---
title: 'The Missing $Quadrillion'
subtitle: 'What Maxwell''s Demon Was Trying to Tell Us for 158 Years—and the Economic Channel That Every Major AI Forecast Has Missed'
slug: 'missing-quadrillion'
date: 2026-02-18
type: 'essay'
status: 'published'
tags: ['enviroai', 'thermodynamics', 'information-theory', 'landauer', 'paper', 'maxwell']
abstract: 'Identifies a second economic channel that every major AI-impact forecast (Goldman, McKinsey, PwC) has missed: the bond-bit asymmetry. Channel A asks what happens when AI substitutes for cognitive labor; Channel B asks what happens when information substitutes for physical manipulation across the entire material economy. The second channel is roughly twice the size of the first and reframes the path to a $1-quadrillion economy.'
license: 'CC-BY-4.0'
author: 'Jed Anderson'
co_authors: []
canonical_url: 'https://jedanderson.org/essays/missing-quadrillion'
pdf: '/pdfs/missing-quadrillion.pdf'
hero_image: '/images/missing-quadrillion-hero.png'
hero_image_alt: 'First page of The Missing $Quadrillion'
supporting_files: []
---

The Missing $Quadrillion What Maxwell's Demon Was Trying to Tell Us for 158 Years—and the Economic Channel

That Every Major AI Forecast Has Missed A First-Principles Economic Analysis Jed Anderson / EnviroAI, February 2026

The Thesis Every major economic projection of AI's impact on global GDP . . . from

Goldman Sachs (~$7T), McKinsey ($13T in 2018; $2.6–4.4T/year for GenAI in 2023), and PwC ($15.7T) . . . overwhelmingly models the same phenomenon: the value of substituting intelligence for human cognitive labor, plus downstream consumption effects from AI-enhanced products.

They are measuring one channel. There are two.

The second channel—revealed by a measured asymmetry in the structure of reality called the Bond-Bit

Asymmetry—is approximately twice the size of the first. No major economic forecaster has built a framework capable of fully capturing it.

Channel A asks: What happens when AI can do what humans do?

Channel B asks: What happens when information replaces force across the entire material economy?

These are fundamentally different questions. Channel A substitutes intelligence for cognitive labor. Channel B substitutes information for physical manipulation—whether that means preventing waste, transforming matter, or discovering entirely new configurations of reality.

The blueprint for Channel B has been sitting in a physics thought experiment since 1867. For

158 years, we thought Maxwell's Demon was a paradox about thermodynamics. It was a design pattern for abundance. We just didn't notice.

The difference between one channel and two is tens of trillions of dollars per year in uncounted economic value—and a fundamentally different answer to the question of how fast civilization reaches a $1 quadrillion economy.

## Part I: Two Floors

Everything that follows rests on two measured numbers. Both are experimentally verified.

Neither is debatable.

The Floor of Knowing In 1961, Rolf Landauer proved that erasing one bit of information requires a minimum energy dissipation of:

E_bit = k_B · T · ln(2)

At room temperature (T = 300K):

E_bit = (1.381 × 10⁻²³ J/K) × (300 K) × (0.693) = 2.87 × 10⁻²¹ joules per bit This is not an engineering estimate. It is a consequence of the Second Law of Thermodynamics.

No technology, no matter how advanced, can process information for less energy than this.

Strictly, Landauer's bound applies to logically irreversible operations—erasure of a bit.

Measurement itself can in principle be reversible (Bennett, 1973), but any cyclic informationprocessing system must eventually erase its memory to accept new data, paying the Landauer cost per cycle. For any continuously operating sensor-and-compute system, k_BT·ln(2) per bit per cycle is the irreducible thermodynamic floor. In 2012, Bérut et al. (Nature) verified

Landauer's limit directly by measuring heat dissipation from erasing a single bit stored in a colloidal particle. The measured value approached k_BT·ln(2) in the slow-erasure limit.

The Floor of Moving The energy required to break a single carbon-hydrogen bond is approximately:

E_bond ≈ 413 kJ/mol = 6.86 × 10⁻¹⁹ joules per bond This value derives from quantum mechanics—specifically from the fine-structure constant (α ≈

1/137) and electron mass, which together determine all chemical bond energies. It has been measured to high precision for over a century (CRC Handbook of Chemistry and Physics). The energy required to break a C-H bond in 2025 is identical to what it was in 1900 and will be in
3000. These are fundamental constants of nature.
The Ratio E_bond / E_bit = (6.86 × 10⁻¹⁹ J) / (2.87 × 10⁻²¹ J) ≈ 239 At the molecular level, moving one bond costs approximately 240 times more energy than knowing one bit at the thermodynamic limit.

This is the Bond-Bit Asymmetry at the atomic scale, derived from measured physical constants.

The Landauer limit scales with temperature (E_bit = k_BT·ln2), but chemical bond energies are fixed by the fine-structure constant regardless of temperature. At any temperature where liquidphase chemistry operates—the regime relevant to all industrial activity and all biology—the peroperation ratio holds at roughly 200–250×. This ratio is a structural feature of the universe's electromagnetic physics, not an engineering parameter.

But 240× drastically understates the macroscopic reality.

