Does Technological Innovation Really Lower Prices?

1. The Core Question: Does Technology Make Things Cheaper or More Expensive?

One of the oldest beliefs about technological innovation is that “good technology makes the world cheaper.” Automobiles reduced the cost of transportation compared with horse-drawn carriages. Semiconductors reduced the cost of computation. The internet pushed the cost of distributing information close to zero. Smartphones combined a camera, map, music player, bank, newspaper, and messenger into a single device. Consumers gained access to far more functions at a far lower cost than in the past.

From this perspective, AI appears to be moving along the same path. AI can lower the cost of writing, coding, customer service, translation, design, data analysis, education, and medical support. It can help one person complete tasks in less time, raise the productivity of beginners, and automate repetitive work. This is the core of the technology-deflation argument often associated with Cathie Wood and ARK Invest. Disruptive innovation breaks existing cost structures and provides the same function in a cheaper and more efficient way.

AI models may look like digital software, but their actual foundation is deeply physical. AI consumes electricity, runs on chips, and produces heat inside data centers. High-performance GPUs require HBM and advanced packaging. Data centers require transformers, cooling systems, and grid connections. As AI is used more widely, the cost per unit of computation may fall, but if total computing demand explodes, demand for electricity, equipment, land, and raw materials can rise sharply.

AI is both a technology that lowers costs and a demand shock that raises bottleneck prices.

2. Theoretical Background: Wright’s Law and the Decline in Technology Costs

To understand the logic of technological deflation, we first need to look at Wright’s Law. Wright’s Law originated from Theodore Paul Wright’s 1936 study of aircraft production costs. Its core idea is simple: whenever cumulative production doubles, unit production costs tend to fall by a consistent percentage.

People often assume that technology becomes cheaper simply because time passes. But Wright’s Law says something more precise.

This distinction matters. Costs do not fall simply because years go by. They fall when production experience accumulates, processes improve, supply chains expand, workers become more skilled, equipment becomes standardized, and defect rates decline. The decline in technology prices is not primarily the “effect of time.” It is the “effect of cumulative learning.”

Although Wright’s Law began with aircraft, it has since been applied to many other industries. Semiconductors, solar panels, batteries, wind power, displays, electric vehicles, and robotics are all important examples. In the solar industry, a related idea is known as Swanson’s Law. Solar module prices have tended to fall by roughly 20% whenever cumulative shipments doubled, and as a result, the cost of solar power has declined dramatically over several decades.

The key point is that technological innovation is not just about new invention. Invention is only the starting point. Real cost reduction comes from mass production, standardization, supply chain expansion, and learning effects.

When this virtuous cycle works, technology becomes a force that lowers prices.

3. Historical Examples: Solar Power, Batteries, and Semiconductors

Solar power is one of the clearest examples of technological cost decline. In the 1970s, solar power was an extremely expensive energy source. It was closer to a specialized technology used for satellites and niche applications. But as production increased over several decades and global supply chains expanded, including in China, solar panel prices fell sharply. The cost of generating the same amount of electricity declined significantly, and solar power has now become one of the cheapest sources of electricity in many regions.

Batteries followed a similar path. Lithium-ion batteries were once expensive components mainly used in laptops and mobile phones. But as the electric vehicle market expanded, battery cell production surged, manufacturing processes became more automated, and materials improved. As a result, battery pack prices declined over the long term. There were temporary periods of cost pressure when raw materials such as lithium, nickel, cobalt, and graphite became more expensive, but the long-term trend has still been one of cost reduction.

Semiconductors are another symbol of technological deflation. Moore’s Law reduced the cost of computing performance by allowing more transistors to fit into the same area. Consumers gained access to more powerful computing at lower prices. Smartphones, cloud computing, internet services, and artificial intelligence all grew on top of this long-term decline in semiconductor costs.

In other words, technology lowers the core unit cost of a product or service. But when that technology spreads across society, it often creates new infrastructure costs. This point becomes even more important in the AI era.

4. AI Productivity: The Strongest Candidate for Cost Reduction

The argument that AI can lower prices is centered on productivity growth. In economics, productivity means the ability to produce more output with the same amount of labor and capital. When productivity rises, companies can do more with the same workforce, and over time, the unit cost of goods and services can fall.

