Canonical Claims Dataset

At room temperature (T = 300 K), this gives:

At the thermodynamic floor, **knowing one bit of information costs about 240 times less than breaking one chemical bond.** ([Full derivation and corpus reconciliation.] ) This 240× is a floor ratio, not a universal real-world constant.

Depending on the specific task being computed and the specific physical action being substituted for, the practical ratio can be larger or smaller than 240×.

The well-defended claim is the floor: at the limit physics allows, information manipulation is at least 240× cheaper than force, per microscopic event.

> Knowing the right bit costs at least 240× less than pushing the right molecule.

**Piece 2:** Information manipulation is at least 240× cheaper than force per microscopic event.

For total information bandwidth, Manfred Zimmermann (1989) estimated the conscious nervous system's processing rate at about 50 bits per second (an upper bound; many follow-up studies put it lower).

**Conclusion:** a human's conscious measurement capacity is on the order of 7 items at 50 bits/sec.

(The frontier models now reach 128k, 200k, and 1M+ tokens; we use 32k as a conservative example.) Each token carries roughly 10 bits of information after tokenization.

A modern transformer-based AI like the ones running in late 2024 holds a context window of around 32,000 tokens.

The 32,000 tokens correspond to about 128,000 characters.

Multiply: 8 billion humans, each consciously attending at about 50 bits/sec, for about 16 waking hours per day.

Each generates filtered, processed measurements at a rate of order 1,000 bits per second from sensors (accelerometer, GPS, microphone, camera, touch, network), most of which is processed by AI before reaching a human or another system.

There are roughly 15 billion connected IoT devices, each contributing at least 1 bit per second on average.

The 1,000 bits/sec figure is an order-of-magnitude assumption based on the rough number and continuity of on-device sensor processing.

The 1 bit/sec for IoT is similarly a rough floor.

Suppose we cut the per-phone rate by a factor of 10 (to 100 bits/sec), assume only half of phone processing is meaningfully AI-mediated, and discount for double-counting by another factor of 3.

The stronger claim that the ratio is at least 10× is plausible under the stated assumptions but is not strictly verified.

We still arrive at AI-side rates around 1.2 × 10¹¹ bits/sec—already comparable to the human-side estimate.

Asking ten thousand questions that each chip away 0.001% of the possibility space barely moves the needle.

At the thermodynamic floor, information is 240× cheaper than force.

The bond-bit energy ratio of approximately 240× at room temperature, computed from established constants.

Fission and fusion convert measurably larger fractions of mass (about 0.1% and 0.4% respectively), but even these are small numbers.

Fission and fusion convert measurably larger fractions of mass (about 0.1% and 0.4% respectively), but even these are small numbers.

Gradient-harvesting events sit far below mass-destruction events: one bit at the Landauer bound (17.8 meV), one 500 nm photon (2.48 eV), and one C-H bond (4.28 eV) versus electron-positron annihilation (1.0 MeV), one D-T fusion event (17.6 MeV), and one U-235 fission event (200 MeV).]

Gradient-harvesting events sit far below mass-destruction events: one bit at the Landauer bound (17.8 meV), one 500 nm photon (2.48 eV), and one C-H bond (4.28 eV) versus electron-positron annihilation (1.0 MeV), one D-T fusion event (17.6 MeV), and one U-235 fission event (200 MeV).]

Gradient-harvesting events sit far below mass-destruction events: one bit at the Landauer bound (17.8 meV), one 500 nm photon (2.48 eV), and one C-H bond (4.28 eV) versus electron-positron annihilation (1.0 MeV), one D-T fusion event (17.6 MeV), and one U-235 fission event (200 MeV).]

Gradient-harvesting events sit far below mass-destruction events: one bit at the Landauer bound (17.8 meV), one 500 nm photon (2.48 eV), and one C-H bond (4.28 eV) versus electron-positron annihilation (1.0 MeV), one D-T fusion event (17.6 MeV), and one U-235 fission event (200 MeV).]

Gradient-harvesting events sit far below mass-destruction events: one bit at the Landauer bound (17.8 meV), one 500 nm photon (2.48 eV), and one C-H bond (4.28 eV) versus electron-positron annihilation (1.0 MeV), one D-T fusion event (17.6 MeV), and one U-235 fission event (200 MeV).]

Gradient-harvesting events sit far below mass-destruction events: one bit at the Landauer bound (17.8 meV), one 500 nm photon (2.48 eV), and one C-H bond (4.28 eV) versus electron-positron annihilation (1.0 MeV), one D-T fusion event (17.6 MeV), and one U-235 fission event (200 MeV).]

Matter-antimatter annihilation converts 100% of the participating rest mass.