## Part II: The Twenty Orders of Magnitude

From Molecules to Kilograms The atomic ratio of ~240 is the per-operation asymmetry. In real physical systems, the leverage explodes because of a fundamental feature of nature: information compresses. You do not need to know the position of every molecule to prevent a catastrophe. You need macro-state information—a few billion bits about valve degradation—that prevents micro-state disaster involving trillions of trillions of molecular bonds.

Consider a practical scenario:

Moving: A storage tank valve fails. One kilogram of hydrocarbon disperses into soil and groundwater. Full molecular reconfiguration—breaking and reforming bonds across the contaminated mass—establishes the thermodynamic floor for physical restoration.

• Molecular weight of CH₂ unit: ~14 g/mol

• Moles in 1 kg: 1000/14 ≈ 71.4 mol

• Bonds per CH₂ unit: ~3 (C-C backbone + C-H)

• Total bonds: 71.4 × (6.022 × 10²³) × 3 ≈ 1.29 × 10²⁶ bonds

• Energy: 1.29 × 10²⁶ × 6.86 × 10⁻¹⁹ J ≈ 8.9 × 10⁷ joules

Knowing: A sensor detects micro-vibrations indicating valve degradation. The system processes data and triggers valve closure before failure.

• Sensor data + analysis computation: ~10⁹ bits processed

• Energy at Landauer limit: 10⁹ × 2.87 × 10⁻²¹ J = 2.87 × 10⁻¹² joules

The ratio of thermodynamic floors: (8.9 × 10⁷ J) / (2.87 × 10⁻¹² J) ≈ 3.1 × 10¹⁹ ≈ 10²⁰

Twenty orders of magnitude. One hundred quintillion to one.

What this ratio measures: The ~10²⁰ is not the ratio of two fundamental constants (that is

~240). It is the ratio of two thermodynamic floors—the minimum energy to physically reconfigure a kilogram of dispersed matter versus the minimum energy to computationally process the information that prevents the dispersal. The enormous gap arises because macro-state information (valve status, flow dynamics, vibration signatures) compresses by a factor of ~10¹⁷ relative to the micro-state physical reconfiguration (10²⁶ bonds). This compression is not arbitrary; it reflects the mathematical structure of physical systems, where boundary measurements can characterize volumetric states (established by PDE observability theory,

Bardos-Lebeau-Rauch 1992) and sparse signals can be reconstructed from far fewer samples than classical theory requires (compressed sensing, Candès-Tao-Romberg-Donoho, 2004–2006).

The actuation gap: A complete accounting must include the energy required to physically close the valve—the mechanical work of actuation. For a typical industrial valve, this is on the order of

1–100 joules. Including actuation, the full prevention cost at the Landauer limit is ~10⁰ to 10² joules, and the ratio becomes:

(8.9 × 10⁷ J) / (10² J) ≈ 10⁵ to 10⁷ Even with actuation energy included, knowing where to act—and acting—is one million to one hundred million times cheaper than physical restoration after failure. And only the computation component of this ratio improves over time. The actuation energy is already negligible; the remediation energy is fixed by quantum mechanics.

At the Landauer limit: The ratio of thermodynamic floors (computation-only) is ~10²⁰. This represents the ultimate ceiling on how favorable information becomes relative to physical reconfiguration as computation approaches fundamental limits.

Even Today Current computers operate at approximately 10⁻¹² joules per operation—roughly 10⁹ times above the Landauer limit. Even at today's computational efficiency:

(8.9 × 10⁷ J) / (2.87 × 10⁻³ J) ≈ 3.1 × 10¹⁰ Knowing is already ten billion times cheaper than moving. And this ratio improves every year, because computation gets cheaper while chemistry does not.

There is no Moore's Law for the fine-structure constant.

## Part III: The Blueprint Hidden in a Paradox

What Maxwell's Demon Actually Did In 1867, James Clerk Maxwell imagined a tiny being . . . a "demon" . . . that could observe individual gas molecules and selectively open a door between two chambers, sorting fast molecules from slow ones.

For 158 years, physics treated this as a paradox about the Second Law of Thermodynamics. Does the demon violate it? (No, Bennett showed in

1982 that the demon must eventually erase its memory, paying the Landauer cost.)

But in solving the paradox, we missed what the demon was doing.

The demon started with a gas at uniform temperature. It ended with a temperature gradient—hot on one side, cold on the other. A new configuration that did not previously exist.

The demon did not prevent anything from scattering. The demon TRANSFORMED the system.

It took matter in one configuration and navigated it to a different configuration using information instead of force. It created order that did not previously exist. It assembled a new state.

Maxwell's Demon was never about guarding. It was always about building. For 158 years, we focused on the paradox and missed the blueprint.

Sagawa-Ueda: The Proof of Principle In 2008–2012, Takahiro Sagawa and Masahito Ueda at the University of Tokyo derived a generalized second law of thermodynamics that makes this rigorous:

W_ext ≤ −ΔF + k_BT · I The maximum work extractable from any thermodynamic process equals the free energy change plus k_BT times the mutual information gained through measurement. Information acts as thermodynamic fuel.

This was experimentally verified by Toyabe et al. (Nature Physics, 2010) and Koski et al.

(PNAS, 2014), who extracted work at 90% of the theoretical maximum from a single-electron

Szilard engine.

A note on scope and mechanism: The Sagawa-Ueda equality operates rigorously in the microscopic regime where thermal fluctuations dominate. The direct thermodynamic work extractable via k_BT · I for a macroscopic quantity of information (say, 10⁹ bits) is approximately 10⁻¹² joules—negligible at industrial scales. One cannot power a valve or a supply chain on 10⁻¹² joules.

