A second copy of a PDF costs nothing to make. A second hour of a human lawyer does not. Jeremy Rifkin called the cheap-copy case the "zero marginal cost society" in 2014. AI now puts a similar squeeze on some cognitive work. Energy and housing still have a bill.
Tap to enlargeThe scarcity assumption
AI copies, cheap solar, and open-source tools are pushing the extra unit cost of some goods toward zero. 2026 prices already show the drop in those categories.
Rifkin already had the cheap-copy story in 2014: IoT, renewables, commons. A decade later, AI adds a piece he only half saw. Shipping a file got cheap. Now the production of intelligence, the expensive input in knowledge work, is getting cheaper faster than most cost curves I know.
Three vectors of deflation
The thesis is three cost curves moving at once. Any one of them would bruise a sector. Together they can rearrange a whole economy.
1. The cost of intelligence
For most of industrial history, smart work was expensive. Years of school. Wages that tracked scarcity. A legal memo, a diagnosis, a model of a factory: those needed trained people, and there were only so many.
Generative models broke the symmetry. A training run that costs billions can spit out the millionth answer for a fraction of the first. Contrary Research put the 2024-2025 inference drop at over 75% for comparable tasks. Moore's Law energy, except the output is text and code.
The marginal cost of producing a legal brief, a market analysis, a code module, or a technical summary is approaching the cost of the electricity required to run the inference. The structural implications for any profession built on selling cognitive labor are significant.
Training frontier models still requires billions in capital expenditure. Running them at scale requires data centers, energy, and specialized hardware. AI companies report gross margins of 50 to 60%, far below the 80 to 90% margins of traditional software businesses. The marginal cost is low, not zero.
The economic effect is similar either way. When a task that previously required a $200-per-hour specialist can be completed by a model for a few cents, the market price of that output faces relentless downward pressure. The model is re-pricing the output, whether or not it replaces the specialist.
McKinsey estimates that generative AI may automate 60 to 70% of current work activities. Studies from Wharton and MIT report productivity gains of 15 to 50% for knowledge workers using AI tools. The cost of routine cognitive labor, data summarization, document drafting, basic code generation, administrative policy writing, has fallen so sharply that some researchers describe it as a structural "wage cut" for specific task categories.
2. The cost of energy
The second vector operates on a slower timeline but carries broader implications. Utility-scale solar PV's global weighted-average LCOE fell 90% from 2010 to 2023, from $0.460/kWh to $0.044/kWh, according to IRENA's 2023 costs report. By 2023, that solar LCOE was 56% lower than the weighted-average cheapest fossil-fired option. The 97% and 41% figures that sometimes circulate are not the IRENA table.
The dynamics here are well understood. China's massive investment in clean technology manufacturing has created an overcapacity of solar modules, driving global prices to record lows. N-type cells and bifacial panels have become industry standard, increasing energy yield per square meter. BNEF's 2025 survey put stationary storage pack prices 45% lower than 2024, at $70/kWh. Volume-weighted packs across uses fell 8% to $108/kWh. Solar-plus-storage got cheaper.
Global cumulative installed solar PV capacity surpassed 2,260 gigawatts by the end of 2024 (IEA PVPS). Renewables, led by solar and wind, account for the vast majority of new electricity capacity additions worldwide. The direction of new capacity is settled; the timeline is the open variable.
Energy is the substrate of all economic activity. When the cost of energy declines structurally, the cost floor of everything else declines with it. Manufacturing, transportation, computation, agriculture, water desalination, each of these sectors has energy as a significant input cost. The post-scarcity thesis for energy requires energy to become cheap enough that it is no longer a meaningful constraint on production.
This is already happening in some geographies. In parts of Texas, solar overproduction has driven wholesale electricity prices below zero during peak generation hours. In Chile, Germany, and Australia, similar patterns have emerged. The economics of curtailment, where energy is intentionally wasted because the grid cannot absorb it, point to a system where the problem is shifting from scarcity to management of surplus.
3. The cost of digital goods
The third vector is the oldest and most established. The marginal cost of distributing a digital good, a song, a document, a software package, a video, has been near zero since the broadband era. This is why the music industry collapsed and rebuilt, why journalism is in perpetual crisis, and why open-source software dominates infrastructure.