Nuclear reactions rearrange nucleons inside the nucleus itself; a measurable fraction of mass (about 0.1% for fission, 0.4% for fusion) disappears as binding energy.

Nuclear reactions rearrange nucleons inside the nucleus itself; a measurable fraction of mass (about 0.1% for fission, 0.4% for fusion) disappears as binding energy.

![Horizontal bar chart of the fraction of rest mass converted to energy on a logarithmic axis: a C-H bond converts 3.76×10⁻⁸ percent, U-235 fission 0.091 percent, D-T fusion 0.378 percent, solar pp-chain fusion 0.717 percent, and matter-antimatter annihilation 100 percent.]

- Per single C-H bond: 6.86 × 10⁻¹⁹ J = 4.28 eV (canonical reference, 413 kJ/mol) - Per single visible photon: 3.97 × 10⁻¹⁹ J = 2.48 eV (500 nm) - Per single fission event: 3.20 × 10⁻¹¹ J = 200 MeV (≈ 47 million × a bond) - Per single fusion event: 2.82 × 10⁻¹² J = 17.6 MeV (≈ 4 million × a bond) - Per single bit at 300 K: 2.87 × 10⁻²¹ J = 0.018 eV (1/240 of a bond: the bond-bit ratio)

- Per single C-H bond: 6.86 × 10⁻¹⁹ J = 4.28 eV (canonical reference, 413 kJ/mol) - Per single visible photon: 3.97 × 10⁻¹⁹ J = 2.48 eV (500 nm) - Per single fission event: 3.20 × 10⁻¹¹ J = 200 MeV (≈ 47 million × a bond) - Per single fusion event: 2.82 × 10⁻¹² J = 17.6 MeV (≈ 4 million × a bond) - Per single bit at 300 K: 2.87 × 10⁻²¹ J = 0.018 eV (1/240 of a bond: the bond-bit ratio)

- Per single C-H bond: 6.86 × 10⁻¹⁹ J = 4.28 eV (canonical reference, 413 kJ/mol) - Per single visible photon: 3.97 × 10⁻¹⁹ J = 2.48 eV (500 nm) - Per single fission event: 3.20 × 10⁻¹¹ J = 200 MeV (≈ 47 million × a bond) - Per single fusion event: 2.82 × 10⁻¹² J = 17.6 MeV (≈ 4 million × a bond) - Per single bit at 300 K: 2.87 × 10⁻²¹ J = 0.018 eV (1/240 of a bond: the bond-bit ratio)

- Per single C-H bond: 6.86 × 10⁻¹⁹ J = 4.28 eV (canonical reference, 413 kJ/mol) - Per single visible photon: 3.97 × 10⁻¹⁹ J = 2.48 eV (500 nm) - Per single fission event: 3.20 × 10⁻¹¹ J = 200 MeV (≈ 47 million × a bond) - Per single fusion event: 2.82 × 10⁻¹² J = 17.6 MeV (≈ 4 million × a bond) - Per single bit at 300 K: 2.87 × 10⁻²¹ J = 0.018 eV (1/240 of a bond: the bond-bit ratio)

- Per single C-H bond: 6.86 × 10⁻¹⁹ J = 4.28 eV (canonical reference, 413 kJ/mol) - Per single visible photon: 3.97 × 10⁻¹⁹ J = 2.48 eV (500 nm) - Per single fission event: 3.20 × 10⁻¹¹ J = 200 MeV (≈ 47 million × a bond) - Per single fusion event: 2.82 × 10⁻¹² J = 17.6 MeV (≈ 4 million × a bond) - Per single bit at 300 K: 2.87 × 10⁻²¹ J = 0.018 eV (1/240 of a bond: the bond-bit ratio)

- Per single C-H bond: 6.86 × 10⁻¹⁹ J = 4.28 eV (canonical reference, 413 kJ/mol) - Per single visible photon: 3.97 × 10⁻¹⁹ J = 2.48 eV (500 nm) - Per single fission event: 3.20 × 10⁻¹¹ J = 200 MeV (≈ 47 million × a bond) - Per single fusion event: 2.82 × 10⁻¹² J = 17.6 MeV (≈ 4 million × a bond) - Per single bit at 300 K: 2.87 × 10⁻²¹ J = 0.018 eV (1/240 of a bond: the bond-bit ratio)

- Per single C-H bond: 6.86 × 10⁻¹⁹ J = 4.28 eV (canonical reference, 413 kJ/mol) - Per single visible photon: 3.97 × 10⁻¹⁹ J = 2.48 eV (500 nm) - Per single fission event: 3.20 × 10⁻¹¹ J = 200 MeV (≈ 47 million × a bond) - Per single fusion event: 2.82 × 10⁻¹² J = 17.6 MeV (≈ 4 million × a bond) - Per single bit at 300 K: 2.87 × 10⁻²¹ J = 0.018 eV (1/240 of a bond: the bond-bit ratio)