The significance of Sagawa-Ueda for the macroscopic economy is not that AI literally converts

Shannon entropy into mechanical work via the Jarzynski equality. It is that the equation establishes a thermodynamic proof of principle: information and free energy are fungible at the fundamental level. This principle scales to macroscopic systems through classical mechanisms—algorithmic optimization (finding shorter routes, optimal chemical yields), predictive modeling

(preventing waste before it occurs), and configuration-space navigation (searching molecular space computationally rather than physically). The microscopic equation proves the underlying physics is real; the macroscopic leverage operates through these classical amplification channels.

Read the equation carefully. It does not say "for prevention." It says: for any thermodynamic process where information is available, that information substitutes for free energy.

This means the Bond-Bit Asymmetry applies—through both the direct thermodynamic channel and the vastly larger classical optimization channel—to:

• Prevention—keeping a system at its current configuration ("stay here")

• Transformation—moving a system to a new configuration ("go there")

• Discovery—finding the right configuration in a vast possibility space ("find where to

go")

These are not three different phenomena. They are three expressions of one thermodynamic fact: the universe charges enormously more to navigate reality with force than with information. The

~10²⁰ at the Landauer limit is the floor. For complex systems, the leverage grows without bound.

The demon was never just a guard. It was always a builder. And the blueprint applies to everything.

The Configuration Space Ceiling The 10²⁰ ratio is the floor. For complex problems involving vast configuration spaces, the leverage is astronomically higher.

A 100-residue protein has 20¹⁰⁰ ≈ 10¹³⁰ possible sequences. To find a specific functional sequence by blind physical synthesis:

• Sequences to search (birthday-bound): ~10⁶⁵

• Energy per synthesis: ~100 peptide bonds × ~10⁻¹⁸ J ≈ 10⁻¹⁶ J

• Total blind-search energy: 10⁶⁵ × 10⁻¹⁶ J = 10⁴⁹ joules

• (For scale: the Sun outputs ~4 × 10²⁶ joules per second)

To specify the same sequence informationally and synthesize it once:

• Information to specify: log₂(20¹⁰⁰) = 432 bits → 1.24 × 10⁻¹⁸ J at Landauer limit

• One physical synthesis: ~10⁻¹⁶ J

• Total: ~10⁻¹⁶ joules

Ratio: 10⁶⁵.

Not 10²⁰. 10⁶⁵. The leverage grows with the size of the space being searched. For drug-like chemical space (~10⁶⁰ molecules), the ratio reaches approximately 10⁶⁰. For practical materials discovery, 10²⁰ to 10⁴⁰.

The 10²⁰ prevention leverage is the floor. There is no ceiling.

## Part IV: What Everyone Sees: Channel A

Current Global GDP Global nominal GDP in 2025: approximately $117.2 trillion (IMF World Economic Outlook,

October 2025).

The Value of Human Labor Global labor compensation as a share of GDP: approximately 52.4% (ILO, 2024), yielding direct employee compensation of roughly $61 trillion. With self-employment, informal economy, and imputed unpaid labor: approximately $80–100 trillion addressable.

This is what every major AI GDP forecast measures:

Source AI GDP Impact Timeframe Mechanism Goldman Sachs Over ~10-year +7% of GDP ($7T) Labor productivity

(2023) diffusion +$2.6–4.4T/year Task automation across use McKinsey (2023) By 2040

(gen AI) cases +$13T additional McKinsey (2018) By 2030 Broad AI adoption activity

+$15.7T (14% of $6.6T productivity + $9.1T PwC (2017) By 2030 GDP) consumption Acemoglu/MIT

+~1% of GDP Over 10 years Conservative task exposure (2024)

Serious estimates from serious institutions. All measuring the same fundamental question: what happens when machines perform cognitive tasks currently done by humans.

What These Forecasts Include and What They Miss A fair assessment requires examining what these forecasts actually model. PwC's $15.7T comprises $6.6T in labor productivity gains (automating processes, augmenting workforce) and

$9.1T in consumption-side effects (increased consumer demand for AI-enhanced, personalized, higher-quality products). McKinsey's GenAI report models task automation across existing workflows, including some R&D acceleration.

These forecasts do capture incremental physical optimization—manufacturing defect reduction, logistics routing, predictive maintenance. They model these as productivity improvements within existing economic activity, using linear economic multipliers.

What they do not model—and structurally cannot model within a task-substitution framework—is the thermodynamic phase transition: the replacement of brute-force physical search with computational navigation across vast configuration spaces. Making R&D 20% faster is Channel

A. Accessing 10⁵⁴ more of chemical space is Channel B. The first is an efficiency improvement within existing process. The second is a change in what is physically possible at the interface of information and matter.

No major economic forecast models the Bond-Bit Asymmetry, the configuration-space leverage ratios, or the systematic repricing of physical uncertainty that these imply. The channel they undercount is the subject of the next section.

## Part V: What Gets Undercounted: Channel B

The Tax on Ignorance The global material economy—manufacturing, energy, agriculture, construction, logistics, extraction—represents approximately $55–75 trillion of GDP. These sectors do not merely process information. They move, transform, and configure matter.

Channel B asks: how much value does the material economy currently destroy or fail to create because it navigates reality with force instead of information?

