> ## Content Index > Fetch the complete content index at: https://www.instituteforstrategicintelligence.com/llms.txt > Use this file to discover other available public pages before exploring further. # AI COMMODITIZES VIRTUALLY EVERYTHING - URL: https://www.instituteforstrategicintelligence.com/ai-commoditizes-virtually-everything/ - Published: 2026-08-29T07:02:11.000Z - Updated: 2026-09-04T03:37:16.000Z - Description: AI is making intelligence cheaper, development faster and competition easier. What happens when the technology creating extraordinary value also destroys the scarcity required to capture it? - Author: Jerry Sutter - Tags: AI Commoditization, Artificial Intelligence **Artificial intelligence may become one of the most valuable technologies ever created. That does not mean the companies creating it will capture anything close to that value.** Something unusual is happening across the artificial-intelligence industry. AI is becoming more capable. But at the same time, access to AI capability is becoming cheaper, faster and more widely available. Models improve. Coding becomes easier. Development cycles shrink. Open models proliferate. AI agents perform work that previously required skilled people. Companies learn from one another. Customers gain more alternatives. And competitors can emerge with astonishing speed. The conventional investment thesis assumes that extraordinary technological capability will produce extraordinary economic returns. But there is another possibility: ## AI MAY CREATE EXTRAORDINARY VALUE WHILE SYSTEMATICALLY DESTROYING THE SCARCITY REQUIRED TO CAPTURE IT. That possibility deserves considerably more attention. Because scarcity is what makes extraordinary margins possible. And AI increasingly creates abundance. --- ## THE COST OF CAPABILITY IS FALLING This is no longer theoretical. According to Stanford University's 2025 AI Index, the cost of using a model performing at roughly GPT-3.5 level fell from $20 per million tokens in November 2022 to just $0.07 by October 2024—a decline of more than 280-fold in approximately 18 months. AI hardware price-performance was improving by roughly 30% annually, while energy efficiency improved about 40% annually. And smaller models were becoming dramatically more capable. In 2022, Stanford found that the smallest model exceeding 60% on the MMLU benchmark contained 540 billion parameters. By 2024, Microsoft's Phi-3-mini crossed the same threshold with just 3.8 billion. That is a 142-fold reduction in model size for comparable performance on that benchmark. The direction is difficult to ignore: **Capability is becoming cheaper.** For most of economic history, sophisticated capability was expensive to reproduce. Building a software company required engineers. Building sophisticated technology required specialized expertise. Designing complex systems required experienced teams. Entering advanced industries often required years of accumulated institutional knowledge. Those requirements created barriers to entry. AI is beginning to attack those barriers. A capable engineer working with increasingly sophisticated coding agents can accomplish work that once required considerably more human labor. A small company can access powerful foundation models without spending billions of dollars training one. Open-weight models allow developers to modify and deploy increasingly capable systems themselves. AI can assist with research, programming, design, marketing, customer service, analysis and increasingly complex business functions. This doesn't eliminate expertise. But it can dramatically increase the amount of capability each expert commands. And that changes the economics of competition. --- # AI CREATES COMPETITORS Consider software. Suppose a company develops an enormously successful AI application. Under the traditional technology model, competitors might require years to reproduce its functionality. They would need engineers. Capital. Infrastructure. Specialized knowledge. Time. But what happens when the competitor has access to powerful foundation models, open-source software, rented computing resources and increasingly capable AI coding agents? The cost of creating the next competitor falls. Then AI improves again. The coding agent becomes better. Development gets faster. The models become cheaper. The infrastructure becomes more accessible. And the next competitor may become cheaper still. The extraordinary paradox of the AI industry may therefore be this: ## THE TECHNOLOGY THAT ALLOWS A COMPANY TO BUILD AN EXTRAORDINARY PRODUCT ALSO MAKES IT EASIER FOR SOMEONE ELSE TO BUILD THE COMPETING PRODUCT. AI doesn't merely create products. **AI creates competitors.** --- ## DEEPSEEK SHOULD HAVE CHANGED THE CONVERSATION DeepSeek provided one of the clearest demonstrations of this phenomenon. Its DeepSeek-V3 technical report described a 671-billion-parameter mixture-of-experts model that activated 37 billion parameters for each token. More importantly, DeepSeek reported performance comparable to leading closed models while requiring approximately 2.8 million Nvidia H800 GPU-hours for full training. The exact economics of developing frontier models remain complicated, and training cost alone does not capture research, personnel, infrastructure or earlier experimentation. But DeepSeek demonstrated something strategically important: **Matching useful AI capability does not necessarily require matching every dollar spent by the largest American laboratories.