- Per single C-H bond: 6.86 × 10⁻¹⁹ J = 4.28 eV (canonical reference, 413 kJ/mol) - Per single visible photon: 3.97 × 10⁻¹⁹ J = 2.48 eV (500 nm) - Per single fission event: 3.20 × 10⁻¹¹ J = 200 MeV (≈ 47 million × a bond) - Per single fusion event: 2.82 × 10⁻¹² J = 17.6 MeV (≈ 4 million × a bond) - Per single bit at 300 K: 2.87 × 10⁻²¹ J = 0.018 eV (1/240 of a bond: the bond-bit ratio)

- Per single C-H bond: 6.86 × 10⁻¹⁹ J = 4.28 eV (canonical reference, 413 kJ/mol) - Per single visible photon: 3.97 × 10⁻¹⁹ J = 2.48 eV (500 nm) - Per single fission event: 3.20 × 10⁻¹¹ J = 200 MeV (≈ 47 million × a bond) - Per single fusion event: 2.82 × 10⁻¹² J = 17.6 MeV (≈ 4 million × a bond) - Per single bit at 300 K: 2.87 × 10⁻²¹ J = 0.018 eV (1/240 of a bond: the bond-bit ratio)

- Per single C-H bond: 6.86 × 10⁻¹⁹ J = 4.28 eV (canonical reference, 413 kJ/mol) - Per single visible photon: 3.97 × 10⁻¹⁹ J = 2.48 eV (500 nm) - Per single fission event: 3.20 × 10⁻¹¹ J = 200 MeV (≈ 47 million × a bond) - Per single fusion event: 2.82 × 10⁻¹² J = 17.6 MeV (≈ 4 million × a bond) - Per single bit at 300 K: 2.87 × 10⁻²¹ J = 0.018 eV (1/240 of a bond: the bond-bit ratio)

- Per single C-H bond: 6.86 × 10⁻¹⁹ J = 4.28 eV (canonical reference, 413 kJ/mol) - Per single visible photon: 3.97 × 10⁻¹⁹ J = 2.48 eV (500 nm) - Per single fission event: 3.20 × 10⁻¹¹ J = 200 MeV (≈ 47 million × a bond) - Per single fusion event: 2.82 × 10⁻¹² J = 17.6 MeV (≈ 4 million × a bond) - Per single bit at 300 K: 2.87 × 10⁻²¹ J = 0.018 eV (1/240 of a bond: the bond-bit ratio)

- Per single C-H bond: 6.86 × 10⁻¹⁹ J = 4.28 eV (canonical reference, 413 kJ/mol) - Per single visible photon: 3.97 × 10⁻¹⁹ J = 2.48 eV (500 nm) - Per single fission event: 3.20 × 10⁻¹¹ J = 200 MeV (≈ 47 million × a bond) - Per single fusion event: 2.82 × 10⁻¹² J = 17.6 MeV (≈ 4 million × a bond) - Per single bit at 300 K: 2.87 × 10⁻²¹ J = 0.018 eV (1/240 of a bond: the bond-bit ratio)

- Per single C-H bond: 6.86 × 10⁻¹⁹ J = 4.28 eV (canonical reference, 413 kJ/mol) - Per single visible photon: 3.97 × 10⁻¹⁹ J = 2.48 eV (500 nm) - Per single fission event: 3.20 × 10⁻¹¹ J = 200 MeV (≈ 47 million × a bond) - Per single fusion event: 2.82 × 10⁻¹² J = 17.6 MeV (≈ 4 million × a bond) - Per single bit at 300 K: 2.87 × 10⁻²¹ J = 0.018 eV (1/240 of a bond: the bond-bit ratio)

The C-H bond enthalpy and the Landauer bound at 300 K give the canonical bond-bit ratio of approximately 240 (Anderson, 2026).

The Landauer bound has been experimentally verified to within 10% at the lab scale (Bérut et al., 2012).

The demon monitored electron positions in real time and applied feedback that extracted approximately kT ln 2 of work per bit of information acquired, within 10% of the theoretical Landauer ceiling.

![Bar chart of continuous power on a logarithmic terawatt axis: industrial civilization delivers about 20 TW, the biosphere's net primary productivity about 130 TW, and solar exergy reaching Earth's surface about 115,000 TW—roughly 5,800 times human demand.]

![Bar chart of continuous power on a logarithmic terawatt axis: industrial civilization delivers about 20 TW, the biosphere's net primary productivity about 130 TW, and solar exergy reaching Earth's surface about 115,000 TW—roughly 5,800 times human demand.]

