays that manage AI memory and execution, such as E2B, as these "operating systems" are better positioned to capture value than niche vertical software. Avoid companies that merely "burn tokens" as middlemen; instead, seek out platforms like OpenCrawl that own user context and local data to create high switching costs. While current valuations for top-tier AI labs are high, the real "boiling point" for corporate revenue growth is projected for 2026, making this a critical accumulation phase for patient investors.

Investors should prioritize NVIDIA (NVDA) as big tech capital expenditure continues to be revised upward, providing high revenue visibility through long-term chip and power supply agreements. Be cautious with traditional SaaS and software companies that rely on per-seat subscriptions, as these "human-centric" models are being disrupted by usage-based AI "intelligence" platforms. Keep a close watch on Anthropic and OpenAI for potential mega-IPOs, which may trigger a significant rotation of liquidity out of existing Magnificent 7 stocks. Meta (META) offers potential margin growth through internal AI efficiency, but investors should monitor for "waste-driven pullbacks" if their massive compute spending doesn't yield immediate revenue. Look for emerging opportunities in Agentic Infrastructure and "New Labs" that focus on coding agents and "dollar-per-outcome" pricing rather than simple raw data processing.

Investors should prioritize Google (GOOGL) as it leverages its proprietary TPU hardware and vertical integration to secure a long-term infrastructure moat over competitors. Anthropic remains the "gold standard" for AI coding; while private, its dominance in this sector suggests Microsoft (MSFT) and GitHub are the primary public plays to capture the massive productivity gains in software engineering. The industry is shifting from simple chatbots to "agentic" models that manage entire projects, making companies focused on AI Coding Tools the most immediate source of high ROI. While NVIDIA (NVDA) maintains hardware leadership, watch for a transition toward "inference-time scaling," which favors companies that can optimize how models "think" rather than just those with the most raw compute. For exposure to consumer-facing AI innovation, monitor the Chinese AI landscape, as they are currently outperforming Western labs in user engagement and application design.

Investors should prioritize Big Tech leaders like Microsoft (MSFT), Alphabet (GOOGL), and NVIDIA (NVDA) as they provide the essential "frontier models" and hardware for the shift from AI tools to autonomous "AI Labor." Focus on Anthropic (via Amazon or Google partnerships) for its leadership in "Computer Use" technology, which allows agents to automate legacy software without complex integrations. Look for high-conviction opportunities in Vertical AI startups that solve "hard problems" in legal, medical, or coding sectors, as these specialized "Expert Agents" maintain a stronger competitive moat than general chat models. Monitor the 2025-2026 timeframe for the commercial rollout of "Self-Learning" agents, which is expected to significantly optimize enterprise business processes and high-end knowledge work. Be cautious of "simple" AI startups, as general models like GPT-4o and Claude 3.5 are rapidly absorbing basic features and displacing under-capitalized competitors.

The next 2–3 months represent a critical window for the transition from simple chatbots to AI Agents, with value shifting toward framework layers like OpenClaw and Llama. Investors should prioritize companies utilizing efficient architectures like MTP and MIA (e.g., DeepSeek and Xiaomi), as these "cost-performance" leaders are narrowing the gap with US models to just 2–3 months. Expect a 10x surge in demand for inference-specific hardware as persistent agents begin scanning screens and performing long-term tasks. The most reliable ROI over the next 24 months will likely come from AI for Software Engineering, as coding provides the most stable environment for AI self-evolution. Avoid AI applications in sectors with "messy" feedback loops, such as certain quantitative finance models, where the lack of clear rewards hinders model training.

Investors should monitor Google (GOOGL) and Amazon (AMZN) as they lead the transition from "probabilistic" AI to "deterministic" reasoning through projects like AlphaProof and AWS automated reasoning teams. High-conviction opportunities are emerging in Vertical AI startups that focus on Formal Verification and the Lean programming language to eliminate AI hallucinations in software and chip design. While Axiom is currently a private unicorn, its "SpaceX-like" risk profile suggests a binary investment outcome: either achieving super-intelligence in mathematics or total failure. For 2025-2026, look for specialized labs that prioritize Reinforcement Learning and Post-training over expensive pre-training, as these models offer a more capital-efficient path to market. The most actionable shift for the next two years is toward Verified Code Generation, where AI provides mathematically proven, bug-free code rather than simple suggestions.

Investors should prioritize Anthropic (private) as the current leader in high-value coding agents, with the company projected to reach over $2 billion in ARR next year. While OpenAI (private) temporarily lost focus, watch for the release of their "Sparrow" model as the critical catalyst for them to reclaim the lead in agentic workflows. For a stable, long-term play, Google (GOOGL) remains a high-conviction "safe bet" due to its vertical integration with proprietary TPU hardware and massive distribution via Workspace. High-growth specialized startups like Manus (agents), Eleven Labs (voice), and Perplexity (search) offer significant upside as they capture specific niches of the AI economy. A balanced AGI Portfolio should allocate 20% to top-tier models like the "Big Three," 10% to agent infrastructure, and 10% to AI-driven science.

Investors should prioritize exposure to OpenAI and Meta, as their massive data moats position them to dominate the transition from AI "tools" to "personal agents" that act as digital proxies. Look for opportunities in the emerging AI-native social networking sector, which is disrupting a 15-year period of stagnation by using "context-driven" models to automate social friction and networking. High-conviction value lies in major IP holders like Disney and Nintendo, who can leverage deep character lore to create high-margin, interactive "digital souls" for fans. Monitor the Shenzhen tech hub for early-stage startups like Natural Selection, which are utilizing local AI talent to build "cyber-doppelgangers" that manage personal branding and information filtering. Be cautious of "pixel-level" copycats and prioritize companies whose competitive advantage is built on proprietary user data and "context flow" rather than just interface design.

Investors should prioritize NVIDIA (NVDA) as it cements its role as the essential infrastructure provider for autonomous driving and robotics through its Orin chips and Omniverse simulation platform. Meta Platforms (META) offers a unique data-collection advantage via its Ray-Ban smart glasses, which serve as a "Trojan Horse" to capture the first-person physical world data necessary for training future AI agents. While Tesla (TSLA) remains a leader in data scaling through its massive vehicle fleet, the industry is shifting toward "Embodied AI" and Vision-Language-Action (VLA) models, making specialized robotics labs and "Data Infrastructure" firms high-conviction themes. For exposure to the hardware side of this transition, monitor Chinese manufacturers like Xiaomi and Xpeng, which are leveraging their supply chains to lead in the physical deployment of humanoid robotics. Focus on companies that prioritize "AI Education" and high-quality synthetic data, as the ability to simulate and evaluate complex physical tasks is the next major bottleneck for the industry.