What AI adds to this existing trend is the production side. Previously, creating a high-quality digital good still required expensive human labor from a software engineer, a designer, a writer, or a musician. AI compresses the production cost alongside the distribution cost. When both production and distribution trend toward zero, the entire value chain of digital goods faces restructuring.
The post-scarcity thesis is strongest where all three vectors converge. A knowledge product, powered by cheap intelligence, running on cheap energy, distributed at near-zero cost. In that intersection, the traditional cost structure of an industry can collapse in years, not decades.
What post-scarcity does not mean
The term "post-scarcity" invites misunderstanding. At its extreme, it implies a world where everything is free and abundant. That world does not exist, and it may never exist. The useful version of the concept is more precise.
Post-scarcity in specific categories. Digital goods, cognitive labor, and energy are trending toward abundance. Physical goods, land, rare minerals, clean water, and human attention remain scarce. The economy is bifurcating into categories of abundance and categories of persistent scarcity.
New forms of scarcity emerge. As AI makes cognitive output abundant, the scarce resource shifts. In 2026, the most valuable inputs are proprietary data (the context that general models lack), physical execution capability (robotics, manufacturing, logistics), and human judgment in novel situations. Scarcity migrates.
Infrastructure concentrates. Even if the marginal cost of an AI query trends toward zero, the fixed cost of building the infrastructure to run that query is enormous. Data centers, GPU clusters, energy contracts, and training pipelines require billions in capital. The economics of post-scarcity at the consumer level may coexist with extreme concentration of infrastructure at the producer level. A few entities may control the "utilities" of abundance.
Distribution is political, not automatic. The most persistent critique of post-scarcity optimism is that abundance does not automatically translate to access. The United States produces enough food to feed its population several times over, yet food insecurity persists. Technology makes abundance possible; institutions determine whether it is shared.
The gap between technological abundance and equitable distribution is a governance problem. Every prior wave of cost reduction, from the printing press to the internet, generated immense surplus value. Who captured it was the live question.
The macroeconomic paradox
Central banks, the institutions most responsible for managing modern economies, are built for scarcity. Their primary tools, interest rates, money supply adjustments, and inflation targeting, assume an economy where demand can outstrip supply, where wages can drive prices upward, and where the normal state of affairs involves managing cycles of growth and contraction within a framework of finite resources.
AI-driven deflation presents a category problem for these institutions. When the cost of cognitive labor drops by 75% in a year, that is deflationary. It is "good deflation," the kind that improves living standards by making goods and services more affordable, rather than the demand-deficient kind (recession, unemployment spirals) that central banks fear.
The problem is that existing monetary policy frameworks do not clearly distinguish between these two types. A sustained period of falling prices, regardless of cause, can trigger policy responses designed for recession (lower interest rates, quantitative easing) that may be inappropriate for a technology-driven cost reduction.
Japan's experience offers a partial case study. Three decades of deflationary pressure, driven in part by demographic decline and efficiency gains, challenged the Bank of Japan's ability to stimulate growth using conventional tools. The AI era may generalize this dynamic globally, not from demographics, but from technology. Economists are beginning to frame this as a structural shift that requires new macroeconomic thinking rather than new policy tools alone.
The wage-consumption question
Classical economics assumes a feedback loop. Workers produce goods. They earn wages. They spend wages on goods. Their spending creates demand. That demand creates more production, more jobs, more wages. This loop has sustained market economies for two centuries.
AI introduces a potential break in this loop. If machines perform a growing share of productive labor, the mechanism for distributing purchasing power (wages) weakens. Output may continue to grow, but income may not keep pace. The result is a demand problem that looks different from a traditional recession.
Keynes anticipated it in 1930 with his essay "Economic Possibilities for our Grandchildren," predicting that by 2030, the "economic problem" of scarcity would be largely solved, replaced by the challenge of how to use the resulting leisure. He did not foresee that the transition period would be the most disruptive part.
Several proposed mechanisms exist for addressing this structural gap.
Universal Basic Income (UBI). Direct cash transfers to all citizens, funded by taxation of productivity gains. Finland, Kenya, and several U.S. municipalities have conducted trials with mixed but generally positive results on well-being, though scalability and fiscal sustainability remain debated.