![Bar chart of continuous power on a logarithmic terawatt axis: industrial civilization delivers about 20 TW, the biosphere's net primary productivity about 130 TW, and solar exergy reaching Earth's surface about 115,000 TW—roughly 5,800 times human demand.]

Solar exergy at Earth's surface is roughly 115,000 TW after albedo, about 5,800× human energy consumption.

Solar exergy at Earth's surface is roughly 115,000 TW after albedo, about 5,800× human energy consumption.

Industrial civilization delivers ~20 TW of high-exergy power.

The biosphere generates ~130 TW of gross chemical-bond formation (most of which cycles back to CO₂).

The hot reservoir is the sun at 5800 K.

The cold reservoir is the ambient environment at 300 K.

AE exceeds biological photosynthesis on energy density (silicon PV averages around 200 watts per square meter at peak, against 0.5 watts per square meter for a leaf), on conversion efficiency (47% in laboratory multijunction PV against ~6% for a typical C3 plant), and on output flexibility (electricity, hydrogen, ammonia, designer molecules).

AE exceeds biological photosynthesis on energy density (silicon PV averages around 200 watts per square meter at peak, against 0.5 watts per square meter for a leaf), on conversion efficiency (47% in laboratory multijunction PV against ~6% for a typical C3 plant), and on output flexibility (electricity, hydrogen, ammonia, designer molecules).

This is the bond-bit ratio, derived in full in the canonical source (Anderson, The Bond-Bit Ratio, 2026), and it holds for the C-H reference at 300 K; it is not a universal constant across all bonds or temperatures.

Steering a chemical process requires handling information, and the minimum thermodynamic cost of irreversibly handling one bit is the Landauer bound, kT ln 2, about 2.87 × 10⁻²¹ joules at 300 K.

Per quantum carrier, the cost of irreversibly handling a bit at 300 K is approximately 18 millielectron-volts: about 140 times less than a single visible photon, and roughly 240 times less than the canonical carbon-hydrogen bond, the bond-bit ratio established earlier.

That ratio is specific to the C-H reference at 300 K, not a universal constant; weaker bonds narrow it and the Landauer cost scales with temperature.

| Dimension | Traditional energy | Artificial energy | | --- | --- | --- | | Resource ceiling | Finite stock; depletes as used | ~115,000 TW solar at surface; renewed daily | | Energy return (EROI) | Declines as easy reserves deplete | PV ~10-30, wind ~15-35, and rising | | Efficiency limit | Carnot-bound; must dump >40% as heat | Not Carnot-bound; heat pumps reach 300-500% | | Marginal fuel cost | Floored by extraction; never zero | Zero; the gradient is not invoiced | | Lifecycle carbon | Coal ~820, gas ~490 g/kWh | Solar ~48, wind ~11 g/kWh | | Levelized cost (generation) | Coal $69-168, gas $45-108 /MWh | Solar $29-92, wind $27-73 /MWh | | Waste profile | CO2 intrinsic to the reaction | Recycled material flows; heat re-radiated |

| Dimension | Traditional energy | Artificial energy | | --- | --- | --- | | Resource ceiling | Finite stock; depletes as used | ~115,000 TW solar at surface; renewed daily | | Energy return (EROI) | Declines as easy reserves deplete | PV ~10-30, wind ~15-35, and rising | | Efficiency limit | Carnot-bound; must dump >40% as heat | Not Carnot-bound; heat pumps reach 300-500% | | Marginal fuel cost | Floored by extraction; never zero | Zero; the gradient is not invoiced | | Lifecycle carbon | Coal ~820, gas ~490 g/kWh | Solar ~48, wind ~11 g/kWh | | Levelized cost (generation) | Coal $69-168, gas $45-108 /MWh | Solar $29-92, wind $27-73 /MWh | | Waste profile | CO2 intrinsic to the reaction | Recycled material flows; heat re-radiated |

| Dimension | Traditional energy | Artificial energy | | --- | --- | --- | | Resource ceiling | Finite stock; depletes as used | ~115,000 TW solar at surface; renewed daily | | Energy return (EROI) | Declines as easy reserves deplete | PV ~10-30, wind ~15-35, and rising | | Efficiency limit | Carnot-bound; must dump >40% as heat | Not Carnot-bound; heat pumps reach 300-500% | | Marginal fuel cost | Floored by extraction; never zero | Zero; the gradient is not invoiced | | Lifecycle carbon | Coal ~820, gas ~490 g/kWh | Solar ~48, wind ~11 g/kWh | | Levelized cost (generation) | Coal $69-168, gas $45-108 /MWh | Solar $29-92, wind $27-73 /MWh | | Waste profile | CO2 intrinsic to the reaction | Recycled material flows; heat re-radiated |

