Who Made Technology Cheap?

Innovation created the possibility. Globalization made it scalable.

Antwan Baker, Founder

Research status: This paper is a literature-grounded synthesis of existing evidence. It does not yet contain Laray.ai's planned original compute and battery decompositions. The reviewed empirical methodology is preserved separately in methodology-v0.3.md.

Abstract

Why have computers, batteries, solar panels, televisions, and other technologies become dramatically more capable without becoming proportionally more expensive?

The usual explanations divide into two camps. One says technology became cheaper because engineers made it better. The other says technology became cheaper because corporations moved manufacturing to countries with lower wages, supportive governments, and dense supplier networks.

Neither explanation is sufficient by itself.

Scientific and engineering progress determines what a product can do and how efficiently it can be made. Global manufacturing determines how quickly that progress can be produced at scale and delivered at an affordable price. Government policy, access to capital, commodity prices, competition, logistics, and producer margins determine how much of the resulting cost decline reaches the customer.

The balance differs by industry. In general-purpose computing, improvements in performance and product quality explain far more of the decline in price per unit of service than manufacturing relocation alone. In lithium-ion batteries, chemistry, energy density, material use, factory scale, production experience, policy, and Asian manufacturing ecosystems are deeply intertwined. In solar modules, both technical progress and global supply chains have made large, measurable contributions. In assembly-heavy consumer electronics, manufacturing location and supplier clusters likely account for a larger portion of cost decline than they do in compute.

The central conclusion is therefore not that innovation or offshoring “won.” It is that they played different roles.

Innovation moved the technological frontier. Globalization compressed the cost and time required to commercialize it. Competition determined who captured the savings.

Introduction

Modern technology routinely violates the intuition people develop from ordinary goods.

A chair may become cheaper because its manufacturer finds less expensive wood, automates part of the factory, or moves production to a lower-cost country. But a computer can remain near the same nominal price while becoming hundreds or thousands of times more capable. A battery may become less expensive per kilowatt-hour while also becoming lighter, safer, longer-lasting, and more powerful. A solar module may fall in price while generating more electricity from the same physical area.

This creates a measurement problem before it creates an economic one.

The relevant question is not merely whether the sticker price fell. It is whether the price of the service delivered fell. For a processor, the service might be useful compute performance. For a battery, it might be usable energy delivered over its lifetime. For a solar module, it might be lifetime electricity generation per unit of area.

A product can therefore become cheaper in at least two fundamentally different ways:

1. The physical product becomes less expensive to manufacture.
2. The product delivers more useful service for the same price.

The first route is where labor costs, factory location, supply chains, scale, commodity inputs, and margins often matter most. The second is where architecture, chemistry, materials science, software, design, and engineering progress dominate.

Public arguments about technology and globalization often collapse these routes into one. This paper separates them.

Its conclusion is necessarily heterogeneous: different technologies became cheap for different reasons, and even within the same industry the dominant mechanism changed over time.

I. The Price Tag Is Not the Product

Imagine two processors that both sell for $300. The newer processor completes twice as much useful work in the same amount of time.

Its sticker price has not declined. Its price per unit of compute service has fallen by half.

Statistical agencies attempt to capture this distinction through quality adjustment. The U.S. Bureau of Labor Statistics uses hedonic and related techniques for rapidly changing products because simply following the price of the same model can miss improvements introduced through product replacement. BLS explains that hedonic methods separate price-determining characteristics from observed price change, and it has repeatedly revised its microprocessor methods as product design and pricing behavior evolved. BLS, Hedonic Models in the Producer Price Index; BLS, Microprocessor Quality Adjustment.

This is not a minor technical detail. The measured history of technological progress can change depending on whether a price index follows identical models, compares product characteristics, or directly evaluates performance.

Recent BLS research emphasizes that static list pricing can make quality adjustment especially important in microprocessor indexes: producers may hold a model's price relatively steady while introducing more capable generations around it. Adams and Sawyer, 2024.

The implication is profound:

Some apparent price stability is rapid technological deflation hidden inside a better product.

Offshoring cannot fully explain this type of decline. Lower-cost assembly may reduce the price of a physical unit, but it cannot by itself explain why the unit performs dramatically more work, stores more energy, produces more electricity, or lasts longer.

II. Computing: The Strongest Case for Innovation

General-purpose computing provides the clearest example of a technology becoming cheaper primarily because the service delivered by the product improved.