** A competitor does not need identical infrastructure. It needs sufficient capability. This distinction matters enormously. The question isn't whether two models are identical. They aren't. The question is whether the less expensive model becomes **good enough for the customer's task**. If one model costs dramatically less while accomplishing that task adequately, the premium provider eventually has a pricing problem. And as models improve across the industry, "good enough" continually gets better. That is how commoditization begins. Not when every product becomes identical. But when the differences between products become less economically important than the differences in their prices. --- ## OPEN MODELS ACCELERATE THE PROCESS Open-weight AI introduces another powerful competitive force. Stanford's AI Index found that the performance gap between leading closed- and open-weight models on one major leaderboard fell from 8.0% in early 2024 to 1.7% by early 2025. It also found substantial convergence among frontier models generally. That does not mean open and closed models have become interchangeable. They have not. But the direction matters. Knowledge that once existed inside a handful of laboratories increasingly circulates through a global ecosystem of researchers, developers and companies. One laboratory discovers a better technique. Researchers study it. Developers reproduce it. Competitors incorporate variations. AI itself increasingly helps engineers understand and implement new approaches. The cycle repeats. Not every breakthrough spreads immediately. Some advantages remain proprietary. Some require enormous computing resources, specialized data, manufacturing capacity, distribution or accumulated expertise. But the relevant question for investors isn't merely whether competitive advantages exist. It is: # HOW LONG DO THEY LAST? A moat that lasts twenty years is enormously valuable. A moat that lasts twenty months is considerably less valuable. A moat that lasts twenty weeks may barely be a moat at all. And the faster AI accelerates technological diffusion, the more important the **duration of advantage** becomes. --- ## EVEN AI TALENT MAY BECOME LESS SCARCE The industry currently places extraordinary value on elite AI researchers and engineers. There are good reasons. The very best people remain extraordinarily valuable. But AI creates an uncomfortable recursive effect. AI engineers are building systems that make engineering easier. Coding agents improve. Research assistants improve. Models become better at mathematics, reasoning and software development. One exceptional engineer can increasingly command a virtual staff of AI tools and agents. Perhaps a project that once required 30 people eventually requires ten. Then five. Perhaps three highly capable humans working with agents can eventually build something that previously required an entire organization. We should be careful here. Today's agents remain unreliable at many long-duration, complex tasks, and organizations attempting aggressive AI-driven workforce reductions have encountered real limitations. Human judgment, coordination, security, institutional knowledge and accountability remain important. But that doesn't invalidate the direction. If AI allows fewer highly capable people to accomplish substantially more work, the economics of skilled labor change even without anything approaching complete human replacement. AI doesn't have to eliminate exceptional people. It merely has to **amplify them while reducing the quantity required for a given amount of output.** --- ## THE SAME FORCE IS REACHING HARDWARE The commoditization argument does not stop with software. The extraordinary economics surrounding AI chips are attracting extraordinary competition. Microsoft, Google, Amazon and others have developed custom AI silicon alongside their continued purchases from Nvidia and other semiconductor companies. In January 2026, Microsoft introduced Maia 200, a 3-nanometer inference accelerator that it says delivers more than 30% better performance per dollar than the latest-generation hardware previously operating in its fleet. Microsoft explicitly describes the objective as improving the economics of AI inference. None of this means Nvidia suddenly loses its technological leadership. It means something more fundamental. **Extraordinary margins attract capital.** Capital attracts competitors. AI increases the capability of those competitors. And customers purchasing enormous volumes of computing hardware have powerful incentives to develop alternatives themselves. Today's customer can become tomorrow's competitor. That process has occurred repeatedly throughout economic history. AI may simply accelerate it. --- # WHAT ABOUT THE DATA CENTERS? The same question applies to the enormous infrastructure buildout underway around artificial intelligence. McKinsey estimates that meeting worldwide AI computing demand could require approximately **$5.2 trillion of data-center investment by 2030** under its central scenario. Meanwhile, estimates for 2026 capital expenditures by the major U.S. hyperscalers have reached roughly **$700 billion**. Perhaps this extraordinary infrastructure buildout will prove necessary. Demand for AI computing continues to grow rapidly. Training frontier systems remains extraordinarily compute-intensive, and inference demand could become enormous if AI adoption spreads throughout the global economy. But there is another possibility worth considering. What happens if: # AI INTELLIGENCE COMPOUNDS FASTER THAN AI'S NEED FOR CENTRALIZED INFRASTRUCTURE? Models