Universal Basic Compute. A more recent proposal that suggests distributing access to AI compute as a public utility, rather than distributing cash. The logic is that if AI is the new means of production, access to that production capacity is more valuable than a fixed income transfer. Sam Altman and others in the AI industry have proposed variations of this concept.
Stakeholder models. Restructuring corporate governance to distribute productivity gains more broadly, through profit-sharing, equity ownership, or cooperative structures, rather than concentrating them in capital returns to shareholders.
Shortened work weeks. Using productivity gains to reduce labor requirements rather than reduce headcount, distributing the same output across fewer working hours per person.
The industrial revolution answered "how does one produce enough?" The live question is how to distribute the value of abundance when wages for labor no longer scale with output.
Abundance infrastructure
If the production of intelligence, energy, and digital goods is trending toward abundance, a new category of strategic investment emerges. Rather than investing in the goods themselves (which are becoming commodities), the opportunity may lie in the infrastructure that enables and manages abundance.
Energy storage and grid management. As solar overproduction creates periodic surplus, the value migrates from generation to storage and distribution. The ability to store energy cheaply and distribute it efficiently becomes the scarce capability in an energy-abundant world.
AI orchestration and safety. As baseline intelligence becomes commoditized, the value migrates to orchestration (managing multiple AI systems to achieve complex goals), safety (ensuring AI systems operate within acceptable parameters), and domain-specific context (the proprietary data and workflows that general models lack).
Physical execution. Robotics, advanced manufacturing, and logistics remain tied to physical geography, material science, and engineering constraints that software cannot bypass. The "latency bound" of physical execution, the time it takes to move atoms rather than bits, creates a category of persistent scarcity even in an otherwise abundant economy.
Verification and trust. In a world where AI can generate infinite content, code, and analysis, the ability to verify authenticity, accuracy, and provenance becomes increasingly valuable. Cryptographic provenance, human attestation frameworks, and audit systems represent a growth category built on the need to manage abundance rather than create it.
The bifurcated economy
The most likely near-term outcome is a bifurcated economy. Certain sectors, primarily digital goods, cognitive services, and energy, may experience deflation so persistent that their pricing models fundamentally change. Other sectors, primarily physical goods, real estate, healthcare delivery, and anything requiring human presence or rare materials, may continue to operate under scarcity dynamics.
This bifurcation creates unusual dynamics. A person might access frontier AI tutoring for free while being unable to afford housing. A company might generate sophisticated market analysis at near-zero cost while facing rising costs for raw materials and logistics. The "post-scarcity" label applies unevenly, and the economic stress falls disproportionately on those whose livelihoods depend on the sectors experiencing deflation.
The transition to partial post-scarcity is asymmetric. Capital flows toward abundance infrastructure (energy storage, AI compute, verification). Labor value concentrates in execution, judgment, and relationship management. The sectors experiencing deflation shed traditional jobs. The sectors experiencing persistent scarcity face cost inflation from demand pressure. Managing this asymmetry may define economic policy for the next generation.
What history suggests
Every major deflationary wave in history, the printing press, the steam engine, electrification, the internet, followed a similar pattern: an initial period of disruption and displacement, a middle period of institutional adaptation, then a long-term expansion of total wealth and well-being.
The printing press destroyed the economics of manuscript copying but created publishing, journalism, and mass literacy. The steam engine displaced artisanal manufacturing but created the industrial middle class. The internet collapsed the economics of physical media distribution but created the digital economy.
In each case the transition was painful, uneven, and politically contested, the new equilibrium generated more total value than the old one, and the key variable was the institutional response: the laws, regulations, social contracts, and governance structures that determined how the new abundance was distributed.
The post-scarcity technology thesis is an observation that specific, measurable cost functions in intelligence, energy, and digital goods are declining fast enough to challenge the institutional frameworks designed around their scarcity. The evidence suggests the trends continue. Whether the institutions responsible for managing economic life can adapt at a matching pace is the open question.
Texas wholesale power has already gone below zero in solar overproduction hours. Chile, Germany, and Australia show the same surplus-management problem. Japan's three decades of deflation showed that conventional stimulus does not fix a cost-drop. Fiscal design has to follow the surplus, which now sits in data centers and energy contracts, before the wage tax base finishes shrinking.