The real problem behind the technology boom.
Artificial intelligence is no longer a simple technical trade, but a rewrite of a production function.2025Global Enterprise YearAIInvestments reached5817Billions, up from the previous year130%; privateAIInvestments reached3447$ billion, of which the generatorAIThe number of capital flows is the most central. That means the capital market is no longer in the capital market.AIAs laboratory technology, it is considered as a core variable for changes in business efficiency, software patterns and industrial structure over the next decade.
But now.AI[Ration] It is closer to “early high growth”: technological capacity has been proven, user demand has been validated, capital input has begun to be scaled, but business closures, profit-rate stability and organizational transformation are not yet fully completed.ChatGPTIt proves, “People will use it.”AIThe company is willing to do so.AIPaying, but not fully proven.AIIt is possible to create quantifiable profits steadily in most industries.”
And so is…AIThe most worthwhile thing to study is not the story that’s over, but the one that’s just gone from model display to work flow. The biggest change in the industry in the next five to ten years is not that the model is more human, but rather that the model is more human.AIStart replacing, compressing, and enhancing a large number of white collar and engineering processes.SaaSAccount number, may buy a job in the future.Agent。
Industry phase: high growth, but not risk-free prosperity
FACTMenlo VenturesData show that the enterprise is generatingAIExpenditure in2025Year reached370$ billion, higher than2024Yearly.115Billions, around the same time.3.2More than half of the expenditure goes to the application layer, which means that the enterprise is not simply buying the bottom model, but is looking for software that directly enhances productivity.McKinsey 2025The year of the survey also shows that the government is not in a position to do so.AIIt has been widely used, but many enterprises are still at the pilot, partial deployment and process alignment stages.
[Limentation] That explains it.AIThe industry has moved beyond the “need for demand” phase to the “can you scale the value of delivery?”IT♪ And the system ♪AICompanies now need to prove that they can really embed into business processes. This phase is characterized by rapid income growth, but costs, organizational friction and product reliability remain the main constraints.
The next five to ten years,AIThe core drivers are drawn from four directions: increased reasoning, reduced reasoning costs, re-engineering of business processes, multimodels and robotics. This prediction is premised on continued progress in modelling capacity, no severe disruption in the supply of arithmetic, and no complete restriction in the use of data and models by regulation.AIGrowth will continue, but will be significantly slower than current market expectations.
Business model: excellent, but not evenly distributed
AIThe industry’s business model is simple on the surface: businesses, developers, and consumers pay for model capabilities, software functionality, or automated results. But the real complexity is value distribution.AIApplications and industriesAgentEach layer makes money, but the Maori rate, the moat and capital intensity are different.
FACTNVIDIAThese chip companies benefit from training and reasoning needs, with strong supply constraints and ecological barriers; cloud manufacturers pass throughAzure、AWS、Google Cloud(b) Carrying out the capacity requirements;OpenAI、Anthropic、Google DeepMindWait for model companies to sell.API, enterprise subscriptions and platform capacity; the application level is usingAIPackaging is included in specific scenarios such as programming, law, customer service, design, financial analysis and office collaboration.
[Limentary] Long-term profits are most likely to come from two types of companies: the first is those that control critical infrastructure, for example.GPUClouds and data centres; second type is embedded in enterprise processes in depthAIApplying companies. A mere “smarter model” does not necessarily constitute a permanent moat, because model capacity may converge, prices may fall, and open-source models will continue to depress the premium on basic capacity.
The biggest controversy in business patterns is thatAIIt’s gonna be stronger.SaaSOr destroy it?SaaSOne view is thatAIIt’s gonna make the traditional software go by.AgentReplaces because the user no longer needs to open more than a dozen tools, just letAIThe other view is thatAIIt increases the value of the software because the current software has access to customers, data and processes. My judgment is that both are likely to happen: weak.SaaSIt’s gonna be compressed, strong.SaaSYes.AI♪ I’m gonna be a little bit more ♪AIThe raw software grows in a vertical setting.
