Chapter 15: The Deflationary Cascade

In 1977, installing a single watt of solar photovoltaic capacity cost $76.67. In 2024, $0.24. Every dollar spent on solar in 1977 now buys 319 times as much capacity. The force responsible has a name: Wright's Law. Theodore Wright documented the pattern in 1936 studying airplane manufacturing. Every time cumulative production of a good doubled, unit cost fell by a consistent percentage. The percentage held across industries. Across decades. Across technologies. The Boston Consulting Group renamed it the "experience curve" in the 1960s. The renaming obscured the deeper point. Wright's Law is about information. Each unit produced teaches the production system something, and that teaching compounds.

Santa Fe Institute researchers have validated Wright's Law against Moore's Law and other technology forecasting methods across 62 technologies. Wright's Law produced more accurate predictions than alternatives in the majority of cases. The law holds because it is rooted in a deeper truth: production is learning, and learning compounds.

The compounding has reached a threshold that changes the structural conditions of civilization. All the curves are falling at once, and the fall is accelerating in the most transformative sector of all: intelligence itself.

The Data

Solar PV: $76.67 per watt (1977) to $0.24 (2024). 99.7% decline, approximately 10% per year. China-manufactured modules approach $0.10 per watt. Utility-scale solar levelized cost of energy fell approximately 90% from $360/MWh in 2010 to $30-50/MWh in 2022.

Lithium-ion batteries: $1,100 per kilowatt-hour (2010) to $108 (2025). BloombergNEF reports the lowest observed pack price at $50/kWh in lithium iron phosphate chemistry from Chinese manufacturers. A 90% decline in fifteen years. The steepest section of the curve lies ahead.

Genome sequencing: $95 million (2001) to $200 (2022). 99.9998% decline. After next-generation sequencing was adopted in 2008, cost fell faster than Moore's Law by an order of magnitude. The cost of reading the code of life dropped faster than the cost of computing.

Compute: $18.75 million per GFLOPS (1984) to $0.03 (2017). More than twelve orders of magnitude. A ratio so vast the human mind cannot hold both endpoints at once.

AI inference: $20 per million tokens to $0.07. A 99.65% decline. Median cost falls roughly 50x per year. DeepSeek trained frontier-competitive models for approximately $5.5 million, against hundreds of millions spent by Western labs on comparable performance.

Digital storage: $193,000 per gigabyte (1980) to $0.014 (2022). LED lighting: $90 per kilolumen (2008) to $1-3 (2020). Internet transit: $1,200 per Mbps (1998) to $0.50 (2020). Satellite launch: $54,500 per kilogram to low earth orbit (Space Shuttle era) to $2,720 (Falcon 9).

Each curve is well documented by itself. The synthesis, that they are falling simultaneously across every input to civilization, driven by the same dynamic, is what makes this moment structurally different from the printing press, the steam engine, or the early internet.

Why Simultaneous Matters

Steam deflated the cost of mechanical labor. Energy remained scarce. Electricity deflated distance. Intelligence remained scarce. The internet deflated information cost. Physical production remained expensive. Each revolution produced abundance in one domain, generating a coordination challenge that the next scarce resource was called upon to manage. Each created new jobs organized around the remaining scarcity.

The deflationary cascade deflates energy, intelligence, computation, and production at the same time.

When one input deflates, the economy adjusts. Workers move to new industries. New scarcities emerge to organize around. Skilled labor after the printing press. Management after steam. Attention after the internet.

When every input deflates simultaneously, no new scarcity emerges to anchor the economic system. The scarcity-based architecture does not adjust. It reaches a phase transition, the way water at 100 degrees Celsius reorganizes into steam. One degree, but the properties of the system change categorically. Sequential deflation is adjustment. Simultaneous deflation may be transformation.

The money-as-scarcity-tool analysis makes this concrete. Money's three functions each require scarcity to operate. Store of value: if anyone can create unlimited money, it stores nothing. The denarius held value because 3.9 grams of silver could not be conjured from air. Unit of account: the measuring stick must be stable, and scarcity provides stability. When Spain flooded Europe with Potosi silver, the stick stretched and prices rose across the continent, not because goods changed but because the unit warped. Medium of exchange: the token must be costly enough to prevent counterfeiting. All three functions assume that what is being stored, measured, and exchanged is limited. When solar energy costs $0.24 per watt and falling, when AI inference costs $0.07 per million tokens and falling, when the things the economy produces approach free, the tool designed to manage their scarcity loses its structural foundation.

The measurement system confirms the mismatch. The BLS Producer Price Index for prepackaged software declined approximately 74% from 1997 to 2022. GDP records this as the software industry shrinking. Users experienced software capabilities expanding by orders of magnitude. The BEA's attempt to add unpaid domestic work would expand measured US GDP by 25%. Including ecosystem services would roughly double it. The economy produces abundance. The measurement tool records scarcity. The lossy-compression problem, scaled to civilizational accounting.

The empire-collapse-pattern has seen the precursors. The $300 trillion in phantom wealth McKinsey identified, the $699 trillion derivatives market at 6.4 times global GDP, the Buffett Indicator at three times its historical average: these are the system's attempt to maintain scarcity-based accounting in the face of abundance. Paper claims growing faster than the productive economy can honor them. The same dynamic preceded Rome's collapse, Spain's decline, and Britain's loss of the reserve currency. Every previous cycle ended in collapse into the next scarcity regime. Whether this one follows the same path or creates a transition to abundance depends on the simultaneous nature of the cascade. Previous empires faced abundance in one domain. This cascade produces abundance across all domains at once.