| Dimension | Traditional energy | Artificial energy | | --- | --- | --- | | Resource ceiling | Finite stock; depletes as used | ~115,000 TW solar at surface; renewed daily | | Energy return (EROI) | Declines as easy reserves deplete | PV ~10-30, wind ~15-35, and rising | | Efficiency limit | Carnot-bound; must dump >40% as heat | Not Carnot-bound; heat pumps reach 300-500% | | Marginal fuel cost | Floored by extraction; never zero | Zero; the gradient is not invoiced | | Lifecycle carbon | Coal ~820, gas ~490 g/kWh | Solar ~48, wind ~11 g/kWh | | Levelized cost (generation) | Coal $69-168, gas $45-108 /MWh | Solar $29-92, wind $27-73 /MWh | | Waste profile | CO2 intrinsic to the reaction | Recycled material flows; heat re-radiated |

| Dimension | Traditional energy | Artificial energy | | --- | --- | --- | | Resource ceiling | Finite stock; depletes as used | ~115,000 TW solar at surface; renewed daily | | Energy return (EROI) | Declines as easy reserves deplete | PV ~10-30, wind ~15-35, and rising | | Efficiency limit | Carnot-bound; must dump >40% as heat | Not Carnot-bound; heat pumps reach 300-500% | | Marginal fuel cost | Floored by extraction; never zero | Zero; the gradient is not invoiced | | Lifecycle carbon | Coal ~820, gas ~490 g/kWh | Solar ~48, wind ~11 g/kWh | | Levelized cost (generation) | Coal $69-168, gas $45-108 /MWh | Solar $29-92, wind $27-73 /MWh | | Waste profile | CO2 intrinsic to the reaction | Recycled material flows; heat re-radiated |

| Dimension | Traditional energy | Artificial energy | | --- | --- | --- | | Resource ceiling | Finite stock; depletes as used | ~115,000 TW solar at surface; renewed daily | | Energy return (EROI) | Declines as easy reserves deplete | PV ~10-30, wind ~15-35, and rising | | Efficiency limit | Carnot-bound; must dump >40% as heat | Not Carnot-bound; heat pumps reach 300-500% | | Marginal fuel cost | Floored by extraction; never zero | Zero; the gradient is not invoiced | | Lifecycle carbon | Coal ~820, gas ~490 g/kWh | Solar ~48, wind ~11 g/kWh | | Levelized cost (generation) | Coal $69-168, gas $45-108 /MWh | Solar $29-92, wind $27-73 /MWh | | Waste profile | CO2 intrinsic to the reaction | Recycled material flows; heat re-radiated |

The honest comparison is between human demand (~20 TW) and total solar exergy at Earth's surface (~115,000 TW), a ratio of roughly 5,800.

The honest comparison is between human demand (~20 TW) and total solar exergy at Earth's surface (~115,000 TW), a ratio of roughly 5,800.

For 200 years, we drew most of our energy from disassembling matter.

The Bond-Bit Ratio: A derivation of why information is at least 240× cheaper than force.

The IPCC's 67%-confidence 1.5°C remaining carbon budget (400 GtCO₂ from January 2020) will be exhausted around 2030 at current emission rates near 41.6 GtCO₂/yr.

Depends on the budget confidence level (67% vs 50%), the assumed non-CO₂ forcing trajectory, and TCRE distribution; central year is robust to ±2 years.

The Bekenstein bound and the holographic principle indicate the total information content of any region of spacetime is bounded by its boundary area, not its volume.

Bekenstein bound is established in the gravitational regime; the holographic principle's extension to non-gravitational systems is a theoretical conjecture, not a measurement.

In September 2022 the DART spacecraft's collision with Dimorphos shortened the moonlet's orbital period by 32 minutes — first time in 4.5 billion years a species has moved a celestial body off its orbit.

Measured directly post-impact; uncertainty ±2 minutes per NASA confirmation.

Three versions of the environmental superintelligence thesis: weak (AI is useful — settled); middle (AI-scale information processing is necessary — defensible from verified physics); strong (superintelligent AI is the only physically adequate response — philosophical, supported by evidence but not derived).

Strong version is a philosophical commitment, not a verified theorem. Middle version is defensible from current physics + Earth-system science. Weak is settled.

DNA sequencing cost fell from $95 million per genome in 2001 to a few hundred dollars by 2024 — roughly 5 orders of magnitude.

NHGRI numbers; minor methodological breaks (raw sequencing vs. fully assembled and validated) make the exact endpoint a small range, not a single number.