The tax comes in two forms.

The Waste Tax: Value Destroyed Costs the global economy currently bears because it lacks sufficient information to prevent them:

Information Deficit Annual Cost Source Notes Welfare-cost basis (VSL Environmental externalities Lancet Commission

$4.6T methodology); see caveat (pollution) (2017; 2022 update) below Food waste ~$1T FAO (2023)

Material waste World Bank; industry (manufacturing, construction, $2.5–3T analyses mining)

IEA; primary Energy waste (addressable $1.5–2.5T conversion + end-use Adjusted; see caveat below fraction) losses

WHO; National Preventable healthcare costs $2.5–3T Academy of (~30% of ~$9T global spend)

Medicine Logistics inefficiency (~20– 30% of $9–10T logistics $1.8–3T Industry analyses market)

Insurance risk premiums from Swiss Re; industry $1–2T uncertainty data Reactive infrastructure

$1–2T Industry studies maintenance Regulatory compliance $1–2T Industry estimates friction

Waste Tax Total $16.3– 24T/year Central estimate ~$20T/year Critical accounting notes:

On the Lancet pollution figure: The $4.6T is calculated using the Value of Statistical Life

(VSL) methodology—it represents welfare losses (what society would pay to avoid these deaths), not direct nominal GDP. Eliminating pollution prevents 9 million premature deaths per year and their associated suffering, healthcare costs, and lost productivity, but the welfare figure is not directly additive to nominal GDP. The direct economic costs (healthcare expenditure, lost labor productivity, crop damage) are a subset of this figure, estimated at 1.3–2% of GDP in affected countries. We retain the $4.6T welfare figure because it represents the truest measure of value destroyed, while noting that its relationship to measured GDP is indirect. If one substitutes the direct economic cost (~$2T), the Waste Tax total falls to approximately $15–20T, with a central estimate of ~$17T.

On energy waste: Global primary energy waste is approximately 60–67% of total primary energy consumed. However, a substantial portion of this waste is thermodynamically irreducible—dictated by the Carnot limit (η ≤ 1 − T_c/T_h) for any heat engine. No amount of information can make a coal plant or internal combustion engine 100% efficient. The addressable fraction—waste attributable to suboptimal operation, poor load matching, transmission losses, and processes where information could improve efficiency—is estimated at approximately 30–

50% of total energy waste, or roughly $1.5–2.5T of the $4–5T in total wasted energy value. This revised figure replaces the earlier $2–3T estimate and is more conservative.

On double-counting and overlap: These categories are not independent. Pollution contributes to healthcare costs. Food waste contributes to pollution. Energy waste drives pollution. Some logistics inefficiency is an energy waste problem. Summing these as fully independent buckets overstates the total. The range of $16–24T attempts to account for this overlap at the low end, but an honest assessment is that cross-category correlation could reduce the independent total by

15–30%. We present the components to establish that the magnitude of the material-economy inefficiency wedge is plausibly enormous—measured in tens of trillions—without claiming precision on the exact sum.

The Ignorance Tax: Value Never Created The Waste Tax captures value destroyed. But the Demon was always building, not just guarding—which means the asymmetry also implies value that was never created because configuration spaces were too vast to search by force.

Global R&D spending: $2.87 trillion/year (WIPO, 2024). This is what civilization spends annually searching configuration space—finding the right molecular configurations for drugs, the right material compositions for engineering, the right process parameters for manufacturing.

Most of this spending pays the tax on brute-force search.

The signatures of that tax are unmistakable:

Drug discovery: $2.6–2.8 billion per approved drug (Tufts Center). 10–15 year timelines. ~90% clinical failure rate. Eroom's Law: costs doubling every nine years since 1950. The economic fingerprint of searching 10⁶ molecules out of 10⁶⁰ possible—

0.0000000000000000000000000000000000000000000000000000001% of the space . . . and hoping.

Early evidence of the Demon building:

• AI compresses preclinical drug discovery timelines by 25–70% (multiple sources, 2024–

2025)

• Exscientia: 70% reduction in discovery time, 80% cost reduction for molecules entering

clinical trials

• Insilico Medicine: drug candidate in 18 months vs. traditional 4–5 years

• Caveat: No AI-discovered drug has yet achieved FDA approval as of late 2025

The same physics applies across materials science, protein engineering, chemical processes, agricultural genetics, and battery technology—every domain where the fundamental task is navigating vast configuration space.

Bounding the Ignorance Tax: If AI makes existing R&D 3–5× more productive across domains, and $1 of R&D creates $5–8 in eventual GDP (established R&D elasticity studies), then $2.87T in annual R&D at 3–5× productivity yields $3–8T/year in additional value creation near-term, growing to $8–15T/year as discovery tools mature and configuration-space navigation becomes routine. The ceiling is defined by the transformation leverage itself: as the cost of knowing what to make approaches Landauer, only the cost of physically making it remains.

The Unified Channel Component Annual Value Nature Recoverable as information infrastructure

Waste Tax (value destroyed) $15–24T/year deploys Ignorance Tax (value never $3–15T/year, Unlockable as configuration-space search created) growing improves

Total Channel B $18–39T/year Central estimate ~$25T/year For comparison: Channel A, the entire labor-substitution and consumption-enhancement opportunity that Goldman Sachs, McKinsey, and PwC are measuring, is $7–15 trillion/year by

2030–2040.