A processor is not valuable because it is a small piece of packaged silicon. It is valuable because of the computations it can perform. Improvements in architecture, transistor density, process technology, memory hierarchy, power management, software compatibility, and manufacturing yield allow newer chips to provide more useful work per dollar.

Ana Aizcorbe's Bureau of Economic Analysis study decomposed Intel microprocessor price decline into product-quality improvement, physical cost change, and markup change. For the period studied, the acceleration in the price-index decline was attributed almost entirely to faster quality improvement. Declining Intel markups from 1993 through 1999 accounted for about 6 percentage points of an average quarterly index decline of roughly 24 percent—important, but much smaller than the quality effect. Aizcorbe, BEA, 2005.

This does not mean manufacturing geography was irrelevant.

Semiconductor production is among the most globally fragmented production systems in the world. Design, intellectual property, fabrication equipment, specialty chemicals, wafers, fabrication, packaging, testing, memory, and final system assembly may be distributed across multiple countries.

Global specialization can lower physical costs. It can also improve the final product by allowing firms to use world-leading suppliers at each stage. But the economic value of the resulting chip still depends primarily on what it can do.

A thought experiment makes the distinction clear.

Suppose a company moved production of an unchanged processor to a country where total manufacturing cost was 30 percent lower. The processor's price could potentially decline substantially.

Now suppose architecture and fabrication advances made the processor ten times faster without changing the retail price. The price of compute service would decline by 90 percent.

The relocation is economically important. The technical improvement is transformational.

This is why the history of compute cannot be reduced to wage arbitrage. Even a perfectly optimized global supply chain could not manufacture a modern processor from a 1995 design and produce today's performance. The frontier moved because the underlying technology changed.

At the same time, computing reveals why “innovation versus globalization” is a false binary. Global specialization may accelerate innovation by giving chip designers access to specialized foundries, equipment makers, materials suppliers, packaging firms, and enormous production volumes. This is an inference about the system rather than a cleanly identified percentage, but it is consistent with the structure of the modern semiconductor value chain.

The best-supported conclusion is therefore:

In general-purpose compute, innovation dominates the decline in price per unit of service. Globalization materially affected physical cost, scale, and the distribution of value, but manufacturing relocation did not create the majority of the performance improvement itself.

That conclusion should not be casually extended to all forms of compute. After roughly 2015, accelerators, chiplets, high-bandwidth memory, advanced packaging, specialized workloads, and heterogeneous systems make a single universal measure of compute service increasingly difficult. The older CPU-centered evidence remains important, but it does not automatically describe the economics of modern AI accelerators.

III. Offshoring Did More Than Lower Wages

If the computing story is often used to overstate the power of innovation alone, manufacturing statistics may understate the role of offshoring.

Suppose an American manufacturer buys a component from a domestic supplier for $100. It later switches to a foreign supplier offering a comparable component for $65.

A conventional price index may begin following the foreign supplier only after the switch. It records future changes from the new $65 base but may never record the initial $35 reduction.

The cost saving occurred, but the index missed it.

Houseman, Kurz, Lengermann, and Mandel argued that this supplier-switch problem causes imported input growth to be understated and domestic manufacturing productivity and real value added to be overstated. For 1997–2007, they estimated that annual manufacturing multifactor-productivity growth may have been overstated by roughly 0.1 to 0.2 percentage point and real value-added growth by roughly 0.2 to 0.5 percentage point. Houseman et al., 2011.

This does not prove that apparent productivity growth was simply offshoring in disguise. It proves that the standard measurement system can miss some of the largest discrete savings created by moving to new suppliers.

Offshoring also involves much more than replacing an expensive worker with a cheaper worker.

Production often moves into an ecosystem containing:

• specialized component makers;
• tooling companies;
• trained technical labor;
• ports and logistics infrastructure;
• nearby contract manufacturers;
• engineering teams able to modify designs rapidly;
• governments willing to subsidize land, energy, finance, or construction;
• high-volume customers that support large factories.

Once such a cluster exists, its advantage is no longer just low wages. It becomes a production technology in its own right.

This complicates attempts to separate “technology” from “offshoring.” A supplier cluster may create new process knowledge. A larger factory may justify automation that would not be economical at smaller volume. A global customer base may increase cumulative production and accelerate improvement. Digital design and communications technology may make it possible to coordinate these activities across borders in the first place.