Market space: not an industry, but a new infrastructure
AI”The market space of the world can’t be just about looking.”AIsoftware income. If you only see model subscriptions andAPIIncome, it underestimates its impact; if all of it is taken,AIThe remodeled industries are counted and they can overestimate short-term income.Stanford AI Index 2026Show, Global EnterprisesAII’m investing in2025It’s over.5800$ million, notesAICapital spending, software procurement, talent recruitment and infrastructure investment have developed a systematic trend.
[Licensation]AIIt’s…TAMIt should be seen in three layers. The first level is infrastructure, including:GPU , data centres, electricity, cloud services and networks; second level is software, includingAIOffice,AIProgramming,AI- The guest service.AIMarketing,AILaws,AIFinancial analysis; the third level is the real economy, including robotics, autopilot, manufacturing, medical care, energy and logistics. The first level is the first to pay, the second level determines the profit margin, and the third level decides.AIReal access to long-term markets at the trillion-dollar level.
The next 10 years,AIThe biggest increase is not just talking robots, but in companies.AgentandPhysical AI- Business.AgentThe blog is a blog of the blog “The White-Led” (“The White-Little”) which is a blog of the blog.Physical AI♪ Will be ♪AIThe assumption is that you can get it from the screen to the factory, warehouse, hospital, and home.AISystems are still becoming more reliable, robotic hardware costs are falling, and businesses are willing to adapt processes ifAgentLong and unstable operations, or slower than expected progress in robotic hardware,AIMarket space remains large, but the path to realization will be more software-oriented than the real economy.
Competition patterns: Models are not the only battlefields
Current globalAIThe pattern of competition can be divided into four groups. The first group is the infrastructure winner, representing theNVIDIA、AMD、BroadcomThe second group is cloud and platform companies, which are represented byMicrosoft、Amazon、GoogleThe third group is the base model company, which is represented byOpenAI、Anthropic、Google DeepMind、Meta、xAIThe fourth group is the application level company, includingCursor、Perplexity、Harvey、Glean、ServiceNow、SalesforceWait.
The most secure winner of the [extraction] is still at the infrastructure level, as all models and applications require arithmetic.NVIDIAThe moat is more than just a moat.GPUBut…CUDAThe advantage of cloud producers is the company’s customer entrance and computing resources. Model companies’ advantage comes from technology, brand, and developers’ ecology, but competition at the model level can be brutal, because training costs, prices fall fast, and catch-up persists.
What is really uncertain is the application layer. The application layer could be the largest new company, because it is the closest to the client’s budget and the easiest to capture.ROI. But the application layer is also most easily copied by the big platform.AIApplications are weak if they are simply “talk boxed on old software”, the moat river, and if they can rewrite a high-value process, such as contract review, code generation, sales follow-up, insurance claims, medical documentation or financial research, it has the opportunity to become a permanent company.
No more than 10 key indicators
- FACT2025Global Enterprise YearAIInvestment contract5817Billions of dollars, year-on-year growth130%。
- FACT2025PersonalAIInvestment contract3447Billions of dollars, year-on-year growth127.5%。
- Business GenerationAIExpenditure2025Year370 million2024Year3.2Double up.
- [Factual] BusinessAIThe proportion of expenditure in the application layer is more than half, indicating a shift in demand from models to productivity tools.
- [Factual] Newly Financed GlobalAIThe number of companies is2025Annual growth approximate71%。
- ♪ The United States still is ♪AIInvestment Centre,2025PersonalAIInvestments were significantly ahead of China.
- FACTMcKinseyShowAIUse has spread widely, but the scale of the decline and the financial returns remain uneven.
- [Limentation] Arithmetic, power and data centresCAPEX♪ Becoming ♪AIGrowth is hard bound.
- BusinessAIProcurement is moving from “trying” to “Is it possible to bring clarity?ROI”。
- The most important new demand in the next decade will come from the following sources:Agent 、AINative software and robotics.
Five military judgments.
First,AIIt’s nearer the eruption, but not all.AIThe company is about to explode. Capital, business procurement and user use have grown at the same time; the real outbreak is not “talk,” but ratherAIEntering the business process. The test is not a model release, but a willingness to pay for savings, efficiency gains and shorter delivery cycles on a continuous basis.