The channel that gets systematically undercounted is approximately twice the size of the channel everyone counts.

## Part VI: Why It Gets Undercounted

Conventional AI economic analyses undercount Channel B for a precise structural reason: they model AI as a substitute for human tasks within existing economic activity.

The methodology is consistent across every major forecast: decompose the economy into tasks

→ estimate which tasks AI can perform → calculate the labor-cost savings (and, in PwC's case, the consumption-side demand effects from enhanced products).

This framework captures Channel A well. It captures incremental physical optimization—predictive maintenance reducing downtime, logistics routing saving fuel—as productivity improvements within existing sectors.

It systematically undercounts Channel B for two reasons:

It cannot fully price the Waste Tax because waste elimination is not a task to be automated—it is the absence of tasks made possible by information. You don't make remediation faster. You make it unnecessary. Standard economic models capture "reduced downtime" but not the thermodynamic repricing of physical uncertainty at 10²⁰-to-1 leverage ratios.

It cannot see the Ignorance Tax because discovery at the configuration-space frontier is not about doing existing research more efficiently . . . it is about searching spaces that were physically inaccessible. You don't screen drugs 2× faster. You access 10⁵⁴ more of chemical space. Tasksubstitution models capture the 2× improvement. They cannot capture the 10⁵⁴ expansion.

Both blind spots arise from treating AI as a productivity tool within the existing economy, rather than as a technology that changes what is physically possible at the interface of information and matter.

The Bond-Bit Asymmetry has been derivable from known physics since at least 2012. But the technology stack needed to exploit it—$1 sensors, $0.001 inference, LLM-powered reasoning, automated actuation—became available simultaneously between 2020 and 2025. The physics did not change. The lens through which we could see it arrived.

And Maxwell's Demon had been pointing to it since 1867. We were so focused on whether it violated the Second Law that we never noticed what it was building.

## Part VII: The Asymmetry That Only Grows

Channel A is bounded. There is a finite amount of human cognitive labor ($80–100T addressable). Once AI can perform all of it, Channel A saturates.

Channel B has no equivalent ceiling, because it is powered by a ratio between two physical floors . . . one that falls and one that is fixed.

Computation gets cheaper. Koomey's Law: computational energy efficiency doubles approximately every 2.3 years. Current computers operate ~10⁹× above the Landauer limit. The floor of knowing has nine orders of magnitude of room to fall.

Chemistry does not. The C-H bond energy of 6.86 × 10⁻¹⁹ joules is determined by the finestructure constant and electron mass. These are properties of the universe. They cannot be engineered, improved, or negotiated with.

The gap between the cost of navigating reality with information and the cost of navigating reality with force grows every year. The curves diverge monotonically. They can never converge.

There is no Moore's Law for chemistry. There is no Moore's Law for the fine-structure constant.

There is no Moore's Law for the cost of being wrong.

## Part VIII: The Path to One Quadrillion Dollars

What the Missing Channel Means for GDP Start: $117 trillion (2025, nominal). Target: $1,000 trillion ($1 quadrillion).

Scenario 1: No AI (Baseline): Nominal growth ~5.5%/year (3% real + 2.5% inflation). $1

Quadrillion GDP arrives ~2065.

Scenario 2: Channel A Only (What Everyone Projects): AI labor substitution adds ~1–2% to real growth, phasing in over time. Nominal growth ~7%. $1 Quadrillion GDP arrives ~2057.

Eight years earlier.

Scenario 3: Both Channels (What the Physics Reveals):

Nominal Channel A Channel B Starting Ending Period Growth Contribution Contribution GDP GDP

6.7% +0.7% +0.5% $117T $162T 8.3% +1.2% +1.6% $162T $242T 10.0% +1.5% +3.0% $242T $628T

11.8% +1.8% +4.5% $628T $1,114T $1 Quadrillion GDP arrives ~2049. Sixteen years earlier than baseline. Eight years earlier than

Channel-A-only projections.

Scenario $1Q Arrival Acceleration vs. Baseline No AI ~2065—Channel A only ~2057 8 years earlier

Channels A + B ~2049 16 years earlier The Missing Quadrillion is not a metaphor. It is the literal difference between an economy that counts one channel of AI-driven value creation and one that counts both. The channel no one is fully measuring accelerates the arrival of the $1 quadrillion economy by approximately a decade.

Growth Rates in Historical Context The 10%+ nominal growth rates in Scenario 3 are high by historical standards. Sustained rates at this level would be unprecedented for the global economy. This is justified by the unprecedented nature of the event: the simultaneous activation of two distinct value channels, one of which is powered by a thermodynamic asymmetry that widens over time. But reasonable analysts could argue for more conservative adoption curves, which would delay—but not prevent—the arrival of the quadrillion-dollar economy.

The $1Q figure is nominal. In 2025 real dollars, ~$1Q nominal in ~2049 represents roughly

$500–600T in real output—a 4.5–5× real increase. Extraordinary, but supported by two-channel compounding.