Research on production fragmentation supports this two-way relationship: communications and production technologies can make foreign sourcing feasible, while fragmented production changes where firms place different stages of the value chain. Teresa Fort, NBER Working Paper 22550.

Offshoring is therefore not merely an alternative to technological progress.

In many industries, it is one of the systems through which technological progress becomes commercially viable.

IV. Batteries: No Single Cause Survives Inspection

Lithium-ion batteries are the strongest case against assigning one dominant explanation.

The International Energy Agency reports that lithium-ion battery prices declined from about $1,400 per kilowatt-hour in 2010 to less than $140 per kilowatt-hour in 2023. It attributes the broad decline to research and development progress and economies of scale in manufacturing. IEA, Batteries and Secure Energy Transitions, 2024.

But “battery cost” is not one stable object.

A battery cell can change in:

• chemistry;
• cathode and anode composition;
• energy density;
• material quantity per kilowatt-hour;
• cycle life;
• power capability;
• safety characteristics;
• format;
• manufacturing process;
• production yield.

The market can also shift between chemistries. Lithium iron phosphate batteries generally cost less and avoid nickel and cobalt but have historically offered lower energy density than some nickel-rich chemistries. That tradeoff matters differently in a passenger vehicle, a premium long-range vehicle, and a stationary storage installation.

The IEA reported that LFP had risen to approximately 40 percent of electric-vehicle sales and 80 percent of new battery storage in 2023. This illustrates how chemistry mix can change the observed average price independently of improvement within a chemistry. IEA, 2024.

Ziegler, Song, and Trancik found that lithium-ion cell prices had fallen by approximately 97 percent since commercial introduction. Their engineering decomposition examined changes in energy-capacity characteristics, material prices, and non-material manufacturing costs rather than treating the entire decline as an abstract learning rate. Ziegler, Song, and Trancik, 2021.

That approach reveals the multiple layers of “technology.”

An increase in energy density may be recorded as a product improvement because the cell stores more energy for the same mass or volume. But it may also lower manufacturing cost per kilowatt-hour because less casing, separator, electrolyte, or other material is needed for each unit of energy.

A chemistry change may improve safety and reduce expensive material use while lowering energy density. A larger factory may spread fixed costs while also generating enough production experience to raise yields. A government subsidy may cause the factory to be built, after which the scale and learning effects appear to be organic manufacturing improvements.

These effects are real. They are not independent.

A structural study by Barwick, Kwon, Li, and Zahur estimates a learning rate of 9.2 percent during its sample after controlling for industry technological progress, economies of scale, input costs, and vehicle-assembly experience. The study also models the role of government policies in the global EV battery industry. Barwick et al., NBER Working Paper 33378.

That estimate should not be added to an engineering decomposition as another independent share.

Engineering models describe what changed inside the cost structure. A structural learning model attempts to explain how accumulated production experience affected cost after accounting for other variables. Learning may operate through higher yield, lower scrap, better process control, improved material usage, or redesigned equipment—the same changes already visible in the engineering accounts.

China, South Korea, and Japan also developed deep battery production and materials ecosystems. Their role cannot be represented only by a wage differential. The relevant advantages include supplier density, policy support, infrastructure, access to processed materials, engineering talent, production volume, and rapid customer feedback.

The most defensible conclusion is:

Battery cost decline is a joint product of technical improvement, material efficiency, input-price change, scale, manufacturing experience, policy, chemistry substitution, and Asian production ecosystems. No credible single percentage can separate these forces without imposing strong model assumptions.

This is not an analytical failure. It is the correct description of an industry in which the mechanisms reinforce one another.

V. Solar: Technology and Globalization Are Both First-Order Causes

Solar photovoltaics offer perhaps the clearest example of technical progress and globalization making large contributions simultaneously.

Solar modules became cheaper because cells became more efficient, factories grew, manufacturing yield improved, silicon use changed, material costs shifted, supply chains specialized, and production concentrated in large manufacturing ecosystems.

A one-factor learning curve summarizes the relationship between cumulative production and falling cost, but it does not identify the underlying causes.

Gregory Nemet's research concluded that learning from experience only weakly explained changes in several important drivers of photovoltaic cost decline, including plant size, module efficiency, and silicon cost. Those drivers reflected engineering decisions, research, industry structure, input conditions, and investment—not merely repetition of the same process. Nemet, 2006.