Second,AIThe reason for this is not a single application that explodes, but that it can cross-cuttingly access almost all knowledge and parts of the physical work. Offices, programming, customer services, education, law, medical care, manufacturing, finance are all affected.
The third is thatAIIt’s easy to replicate expansion, but it’s not easy to copy high-quality delivery.APIAnd cloud services lower the threshold for entrepreneurship, so low endAIThe apps will be highly homogeneous. The real difficulty is to put it in the right place.AIAnd as a complex process, it’s stable, manageable, compliant, auditable. The future moat is not “I’ve also taken a big model,” but “I can continue to produce results in real business.”
The fourth is that:AI (a) To meet real long-term needs. [Rictation] Businesses have long pursued two things: lower costs and higher incomes.AIIt can reduce duplication of effort and improve marketing, research and development, customer service and operational efficiency. This is not a short-term entertainment demand, but a long-term demand for productivity.
The fifth one.AIIt’s for long-term input, but it’s not for blind input.PromptSkills, which will soon be phased out by tools; if a company packs only model interfaces, they can be easily replicated.AISystem design,AgentWorkstream, industryKnow-howdata engineering,AIInfrastructure, robotics andAISecurity governance.
Maximum opportunity and maximum risk
One of the greatest opportunities is the enterprise.AgentIt is not about answering questions, but about fulfilling its tasks.HRThere are a large number of standardized but judgemental processes available for operation, data analysis.AgentTo be able to do these processes in a stable manner, the company will be able toAIExpenditure was shifted from the innovation budget to the core operating budget.
The second chance is…AIOriginal software. Traditional software is the human center.AIThe original software is mission-centred. Future users may not click on buttons on each other, but rather describe the target, allowing the system to mobilize data, tools and processes to complete its task.
The third opportunity is robotics and real economies.AINumberAIThe solution is the flow of information.Physical AIThe solution is labor, movement and operation in the physical world.AIThe market space for access to warehousing, manufacturing, medical care, home services and energy systems would be much greater than that of pure software.
The risk is also obvious. First, cost risks, computing, electricity, data centres, and reasoning costs may shrink profits. Second, competitive risks, with model capabilities converge, price wars may make it difficult for many model companies to obtain excess profits. Third, regulatory and security risks, copyright, data privacy, model illusions, automated decision-making responsibilities and national security reviews may limit.AIDeployment speed.
Comprehensive diagnosis of institutional investors
From the institutional investor’s point of view, artificial intelligence is a long-term concern, but the investment backbone cannot remain “who has the strongest models”. Model capabilities are of course important, but long-term returns usually come from the most difficult place in the chain of industry.AIIndustry is the most like the early combination of cloud computing and mobile Internet: infrastructure first makes money, platforms redistribute traffic and the application layer eventually gives birth to a new giant.
I’m most concerned with three types of companies. The first is infrastructure companies with a capacity for computing, cloud, chip and data centre resources, because they are.AIThe second is that we can put theAISoftware companies embedded in enterprise core processes as they are closest to budget andROIThe third category is robotics and industry.AIThe company, because they decidedAICan you get into the real economy from the digital world? Instead, I’ll be cautious about applications that are not distributed, data is not available, process depth is not available, and are just shelled on generic models.
For entrepreneurship, the best course is not to create a universal chat tool, but to choose a vertical scene with high frequency, high value, complex processes, and verifiable results. Legal contracts, insurance claims, medical instruments, sales operations, financial auditing, code migration, game content production and industrial transport.
For career choices, the job that is most worthy of entry is not simply a job.Prompt EngineerBut…AIProduct managers,AgentSystems designer,AIApplication Engineer, Data Infrastructure Engineer,AISecurity and governance, robot software engineers, industryAIThe solution architect. The common thing about these jobs is that they’re not just “will use.”AIIt is about understanding operations, systems, data, reliability and business results.
If I can only choose one direction for long-term investment in the next decade, I’ll choose artificial intelligence, but not general.AI. I’ll choose “business.”Agent + AIOriginal software +Physical AIThe reason is that the line is linked to technological advances, enterprise budgets and real productivity gains.AIThe biggest change is not that machines talk, but that software starts to move from tools to labour; once this is established, the cost structure, organizational structure and job structure of the entire business world will be re-pricing.