## Part IX: Caveats and Intellectual Honesty

## 1. The Waste Tax figures carry meaningful uncertainty. The Lancet Commission's

$4.6T is a welfare-cost estimate calculated using the Value of Statistical Life (VSL) methodology with 2015 data; it is the most rigorous available figure but measures welfare losses rather than direct GDP impact. The direct economic costs of pollution (healthcare spending, lost productivity, crop damage) are a smaller subset. Energy waste, healthcare waste, and logistics waste percentages are central estimates from published ranges. The energy waste figure has been adjusted downward to exclude thermodynamically irreducible waste governed by the Carnot limit. The $15–24T range reflects this uncertainty honestly.

## 2. The Ignorance Tax is inherently harder to quantify. It involves bounding the value of

things that do not yet exist. The $3–15T range reflects genuine uncertainty. The underlying physics (configuration-space leverage) is on firm ground; the economic translation is approximate.

## 3. No AI-discovered drug has achieved FDA approval as of late 2025. If AI fails to

improve clinical success rates, not just preclinical timelines, the near-term pharma contribution to the Ignorance Tax will be smaller than estimated. The physics is indifferent to current institutional bottlenecks, but timelines are not.

## 4. Channel overlap exists. Discovery improvements (better catalysts) reduce waste. Labor

automation (Channel A) enables Channel B deployment. The channels are not perfectly additive. The central estimate of ~$25T/year for Channel B may include modest overlap with Channel A projections.

## 5. The ~10²⁰ ratio represents Landauer-limit computation compared to full molecular

reconfiguration. Current computers operate ~10⁹× above Landauer. Today's practical macroscopic ratio (computation to remediation) is ~10¹⁰ . . . still enormously favorable to information, but nine orders of magnitude below the theoretical ceiling. Additionally, the full prevention pathway includes actuation energy (mechanical work to close a valve, redirect flow, etc.), which is on the order of 1–100 joules—negligible compared to remediation but not zero. With actuation included, the current practical ratio is ~10⁵ to

10⁷, improving to ~10¹⁸ at the Landauer limit.

## 6. Sagawa-Ueda and the microscopic-to-macroscopic bridge. The Sagawa-Ueda

generalized second law rigorously proves that information substitutes for free energy in thermodynamic processes. The direct thermodynamic work extractable from macroscopic quantities of information (k_BT · I) is negligible at industrial scales. Macroscopic leverage operates through classical optimization channels—prediction, prevention, and configuration-space navigation—not through direct information-to-work conversion.

Sagawa-Ueda establishes the thermodynamic proof of principle; classical AI implements it at scale.

## 7. Major economic forecasters are not blind to physical optimization. PwC, McKinsey,

and others include some physical efficiency gains in their models—predictive maintenance, logistics optimization, manufacturing quality improvements. What their frameworks structurally undercount is the magnitude of thermodynamic leverage and the configuration-space frontier. They model incremental improvements within existing economic activity using linear multipliers. Channel B as described here—the systematic repricing of physical uncertainty at leverage ratios of 10²⁰ or more—is not captured by task-substitution methodologies.

## 8. The GDP projections are scenarios, not predictions. The physics (Bond-Bit

Asymmetry, configuration-space leverage) is experimentally verified and not in question.

The economic translation—how fast institutions, markets, and societies convert theoretical leverage into captured value—is where uncertainty lives. Rebound effects

(efficiency gains translating into increased consumption rather than net resource reduction) could partially offset waste-reduction benefits, as documented in the energy economics literature. Adoption constraints, regulatory friction, and complementary capital requirements all affect the speed of Channel B realization.

## Conclusion

The tech world sees AI as a substitute for labor. The physics sees something far larger.

There are two channels through which intelligence creates economic value. Channel A, labor substitution and consumption enhancement, is the one everyone measures. It is real and significant: $7–15 trillion per year by 2030–2040, according to the best available estimates.

Channel B, the material leverage of information over force, is approximately twice as large. It captures the ~$25 trillion per year the global economy currently loses to the Waste Tax (value destroyed by navigating with force instead of information) and the Ignorance Tax (value never created because configuration spaces are too vast to search by force).

Channel B rests on the Bond-Bit Asymmetry: the measured fact that the thermodynamic floor for physically reconfiguring matter exceeds the thermodynamic floor for computationally processing information by a factor of ~10²⁰ for typical macroscopic scenarios—a ratio driven by the exponential compression of macro-state information relative to micro-state physical reality. This asymmetry is measured physics. The economic translation—how much of the material economy's waste and unexplored configuration space can be captured, and how fast—is modeldependent and carries genuine uncertainty. But the direction is unambiguous and the magnitude is bounded by published data from the Lancet, IEA, FAO, WHO, and WIPO: the materialeconomy inefficiency wedge is measured in tens of trillions of dollars per year. The SagawaUeda framework proves the underlying principle that information and free energy are fungible; classical optimization channels amplify this principle to industrial scales across prevention, transformation, and discovery. And the asymmetry grows over time because computation gets cheaper and chemistry does not.

And Maxwell's Demon was pointing to all of it for 158 years. We thought it was a paradox about entropy. It was a blueprint for abundance. The demon was never just guarding. It was always building . . . using information to navigate matter through configuration space at a fraction of the cost of force.

That is the missing channel. That is the missing quadrillion.

The tech leaders are right that AI will create unprecedented abundance. They are wrong about the magnitude. It is not because machines will do what humans do. It is because the universe itself has built an asymmetry of 10²⁰ or more into the relationship between knowing and moving .

. . and we are just beginning to exploit it.