At the same time, globalized production created large cost savings.

Helveston and coauthors modeled historical solar deployment in the United States, Germany, and China and compared the globalized supply chain with counterfactual national production paths. They estimated cumulative installer savings from 2008–2020 of approximately $24 billion in the United States, $7 billion in Germany, and $36 billion in China. Helveston et al., Nature, 2022.

This is strong evidence that globalization was not a minor afterthought in solar.

But it would be equally misleading to say China merely took an unchanged technology and produced it with cheaper labor. The manufacturing process itself improved. Plants became larger. Equipment, wafering, cell design, efficiency, material use, yield, and supply-chain coordination changed. The manufacturing ecosystem became a source of technical progress.

Solar therefore supports a more integrated interpretation:

Technical progress made higher-performance, lower-input solar modules possible. Global manufacturing scale and specialized supply chains dramatically reduced the cost of producing and deploying them.

Attempting to assign the same gain exclusively to one side would count the history incorrectly.

VI. Who Captured the Savings?

Even after production cost falls, the customer may not receive the entire gain.

A firm can use lower production costs to:

• lower prices;
• increase product quality;
• expand distribution;
• absorb new regulatory or logistical costs;
• increase gross margin;
• fund research and development;
• underprice competitors temporarily.

The division depends on market structure.

In a highly competitive mature market, cost reductions may pass quickly to customers. The technology becomes cheaper, but manufacturers may earn poor returns.

In a concentrated market with differentiated products, the producer may retain a larger share. Customers still receive better performance, but the company captures part of the technical gain as margin.

This distinction matters for investors.

The company that creates the most impressive cost decline is not automatically the best investment. If competitors can copy the process and customers capture every dollar through lower prices, the social value may be large while shareholder returns remain weak.

Conversely, a firm controlling scarce intellectual property, manufacturing equipment, a difficult process, a critical material, a distribution platform, or a dominant ecosystem may preserve a larger share of the value.

The economic history of cheap technology must therefore answer two questions:

1. Who caused the cost or performance improvement?
2. Who retained the economic surplus?

These are not the same.

VII. A More Accurate Answer

The original question asked how much of technology's falling cost came from moving manufacturing overseas versus technology actually getting better.

The evidence does not support one universal percentage.

It supports a hierarchy of conclusions.

1. For general-purpose compute, technological improvement appears dominant.

Manufacturing geography lowered physical costs and enabled specialization, but the extraordinary decline in price per unit of useful compute came principally from better products and processes. A cheaper location cannot explain orders-of-magnitude performance improvement in an unchanged chip.

2. For assembly-heavy consumer electronics, globalization likely has a larger direct role.

When products rely on modular components and labor-intensive final assembly, moving into dense supplier clusters can create large reductions in landed cost. Quality still improves, but location, logistics, component standardization, scale, and competition matter more directly than in frontier processors.

A precise share remains difficult because the same clusters may also accelerate product and process innovation.

3. For lithium-ion batteries, the mechanisms are inseparable at a high level.

Chemistry, material use, energy density, factory scale, manufacturing experience, subsidies, input prices, and Asian supply chains all contributed. Engineering decomposition can measure specific changes, but broad causal shares require model assumptions.

4. For solar modules, technical progress and globalization are both first-order causes.

Efficiency, material use, manufacturing yield, and factory scale explain important portions of the decline. Global supply chains also generated large measurable savings and accelerated deployment.

5. “Learning” is usually a summary, not a final explanation.

Cost often falls as cumulative production rises. But cumulative production also rises when demand increases, policy subsidizes deployment, factories become larger, suppliers improve, and technology advances. A learning curve describes the pattern unless the underlying mechanisms are separately identified.

6. Offshoring's contribution is probably understated by conventional statistics.

Supplier-switch savings can disappear between price samples, making foreign-sourcing gains look like unexplained domestic productivity.

7. Technology's contribution is also understated when quality is ignored.

A product can stay at the same price while providing far more service. Raw sticker-price analysis misses much of the gain in industries such as computing.

VIII. The Investor Implications

This history provides a framework for evaluating future technologies.

Innovation-Dominant Industries

Look for companies controlling:

• architecture;
• intellectual property;
• design software;
• frontier manufacturing equipment;
• difficult materials science;
• process know-how.

The central risk is technological exhaustion: investors may extrapolate historical improvement rates after the frontier becomes more expensive or physically constrained.