The full loop from health discovery to actual change in the air people breathe under NAAQS implementation runs two to three decades.

Composite empirical observation over four NAAQS cycles; precise loop time varies by pollutant and facility class.

Six of nine planetary boundaries have been transgressed as of 2023.

Per Richardson et al. 2023; specific boundary definitions are framework-dependent but the count is uncontested in the Stockholm Resilience Centre formulation.

TROPOMI on Sentinel-5P images NO₂ plumes from individual industrial facilities at 3.5 by 5.5 km per-pixel resolution from orbit.

Nominal resolution; effective per-pixel SNR varies with cloud cover and solar zenith.

Cities* in the *New England Journal of Medicine*—the landmark Harvard Six Cities Study that established, with a 16-year prospective cohort, that long-term exposure to fine particulate matter at concentrations common in U.S.

In October 2006, EPA tightened the 24-hour standard to 35 µg/m³.

Four years later, in July 1997, EPA promulgated the first National Ambient Air Quality Standard for PM2.5 at 15 µg/m³ annual and 65 µg/m³ 24-hour.

Armstrong McKay and his colleagues, writing in *Science* in 2022, showed that several of the major climate tipping elements—the Greenland ice sheet, the West Antarctic ice sheet, warm-water coral reefs, North Atlantic deep-water formation—have central uncertainty ranges that begin below 2 °C of warming, with some entering the lower bound of plausibility at warming we have already reached.

NOAA and NASA reported on 10 January 2025 that the 2024 global mean surface temperature was 1.46 to 1.47 °C above the 1850–1900 baseline; the World Meteorological Organization's six-dataset synthesis put it at 1.55 °C.

NOAA and NASA reported on 10 January 2025 that the 2024 global mean surface temperature was 1.46 to 1.47 °C above the 1850–1900 baseline; the World Meteorological Organization's six-dataset synthesis put it at 1.55 °C.

The IPCC's remaining carbon budget for a two-in-three chance of holding 1.5 °C—400 GtCO₂ from January 2020—will be exhausted around 2030 at current emission rates near 41.6 GtCO₂ per year, as reported by the Global Carbon Project in *Earth System Science Data* in 2025.

At 300 K—room temperature, planet temperature—Boltzmann's constant times ln 2 works out to 2.87 × 10⁻²¹ joules per bit.

(See [the canonical bond-bit ratio derivation] for the full constants, the C–H reference giving the corpus-standard 240× figure, and the reconciliation of the variants that appear across these essays.) At fundamental physical limits, directing matter by information is roughly two orders of magnitude cheaper than directing matter by force.

Epoch AI documents that training compute for notable AI models has grown 4.1× per year, with a 90% confidence interval from 3.7× to 4.6×, between 2010 and May 2024.

Epoch AI documents that training compute for notable AI models has grown 4.1× per year, with a 90% confidence interval from 3.7× to 4.6×, between 2010 and May 2024.

The Model Evaluation and Threat Research organization (METR) measured something more important than raw compute in their 2025 study on long-horizon tasks: the time-length of tasks an AI agent can reliably complete has been doubling every seven months over 2019–2024, and the post-2023 doubling time has tightened to about 130 days.

Epoch AI documents that training compute for notable AI models has grown 4.1× per year, with a 90% confidence interval from 3.7× to 4.6×, between 2010 and May 2024.

Epoch AI documents that training compute for notable AI models has grown 4.1× per year, with a 90% confidence interval from 3.7× to 4.6×, between 2010 and May 2024.

The Metaculus question on the public emergence of weakly general AI—answered by more than 1,700 forecasters—has compressed from a community median of around 2050 in early 2020 to around November 2027 by late 2024.

Frontier language models have been doubling every 5.2 months since 2020.

The 50%-confidence budget extends the window only modestly.

Lenton and his co-authors, writing as a *Nature* Comment in 2019, called the cluster of active tipping elements a "state of planetary emergency" and warned that "the intervention time left to prevent tipping could already have shrunk towards zero, whereas the reaction time to achieve net zero emissions is 30 years at best." The arithmetic of the IPCC budgets and the Global Carbon Project's 2024 emissions inventory implies that the 67%-confidence 1.5 °C budget will be exhausted around 2030.

Lenton and his co-authors, writing as a *Nature* Comment in 2019, called the cluster of active tipping elements a "state of planetary emergency" and warned that "the intervention time left to prevent tipping could already have shrunk towards zero, whereas the reaction time to achieve net zero emissions is 30 years at best." The arithmetic of the IPCC budgets and the Global Carbon Project's 2024 emissions inventory implies that the 67%-confidence 1.5 °C budget will be exhausted around 2030.