Verification of Key Calculations Calculation Value Source / Derivation k_B × T × ln(2); verified Bérut et

Landauer limit at 300K 2.87 × 10⁻²¹ J/bit al., Nature (2012) 6.86 × 10⁻¹⁹ J (413 CRC Handbook; measured >100

C-H bond energy kJ/mol) years Per-operation Bond-Bit ratio ~240× 6.86×10⁻¹⁹ / 2.87×10⁻²¹

1 kg hydrocarbon reconfiguration 71.4 mol × 6.022×10²³ × 3 × 8.9 × 10⁷ J energy 6.86×10⁻¹⁹

Calculation Value Source / Derivation 1 kg prevention info energy 2.87 × 10⁻¹² J 10⁹ bits × 2.87×10⁻²¹

(Landauer)

Macroscopic thermodynamic-floor ~10²⁰ 8.9×10⁷ / 2.87×10⁻¹² ratio Practical ratio with actuation ~10⁵ to 10⁷ 8.9×10⁷ / (10¹ to 10²)

Current computation gap to Landauer ~10⁹× ~10⁻¹² J/op ÷ ~10⁻²¹ J/bit Current practical ratio (computation

~10¹⁰ 8.9×10⁷ / 2.87×10⁻³ to remediation) 100-residue protein config space 10¹³⁰ 20¹⁰⁰

Bits to specify one sequence 432 log₂(20) × 100 Transformation leverage (protein) ~10⁶⁵ 10⁴⁹ J blind / 10⁻¹⁶ J guided

Sagawa-Ueda experimental 90% of theoretical Koski et al., PNAS (2014) verification max

Direct k_BT·I for 10⁹ bits ~10⁻¹² J Negligible at macroscopic scale Mechanical work against fluid

Typical valve actuation energy ~1–100 J pressure Verification of Key Economic Figures

Figure Value Source Global nominal GDP ~$117.2T IMF WEO, October 2025 Global labor share of

~52.4% ILO, 2024 GDP Lancet Commission (2017); VSL Pollution welfare costs $4.6T/year methodology; 2022 update

Pollution direct ~$2T/year Subset of above; human capital approach economic costs

Global R&D spending $2.87T WIPO GII, 2025 edition Cost per approved drug $2.6–2.8B Tufts CSDD

Drug-like chemical ~10⁶⁰ Standard estimate; Reymond (2012) space Goldman Sachs AI

+$7T by ~2033 GS, June 2023 estimate McKinsey gen AI +$2.6–4.4T/yr by MGI, June 2023 estimate 2040

Figure Value Source PwC, 2017 ($6.6T productivity + $9.1T PwC AI estimate +$15.7T by 2030 consumption)

Global energy waste ~60–67% of primary IEA (total) energy Addressable energy ~30–50% of total

IEA; net of Carnot-limited irreducible losses waste waste All GDP figures: IMF World Economic Outlook (October 2025). Physical constants: NIST.

R&D data: WIPO (2024). Drug discovery costs: Tufts CSDD. Bond-Bit Asymmetry: "The

Intelligence Leverage Equation" and "Thermodynamic Foundations of Entropic Shepherding"

(Anderson, 2025–2026). Sagawa-Ueda framework: Sagawa & Ueda, Physical Review Letters

(2010); experimentally verified Toyabe et al. (2010), Koski et al. (2014). GDP scenarios are the author's synthesis and are presented as scenarios, not predictions.

## References

Physics: Foundations [1] Landauer, R. "Irreversibility and heat generation in the computing process." IBM Journal of

Research and Development, 5(3), 183–191 (1961). https://doi.org/10.1147/rd.53.0183

[2] Bennett, C.H. "The thermodynamics of computation—a review." International Journal of

Theoretical Physics, 21(12), 905–940 (1982). https://doi.org/10.1007/BF02084158 [3] Maxwell, J.C. Theory of Heat. Longmans, Green and Co. (1871). [Maxwell's Demon first described in a letter to P.G. Tait, December 11, 1867.]

[4] Sagawa, T. & Ueda, M. "Second law of thermodynamics with discrete quantum feedback control." Physical Review Letters, 100, 080403 (2008). https://doi.org/10.1103/PhysRevLett.100.080403

[5] Sagawa, T. & Ueda, M. "Generalized Jarzynski equality under nonequilibrium feedback control." Physical Review Letters, 104, 090602 (2010). https://doi.org/10.1103/PhysRevLett.104.090602

[6] Sagawa, T. & Ueda, M. "Fluctuation theorem with information exchange." Journal of

Statistical Mechanics: Theory and Experiment, P01011 (2012). https://doi.org/10.1088/1742-

5468/2012/01/P01011 [7] Parrondo, J.M.R., Horowitz, J.M. & Sagawa, T. "Thermodynamics of information." Nature

Physics, 11, 131–139 (2015). https://doi.org/10.1038/nphys3230 Physics: Experimental Verification

[8] Bérut, A., Arakelyan, A., Petrosyan, A., Ciliberto, S., Dillenschneider, R. & Lutz, E.

"Experimental verification of Landauer's principle linking information and thermodynamics."