Scale-Dominant Industries

Look for companies able to:

• finance large facilities;
• maintain high utilization;
• secure long-term customers;
• translate production volume into higher yield;
• purchase inputs at advantageous terms.

The central risk is excess capacity. Scale lowers unit cost only when factories remain sufficiently utilized.

Cluster-Dominant Industries

Look for ecosystems rather than isolated companies.

A firm may possess intellectual property but still struggle if it lacks nearby suppliers, tooling, skilled labor, infrastructure, logistics, and manufacturing partners.

The central risk is geographic concentration. The same cluster that creates low cost can create geopolitical vulnerability.

Policy-Dominant Industries

Distinguish temporary support from structural advantage.

A subsidy may create enough initial demand for an industry to achieve scale and become competitive. It may also conceal an uneconomic cost structure that reappears when support ends.

Pass-Through-Dominant Markets

Beware industries where technological progress is spectacular but producer economics are weak.

Falling prices and improving products can create enormous value for society while competition transfers most of that value to customers.

Conclusion

Technology did not become cheap because engineers improved it or because manufacturing moved overseas.

It became cheap because those forces operated together—but not equally in every industry.

Scientific and engineering progress increased what products could do. Process innovation reduced the physical resources required to make them. Global production networks expanded scale, concentrated suppliers, lowered input and labor costs, and accelerated commercialization. Governments altered the economics through subsidies, trade rules, infrastructure, and directed finance. Competition determined whether the gain became lower prices, better products, or higher margins.

The cleanest summary is:

Innovation created the possibility. Globalization made it scalable. Market structure determined who captured the benefit.

For compute, the first force appears dominant.

For assembly-heavy electronics, the second carries more weight.

For batteries and solar, the two became so intertwined that forcing them into mutually exclusive percentages would misrepresent the history.

That conclusion is less politically satisfying than declaring either globalization or technological genius the winner. It is also more useful.

It tells investors, companies, and governments what must be preserved.

A country can reshore production and retain the underlying science while accepting higher manufacturing costs. It can preserve low prices through global supply chains while becoming dependent on concentrated foreign clusters. It can subsidize scale without creating durable engineering leadership. It can lead in invention while losing the production experience through which later generations of innovation emerge.

The next era of technological progress will therefore be shaped not simply by who invents the future, or by who manufactures it most cheaply.

It will be shaped by who can connect invention, production, scale, supply chains, and value capture into one durable system.

Committee Review and Accountability

This paper was developed through Laray.ai's multi-model research process. Independent methodological reviews were provided by Gemini (Research), Claude (Risk), DeepSeek (Independent Review), and Grok (Contrarian). ChatGPT performed synthesis and drafting.

Antwan Baker made the final editorial, methodological, and publication decisions and remains the accountable human author.

Selected References

• Aizcorbe, Ana M. Why Are Semiconductor Price Indexes Falling So Fast? Industry Estimates and Implications for Productivity Measurement. U.S. Bureau of Economic Analysis, 2005.
• Adams, Brian, and Sawyer, Kenneth. Hedonic Price Indexes under Static Pricing: An Application to PPI Microprocessors. U.S. Bureau of Labor Statistics, 2024.
• Barwick, Panle Jia; Kwon, Hyuk-Soo; Li, Shanjun; and Zahur, Nahim B. Drive Down the Cost: Learning by Doing and Government Policies in the Global EV Battery Industry. NBER Working Paper 33378, 2025.
• Fort, Teresa. Technology and Production Fragmentation: Domestic versus Foreign Sourcing. NBER Working Paper 22550, 2016.
• Helveston, John Paul, and coauthors. Quantifying the Cost Savings of Global Solar Photovoltaic Supply Chains. Nature, 2022.
• Houseman, Susan N.; Kurz, Christopher J.; Lengermann, Paul; and Mandel, Benjamin R. Offshoring Bias in U.S. Manufacturing. Journal of Economic Perspectives, 2011.
• International Energy Agency. Batteries and Secure Energy Transitions. 2024.
• Nemet, Gregory F. Beyond the Learning Curve: Factors Influencing Cost Reductions in Photovoltaics. Energy Policy, 2006.
• Ziegler, Micah S.; Song, Juhyun; and Trancik, Jessika E. Determinants of Lithium-Ion Battery Technology Cost Decline. Energy & Environmental Science, 2021.