The Reverse Automation Order

AI reverses every previous automation pattern. Mechanization displaced physical labor. Electrification displaced routine manual work. Computerization displaced clerical work. Each wave moved up the skill ladder slowly, from the bottom. AI starts at the top.

Eloundou and colleagues at OpenAI and the University of Pennsylvania, publishing in Science in 2024, found approximately 80% of the US workforce could have at least 10% of their tasks affected by large language models, with higher-income jobs facing the greatest exposure. The IMF found 60% of jobs in advanced economies exposed, with a specific note that AI "challenges the belief that technology affects mainly middle and low-skill jobs." Goldman Sachs estimated 300 million full-time jobs globally facing automation exposure, legal, administrative, and engineering roles first.

The displacement is already measurable. Computer programmer employment fell 27.5% in two years, from approximately 166,000 to 121,200 between 2023 and 2025. Indeed's software engineer postings dropped 35% from January 2020 levels. Microsoft's CEO stated that 30% of company code is now AI-written. Over 50,000 layoffs in 2025 were attributed to AI. CS graduates' unemployment rate exceeded philosophy graduates' for the first time.

Physical labor remains resilient. Construction added 190,000 jobs in 2024. Healthcare support is projected as the fastest-growing employment category through 2034. The World Economic Forum projects farmworkers and delivery drivers among the largest absolute job-growth categories globally by 2030.

A paradox follows. The nations most capable in AI are the most structurally exposed to it. The United States, at 77.8% services GDP, $109 billion in private AI investment, and 73% household-debt-to-GDP, faces maximum disruption. The UK at 73% services faces similar exposure. China at 36.5% industry and South Korea at 32% maintain structural buffers. The countries building the tools of abundance maintain economic structures that abundance dissolves.

Capital Responds

Capital is migrating from bits to atoms. Hyperscaler capital expenditure (Amazon, Microsoft, Google, Meta) grew from $24 billion in 2015 to $211 billion in 2024. Projections: $315-443 billion for 2025, $602 billion for 2026. Seventy-five percent flows to physical AI infrastructure: data centers, GPUs, power systems, cooling. NVIDIA's data center revenue surged from $2.9 billion in fiscal year 2019 to $115 billion in fiscal year 2025. Forty times in six years.

Global clean energy investment crossed $2.2 trillion in 2025, doubling from 2015, at a 2:1 ratio over fossil fuel investment. China invested $800 billion, 4.5% of GDP, more than the US, EU, and UK combined. Private equity infrastructure assets under management quadrupled to $1.3 trillion over the past decade. Infrastructure fundraising surpassed real estate fundraising in 2024 for the first time in recorded history.

Traditional software valuations are being repriced. AI's share of US venture capital surged from 16% in 2021 to 71% in Q1 2025. Public SaaS valuation multiples collapsed from 18-19x EV/Revenue at the 2021 peak to 5.1x in December 2025. Approximately 70% compression. Private software M&A multiples fell from 6.7x to 2.9x.

Hundreds of billions flowing from software margins to physical infrastructure, from financial engineering to energy engineering. The market prices what economic theory has not absorbed: the future is physical, distributed, and energy-intensive. The $699 trillion derivatives market, 6.4 times global GDP, is the last artifact of the compression era's logic. The capital migration says so even when the textbooks do not.

The Fork

Marc Andreessen argues that abundance arrives and the economy adapts. He is right that abundance arrives. He is wrong that it distributes through existing architecture. Every previous wave was captured within a generation. Agriculture, printing, telegraph, radio, internet. The question has never been whether technology produces abundance. The question is where abundance flows.

The cascade guarantees disruption. It does not guarantee direction.

One path: simultaneous deflation makes abundance universal. Energy too cheap to meter. Intelligence available to anyone with a device. Production distributed, local, differentiated. The compression era ends. Something better emerges.

The other path: the same forces concentrate further. AI capabilities accrue to a handful of platform companies. Energy abundance is captured by the entities that own the infrastructure. Production distributes, but whoever controls the coordination layer controls everything. Yanis Varoufakis names this techno-feudalism: cloud capital extracting rent from every transaction, every interaction, every productive act.

Both paths are consistent with the data. The cascade is a forcing function, not a direction. And the forcing is accelerating. China surpassed the United States in absolute R&D spending for the first time in 2024 ($785.9 billion versus $781.8 billion, purchasing power parity). The US maintains higher R&D intensity (3.4% versus 2.6% of GDP). The US is simultaneously the primary creator of AI disruption and the primary target, building the tools that dissolve the service economy constituting 77.8% of its own GDP.

Which future arrives depends on whether the coordination layer distributes along with everything else, or whether it concentrates as production distributes. AI is the first tool in history that can both create abundance and coordinate it. Whether those two functions are built as open protocol or fused into a platform determines which side of the fork civilization takes.

The cascade guarantees the most exciting economic transformation in ten thousand years. It does not guarantee who benefits. The architecture choices of this decade set the trajectory. The next chapter examines why AI sits at the center of that choice and why the transition from shadows to reality determines which path the cascade follows.