become more efficient. Inference costs decline. Specialized chips improve. Power efficiency improves. Smaller models become more capable. More intelligence moves onto PCs, vehicles, robots, phones and industrial equipment. None of these developments eliminates data centers. Indeed, falling costs could stimulate so much additional AI usage that total compute demand continues rising. That is the strongest counterargument to the commoditization thesis: efficiency can create additional demand rather than reduce aggregate infrastructure requirements. But the uncertainty itself creates an extraordinary capital-allocation problem. ## THE AI INDUSTRY IS MAKING DECADES-LONG INFRASTRUCTURE INVESTMENTS WHILE THE TECHNOLOGY THOSE INVESTMENTS SERVE CAN CHANGE MATERIALLY IN MONTHS. That doesn't mean the investments are wrong. It means their economic durability deserves considerably more scrutiny. --- # THEN COME THE AGENTS Agents may accelerate all of this. We normally discuss AI agents as replacements or assistants for individual workers. Programmers. Researchers. Marketers. Accountants. Lawyers. Customer-service representatives. But companies themselves are collections of these functions coordinated toward common objectives. So eventually the more important question may become: ## HOW LONG BEFORE AN AI AGENT CAN HELP CREATE AN AI COMPANY? Give it an objective. Give it a budget. Give it access to models and computing resources. Allow it to deploy specialized agents. Then let the system research a market, analyze competitors, design a product, write software, test it, deploy it, market it and continuously improve it. We are not there yet. Perhaps initially 30 humans supervise the process. Then ten. Then three. The important point isn't predicting the exact number. It is understanding the direction. **The cost of creating organizations capable of competing may fall dramatically.** And if the cost of creating competitors falls, barriers to entry fall with it. --- # THIS DOES NOT MEAN EVERYTHING BECOMES A COMMODITY This is the most important qualification to the thesis. Some scarcity will remain. Physical resources remain scarce. Land remains scarce. Energy can be constrained. Semiconductor fabrication is extraordinarily difficult. Distribution matters. Brands matter. Customer relationships matter. Regulation can create barriers. Proprietary data can matter. Capital matters. Trust matters. And some technological advantages may prove remarkably durable. AI will not literally commoditize everything. That is why our thesis is: # AI COMMODITIZES VIRTUALLY EVERYTHING. It is deliberately provocative. The important proposition isn't that every product becomes indistinguishable. It is that **AI may systematically reduce the cost and time required to reproduce economic capability across an extraordinary range of industries.** If that happens, the consequences extend far beyond technology. --- # THE GREAT AI PARADOX AI could transform medicine. Education. Science. Engineering. Manufacturing. Finance. Transportation. Defense. Software. Professional services. And almost every other major industry. AI adoption could become nearly universal. AI could generate trillions of dollars of economic value. And many individual AI companies could still capture far less of that value than today's valuations imply. There is no contradiction. **Technological value and shareholder value are different things.** Transformative technologies can generate enormous benefits that spread far beyond the companies that originally developed them. AI could follow its own version of that pattern. The technology wins. Society receives enormous value. Competition proliferates. Prices fall. Capability becomes abundant. And economic rents migrate elsewhere. --- # WHERE IS THE MOAT? That may become one of the defining investment questions of the AI revolution. Not: **How powerful will AI become?** It is already becoming extraordinarily powerful. Not: **How large is the total addressable market?** The potential market could be enormous. The harder question is: # HOW LONG CAN ANY AI COMPANY REMAIN SPECIAL? Because extraordinary returns require some form of scarcity. And AI appears increasingly capable of attacking scarcity itself. The mechanism is remarkably simple: **AI creates capability.** **Capability creates competitors.** **Competitors create abundance.** **Abundance destroys scarcity rents.** AI may become one of the most successful technologies humanity has ever created. And that may be precisely why maintaining extraordinary profits from it proves so difficult. --- ## SOURCES & FURTHER READING **Stanford Institute for Human-Centered Artificial Intelligence — 2025 AI Index Report.** Data on falling inference costs, improving hardware economics, smaller-model performance and convergence between open- and closed-weight models. **DeepSeek — DeepSeek-V3 Technical Report.** Model architecture, benchmark results and reported training-compute requirements. **Microsoft — Maia 200: The AI Accelerator Built for Inference.** Microsoft's specifications and reported economics for its custom 3nm AI accelerator. **McKinsey & Company — The Cost of Compute: A $7 Trillion Race to Scale Data Centers.** Estimates of worldwide AI-related data-center investment requirements through 2030. *Published August 2026\. The Institute for Strategic Intelligence distinguishes between documented developments, analytical conclusions and forecasts. Predictions and analytical judgments in this article should be evaluated against subsequent evidence.*