Lenton and his co-authors, writing as a *Nature* Comment in 2019, called the cluster of active tipping elements a "state of planetary emergency" and warned that "the intervention time left to prevent tipping could already have shrunk towards zero, whereas the reaction time to achieve net zero emissions is 30 years at best." The arithmetic of the IPCC budgets and the Global Carbon Project's 2024 emissions inventory implies that the 67%-confidence 1.5 °C budget will be exhausted around 2030.

Pangu-Weather, published by Bi and colleagues in *Nature* the same year, was the first AI model to beat operational numerical weather prediction across all variables at all lead times, at roughly 10,000× the inference speed.

GraphCast, the DeepMind model published by Lam and colleagues in *Science* in 2023, produces ten-day global weather forecasts in under a minute on a single TPU and outperforms the European Centre for Medium-Range Weather Forecasts' gold-standard HRES system on 90% of 1,380 verification targets.

Google's Flood Hub, built on the work of Nearing and colleagues published in *Nature* in 2024, delivers extreme-flood forecasts at five-day lead time that match or beat the Copernicus Global Flood Awareness System's same-day nowcasts, and is now operational in more than 80 countries.

GraphCast, the DeepMind model published by Lam and colleagues in *Science* in 2023, produces ten-day global weather forecasts in under a minute on a single TPU and outperforms the European Centre for Medium-Range Weather Forecasts' gold-standard HRES system on 90% of 1,380 verification targets.

FourCastNet, from a team led by Pathak at NVIDIA in 2022, reported 45,000× speedups.

The DART spacecraft's collision with Dimorphos shortened the moonlet's orbital period by 32 minutes—more than twenty-five times the threshold the mission set for success.

It is what the numbers—the two-to-three-decade health-to-facility loop in NAAQS implementation, the 2030 budget exhaustion, the six of nine planetary boundaries crossed, the 1.55 °C 2024 anomaly—actually say.

The canonical bond-bit asymmetry: information is at least 240× cheaper than force at the thermodynamic floor (C–H bond, 300 K, Landauer's bound).

Floor ratio. Real-world operational ratio is 10^8 to 10^12; 240× is the irreducible minimum.

- k = 1.380649 × 10⁻²³ J/K (exact, 2019 SI redefinition) - T = 300 K (≈ 27 °C; close to global mean surface temperature) - ln 2 ≈ 0.6931

- k = 1.380649 × 10⁻²³ J/K (exact, 2019 SI redefinition) - T = 300 K (≈ 27 °C; close to global mean surface temperature) - ln 2 ≈ 0.6931

- k = 1.380649 × 10⁻²³ J/K (exact, 2019 SI redefinition) - T = 300 K (≈ 27 °C; close to global mean surface temperature) - ln 2 ≈ 0.6931

E_bit ≈ (1.380649 × 10⁻²³ J/K) × (300 K) × (0.6931) ≈ **2.870 × 10⁻²¹ J/bit**

E_bit ≈ (1.380649 × 10⁻²³ J/K) × (300 K) × (0.6931) ≈ **2.870 × 10⁻²¹ J/bit**

E_bit ≈ (1.380649 × 10⁻²³ J/K) × (300 K) × (0.6931) ≈ **2.870 × 10⁻²¹ J/bit**

Exact under 2019 SI redefinition of k. Landauer's principle verified by Bérut et al. 2012.

This is the *Landauer bound at 300 K*.

The bound also scales linearly with temperature—at cryogenic temperatures (4 K) the floor drops by roughly 75×.

The bound also scales linearly with temperature—at cryogenic temperatures (4 K) the floor drops by roughly 75×.

We evaluate at 300 K because every chemical transformation a biosphere cares about happens near planetary surface temperature, which is the relevant regime for environmental claims.

Its mean bond dissociation enthalpy is approximately 413 kJ/mol.

- ΔH_C–H ≈ 413 × 10³ J/mol - N_A = 6.02214 × 10²³ /mol

E_bond ≈ (413 × 10³ J/mol) / (6.02214 × 10²³ /mol) ≈ **6.86 × 10⁻¹⁹ J/bond**

E_bond ≈ (413 × 10³ J/mol) / (6.02214 × 10²³ /mol) ≈ **6.86 × 10⁻¹⁹ J/bond**

R = E_bond / E_bit ≈ (6.86 × 10⁻¹⁹ J) / (2.870 × 10⁻²¹ J) ≈ **239**

R = E_bond / E_bit ≈ (6.86 × 10⁻¹⁹ J) / (2.870 × 10⁻²¹ J) ≈ **239**

To the round number: **approximately 240×**.

Floor ratio; exact value 239× using C–H at 413 kJ/mol. Robust to bond choice within the 200–300× window.