Nature, 483, 187–189 (2012). https://doi.org/10.1038/nature10872 [9] Toyabe, S., Sagawa, T., Ueda, M., Muneyuki, E. & Sano, M. "Experimental demonstration of information-to-energy conversion and validation of the generalized Jarzynski equality." Nature

Physics, 6, 988–992 (2010). https://doi.org/10.1038/nphys1821 [10] Koski, J.V., Maisi, V.F., Sagawa, T. & Pekola, J.P. "Experimental observation of the role of mutual information in the nonequilibrium dynamics of a Maxwell demon." Physical Review

Letters, 113, 030601 (2014). https://doi.org/10.1103/PhysRevLett.113.030601 [11] Koski, J.V., Maisi, V.F., Pekola, J.P. & Averin, D.V. "Experimental realization of a Szilard engine with a single electron." Proceedings of the National Academy of Sciences, 111(38),

13786–13789 (2014). https://doi.org/10.1073/pnas.1406966111 Physical Constants and Bond Energies

[12] Haynes, W.M. (ed.). CRC Handbook of Chemistry and Physics, 97th Edition. CRC Press

(2016). [C-H bond dissociation energy: 413 kJ/mol.] [13] NIST (National Institute of Standards and Technology). "Fundamental Physical Constants." https://physics.nist.gov/cuu/Constants/ [Boltzmann constant, fine-structure constant.]

Economic Forecasts: AI and GDP [14] Briggs, J. & Kodnani, D. "The Potentially Large Effects of Artificial Intelligence on

Economic Growth." Goldman Sachs Economics Research (March 2023). https://www.gspublishing.com/content/research/en/reports/2023/03/27/d64e052b-0f6e-45d7-

967b-d7be35fabd16.html [15] Chui, M. et al. "The economic potential of generative AI: The next productivity frontier."

McKinsey Global Institute (June 2023). https://www.mckinsey.com/capabilities/mckinseydigital/our-insights/the-economic-potential-of-generative-ai-the-next-productivity-frontier

[16] Bughin, J. et al. "Notes from the AI frontier: Modeling the impact of AI on the world economy." McKinsey Global Institute (September 2018). https://www.mckinsey.com/featuredinsights/artificial-intelligence/notes-from-the-ai-frontier-modeling-the-impact-of-ai-on-theworld-economy

[17] PwC. "Sizing the Prize: What's the real value of AI for your business and how can you capitalise?" PwC Global (2017). https://www.pwc.com/gx/en/issues/data-andanalytics/publications/artificial-intelligence-study.html

[18] Acemoglu, D. "The Simple Macroeconomics of AI." NBER Working Paper 32487 (2024). https://www.nber.org/papers/w32487

Economic Data: Waste Tax Sources [19] Landrigan, P.J. et al. "The Lancet Commission on pollution and health." The Lancet,

391(10119), 462–512 (2018). https://doi.org/10.1016/S0140-6736(17)32345-0 [Welfare losses:

$4.6T/year, 6.2% of global GDP.] [20] Fuller, R. et al. "Pollution and health: a progress update." The Lancet Planetary Health,

6(6), e535–e547 (2022). https://doi.org/10.1016/S2542-5196(22)00090-0 [21] FAO. "The State of Food and Agriculture 2019: Moving forward on food loss and waste reduction." Food and Agriculture Organization of the United Nations (2019). https://www.fao.org/state-of-food-agriculture/2019/en/

[22] RMI. "The Incredible Inefficiency of the Fossil Energy System." Rocky Mountain Institute

(June 2024). https://rmi.org/the-incredible-inefficiency-of-the-fossil-energy-system/ [Energy waste: ~$4.6T/year.]

[23] International Energy Agency. World Energy Outlook 2024. IEA (2024). https://www.iea.org/reports/world-energy-outlook-2024

Economic Data: Global Baselines [24] International Monetary Fund. World Economic Outlook, October 2025. IMF (2025). [Global nominal GDP ~$117.2T.]

[25] International Labour Organization. Global Wage Report 2024–25. ILO (2024). [Global labor share of GDP ~52.4%.]

[26] World Intellectual Property Organization. Global Innovation Index 2025. WIPO (2025).

[Global R&D spending: $2.87T.]

Economic Data: Ignorance Tax Sources [27] DiMasi, J.A., Grabowski, H.G. & Hansen, R.W. "Innovation in the pharmaceutical industry:

New estimates of R&D costs." Journal of Health Economics, 47, 20–33 (2016). https://doi.org/10.1016/j.jhealeco.2016.01.012 [Tufts CSDD: $2.6B per approved drug.]

[28] Reymond, J.-L. "The chemical space project." Accounts of Chemical Research, 48(3), 722–

730 (2015). https://doi.org/10.1021/ar500432k [Drug-like chemical space ~10⁶⁰.]

Author's Prior Work [29] Anderson, J. "The Intelligence Leverage Equation: Why Knowing Is 10²⁰ Times Cheaper

Than Moving—And What This Means for Environmental Protection." EnviroAI (2025). [30] Anderson, J. "Thermodynamic Foundations of Entropic Shepherding." EnviroAI (2025).

[31] Anderson, J. "The Physics of Zero-Cost Stewardship." EnviroAI (2026). [32] Anderson, J. "Generalized Functional Efficiency: A Thermodynamic Metric for the

Evolution of Complex Systems." EnviroAI (2026). [33] Anderson, J. "What is Life... and How to Protect It." EnviroAI (2026).

AI Assistance Disclosure: Google Gemini 3.0 Pro Deep Think, Grok-4.1 Deep Research,

ChatGPT 5.2 Deep Research, and Claude 4.6 Deep Research.