Generative AI is already showing this potential, especially in knowledge work. AI is being used as an assistant in customer support, document writing, coding, translation, marketing copy, research, data organization, and educational content creation.

This point is important. The real economic impact of AI does not come only from making one genius even more powerful. It may come more broadly from raising the productivity of average workers, narrowing skill gaps, and automating repetitive tasks.

Still, there is a reason to be cautious. Simply adopting AI does not automatically raise productivity across an entire organization. Companies must clean up data, redesign workflows, and verify AI-generated outputs. They also face costs related to error management, security, copyright, regulation, and internal training. Even if AI reduces the time required for an individual task, that saved time may not immediately translate into higher company revenue or lower operating costs.

“You can see the computer age everywhere but in the productivity statistics.”

— Robert Solow, on the productivity paradox

This quote was used to describe what happened during the spread of computers in the 1980s. Computers appeared throughout the office, but macroeconomic productivity statistics did not immediately improve. Only after companies changed their workflows and organizational structures did the productivity gains from information technology become clearer.

AI may follow a similar path. AI tools are spreading quickly, but their productivity effects may only become visible after organizations redesign the way they work. Therefore, AI’s deflationary effect is likely to appear not immediately, but with a delay.

5. The Counterargument: AI Creates a New Form of Bottleneck Inflation

If we focus only on the claim that AI reduces costs, AI looks like an obvious deflationary technology. But when we examine its physical foundation, the story changes.

AI is not an intangible piece of software. AI is also an electricity industry, a semiconductor industry, and a data center industry. Training and running generative AI models requires high-performance GPUs. GPUs require HBM, advanced packaging, and testing equipment. Servers must be installed in data centers, and data centers require massive amounts of electricity and cooling. If grid connections are delayed, data center completion can also be delayed. Transformers, cables, switchgear, cooling equipment, land, water, and even natural gas power generation can all become bottlenecks.

In the United States, data center electricity demand has already become a major economic issue. Electricity consumption is expected to reach record highs in 2026 and 2027, driven by AI data centers and broader electrification. In some regions, data center electricity demand has become a key variable in grid planning, and securing power has become one of the most important conditions in data center location strategy.

The problem with AI data centers is not just total electricity use. Geographic concentration also matters. If data centers cluster in specific regions, the national power system may appear capable of handling the demand, while local grids still come under intense pressure. This is why data center electricity demand is increasingly discussed as a grid stress factor in places such as Virginia, Oregon, and Ireland.

Therefore, the inflation created by AI may not be something consumers immediately notice at the grocery store. It may first appear in power grids, transformers, generation equipment, cooling systems, copper, data center land, and power purchase agreement prices. Later, these costs may be passed through to cloud prices, AI service prices, corporate IT budgets, and local electricity bills.

6. Expert Perspectives: Optimism and Caution

Expert opinions on the relationship between AI and prices can be divided into several broad perspectives.

Ultimately, the important question is not whether AI is inflationary or deflationary.

Depending on the time horizon and the industry, AI can be both.

7. Analysis: Prices in the AI Era May Move in a K-Shaped Pattern

If we look at prices in the AI era only through a single average figure, we may miss the real story. Looking only at the overall consumer price index may make AI’s impact appear small. But at the industry level, price movements are likely to differ significantly.

Therefore, prices in the AI era may move in a K-shaped pattern. Prices may fall in areas that AI automates or replaces, while prices may rise for the bottleneck assets required for AI expansion.

This has important implications for investors. The winners of the AI era will not be limited to companies that build AI models. Companies that use AI to lower costs can also be winners, and so can companies that solve the bottlenecks created by AI expansion.

8. Investment Perspective: There Are Two Types of AI Beneficiaries

In the AI era, the broad label “AI stocks” is not enough. Investors need a more precise framework. There are two key categories.

These two groups are different in nature. Beneficiaries of cost decline improve margins by using AI. Beneficiaries of bottleneck pricing power become more valuable because AI increases demand for their scarce products and services.

This question creates the investment map for the AI era.

9. Conclusion: Technology Lowers Prices, but Not Immediately