下一个十年胜出的 不会是算力公司、模型公司和应用工具公司 而是那些掌握了“?上下文

This article argues that over the next decade, the winners in the AI industry will not be computing power providers, model developers or AI application tooling companies, but rather companies that master "contextual intelligence". Its key takeaways are as follows:

1. Current state of the AI industry: Gartner projects that by 2030, the cost-efficiency of large language models (LLMs) will increase 100x from 2022 levels, and inference costs for trillion-parameter models will drop by 90%, leaving very little margin for profit from token usage. Barriers to developing all types of AI tools have fallen steadily, but few players have managed to achieve scalable profitability. While general-purpose large models continue to post better benchmark scores, enterprise adoption has an extremely high failure rate: 95% of AI pilot projects and 88% of proof-of-concepts never reach production in overseas markets, with even higher failure rates in China.

2. Core logic and common pain points: AI project performance equals model capability multiplied by contextual intelligence. Even the most powerful model cannot generate business value without context tailored to a company's specific operations. Most AI adoption failures are rooted in context gaps, not flawed inference. Common problems include disconnected identities across systems, inconsistent business semantics, and outdated institutional knowledge—issues that cannot be fixed simply by upgrading models.

The AI adoption framework laid out in this article offers valuable insights for brands looking to build AI capabilities and secure competitive advantages in the AI era. Key takeaways are as follows:

1. Common pitfalls for brands adopting AI: Many brands currently invest heavily in cutting-edge large models and expanded computing power, but overlook fragmented internal data and inconsistent business rules across their organization. As a result, the vast majority of these AI projects end in failure. Even failure rates exceed 90% for mature overseas companies, and domestic Chinese brands face even higher odds of failure.

2. Value of building a contextual layer for brands: A contextual layer solves three core problems that derail AI applications for brands: unifying user identities across systems, standardizing business semantic definitions across departments, and syncing the latest internal rules and organizational knowledge in real time. This prevents AI from generating incorrect or useless outputs and delivers reliable results for business use.

3. Future AI competitive dynamics: In the AI era, a brand's sustainable competitive moat will not be built on large model capabilities, but on its own business-aligned contextual system. Building out a contextual layer early is the only way to ensure AI generates tangible business value.

This article outlines clear industry trends, risks and opportunities for sellers positioning themselves in the AI track. Key takeaways are as follows:

1. Risk outlook: Over the next decade, large model inference costs will fall by 90%, leaving very limited profit opportunity for businesses that rely on selling computing power or token access. Standalone AI application tools struggle to achieve scalable profitability. And companies that focus solely on general-purpose large models face extremely high enterprise adoption failure rates, making it hard to build sustained, stable business value.

2. Opportunity outlook: As large model technology advances rapidly, inference capability has already been solved at scale by a small number of leading research labs. The largest unmet business need today is building contextual layers tailored to enterprise operations, which will be the core growth opportunity in AI over the coming decade.

3. Proven successful framework to learn from: Profitable AI firm Palantir's success is rooted in its mature contextual engineering system. Sellers can follow this model by focusing on products and services that address enterprises' contextual needs for AI adoption, greatly increasing their odds of success.

The core argument on AI adoption in this article offers important insights for factories advancing digital and AI transformation. Key takeaways are as follows:

1. Common pitfalls in factory AI transformation: Many factories blindly invest heavily in state-of-the-art large models and upgraded computing hardware when pursuing AI transformation, but neglect to integrate and govern fragmented data across production, design and customer-facing operations. As a result, most AI projects fail to deliver tangible value, with even higher failure rates than those seen in mature overseas enterprises.

2. Value for production and product design: Building a business-aligned contextual layer integrates fragmented data across a factory's disparate systems, standardizes business semantics, and syncs the latest production and design rules. This eliminates AI output errors caused by inconsistent data, ensuring AI is properly tailored to a factory's production and design requirements.

3. Competitive opportunity: The article makes clear that companies that master contextual intelligence will win over the next decade. By building out their own contextual systems early, factories can establish differentiated competitive moats in the AI era and secure long-term competitive advantages.

This article clarifies industry trends, core customer pain points and future direction for AI service providers. Key takeaways are as follows:

1. Industry growth trends: Large model inference capability will eventually be solved at scale by a small number of leading research labs. Service providers that only focus on large model R&D, computing power services or general AI tool development will struggle to generate sustained long-term profits. Helping enterprises build and maintain contextual layers will become the largest growing demand in the AI service market.

2. Core customer pain points: The biggest barrier to successful enterprise AI adoption today is rooted in lack of context, which manifests in three common problems: disconnected customer identities across systems, inconsistent business semantics across departments, and outdated internal rules and knowledge. These issues are entirely unrelated to model capability, and cannot be fixed by upgrading to larger models.

3. Strategic direction for solutions: Service providers can shift their core focus to contextual layer services, building end-to-end architectural solutions that integrate data unification, semantic standardization, permission management and real-time knowledge updating. By solving the core pain point of difficult enterprise AI adoption, providers can capture massive new growth opportunities.

This article maps out the core demand for enterprise AI adoption and clarifies direction for AI platform strategy, positioning and risk mitigation. Key takeaways are as follows:

1. Core enterprise demand from AI platforms: Most companies today see their AI projects fail even after adopting large models. Their core need is not more powerful models or more computing power, but rather tools and services from the platform to help integrate internal data and build a unified contextual system to solve common adoption barriers.

2. Platform positioning strategy: AI platforms can adjust their service architecture to position the contextual layer as the platform's core infrastructure, providing unified contextual capability support for on-platform AI service providers and enterprise customers alike. This solves common pain points including cross-system identity unification, semantic standardization and real-time knowledge updating, improving overall AI adoption success rates across the platform.

3. Risk mitigation: Platforms should not blindly bet on general-purpose large models with more computing power and better benchmark scores. It is critical to recognize that model capability alone cannot meet enterprise demand for successful AI adoption. Investing early in contextual layer development helps avoid strategic misalignment and captures growth opportunities over the next decade.

This article puts forward a novel assessment of AI industry development, offering valuable reference for research on new industry trends and emerging business models in AI. Key takeaways are as follows:

1. New industry development dynamics: Competition in the AI industry has long centered on model inference capability. As large model technology iterates at an extraordinary pace, inference capability has gradually been solved at scale by a small number of leading research labs. Industry competition is now shifting from model inference capability to the ability to build contextual intelligence, representing a complete shift in competitive logic.

2. New unaddressed industry problems: The industry has widely neglected investment in contextual layers, and missing context creates cascading failures: a 10-step workflow with 85% success per step delivers an overall success rate of just 20%. The well-known problem of AI hallucination is essentially model "guessing" caused by missing context, making this the core new problem the AI industry must solve.

3. Implications for business model research: The winning AI business models of the future will not be cloud computing rental, general model R&D or general AI tool development. Instead, the largest opportunity lies in services for building and maintaining enterprise contextual layers. This model has already been proven viable by Palantir, and leading AI researcher Yann LeCun has already invested $1 billion in this space, making it a high-priority new direction for in-depth research.