Any common chemical bond, divided by Landauer's bound at 300 K, lands in the 200–300× window.

Any common chemical bond, divided by Landauer's bound at 300 K, lands in the 200–300× window.

For a stronger reference bond (O–H, ~463 kJ/mol) the ratio rises to ~270×; for a weaker one (C–C, ~347 kJ/mol) it falls to ~200×.

For a stronger reference bond (O–H, ~463 kJ/mol) the ratio rises to ~270×; for a weaker one (C–C, ~347 kJ/mol) it falls to ~200×.

For a stronger reference bond (O–H, ~463 kJ/mol) the ratio rises to ~270×; for a weaker one (C–C, ~347 kJ/mol) it falls to ~200×.

For a stronger reference bond (O–H, ~463 kJ/mol) the ratio rises to ~270×; for a weaker one (C–C, ~347 kJ/mol) it falls to ~200×.

The 240× figure is a **floor ratio**.

The 240× figure is the **narrowest, most conservative, irreducible version** of the bond-bit asymmetry.

No engineering improvement in computing hardware can take the ratio below 240×, because the denominator is fixed by the second law of thermodynamics and the numerator is fixed by the energetics of chemical bonding.

If you cite the 240× figure in a paper, talk, model, or argument, please cite this derivation as the canonical source:

The 240× figure is load-bearing in:

Added section 7 listing the essays in which the 240× figure is load-bearing.

S = k₂A / 4ℓₚ² (2) where A is the horizon area and ℓ = √(ħG/c³) ≈ 1.616 × 10⁻³⁵ m is the Planck length.

B hole carries S ≈ 10⁷⁷k , vastly exceeding any other object of comparable mass.

Embedding Conjecture, a 40-year-old open problem in operator algebras concerning the structure of von Neumann factors, is false.

Quantum Singularities: The Planck Scale At the Planck scale (ℓ ≈ 1.6 × 10⁻³⁵ m), the Schwarzschild radius of a Planck-mass particle

Analyzing a wide range of recordings—classical, jazz, rock, talk radio—they found that the spectral density of audiopower fluctuations follows a 1/f law over many decades of frequency, from the scale of individual notes down to 5×10⁻⁴ Hz (the full length of a composition).

Schmidhuber's compression-progress theory, developed across 1997 and 2009 papers, provides the theoretical unification.

Asteroid: one civilization-ender every 500,000 years.

That floor is set by Landauer's bound at 300 K and the carbon-hydrogen bond enthalpy.

- **413 kJ/mol.** Energy bound in a single C–H bond, the cost of photosynthesis.

The derivation of why information is at least 240× cheaper than force, set by Landauer's bound at 300 K and the carbon-hydrogen bond enthalpy.

The derivation of why information is at least 240× cheaper than force, set by Landauer's bound at 300 K and the carbon-hydrogen bond enthalpy.

Facts checked: Landauer kT ln 2 (1961); carbon-hydrogen bond enthalpy, 413 kJ/mol; DART impact, September 26, 2022; SpaceX IPO valuation, $1.75 trillion, June 12, 2026; Lloyd, *Computational capacity of the universe*, 2002.

Maxwell's thought experiment proved extraordinarily fruitful, acting as a catalyst for over 150 years of research and debate at the intersection of thermodynamics, statistical mechanics, and information theory.7 It forced physicists to grapple with the physical meaning of information, the thermodynamics of measurement, the nature of computation, and the role of the observer in physical laws.

2.3: Maxwell's Demon, information, and computing - Physics LibreTexts, accessed May 5, 2025, ics/Essential_Graduate_Physics_-_Statistical_Mechanics_(Likharev)/02%3A_Princi ples_of_Physical_Statistics/2.03%3A_Maxwells_Demon_information_and_computi ng

2.3: Maxwell's Demon, information, and computing - Physics LibreTexts, accessed May 5, 2025, ics/Essential_Graduate_Physics_-_Statistical_Mechanics_(Likharev)/02%3A_Princi ples_of_Physical_Statistics/2.03%3A_Maxwells_Demon_information_and_computi ng

The convergence of these fields in the next 10-20 years could indeed produce a prototype “Environmental Holographic Protection System.” Crucially, none of these require new physics – they work within known laws, simply harnessing information better.

Koomey's Law: computational efficiency per joule has doubled approximately every 2.3 years since 1950, with the doubling time lengthening to ~2.6 years post-Dennard scaling.

Trend curve; near-term doubling rate is sensitive to whether you measure at the device, package, or system level.

We establish this claim through seven independent lines of evidence grounded in first-principles physics: (1) nature’s information content exceeds all AI training data by a ratio of 1020–1035, constrained by exact conservation laws that internet data lacks; (2) the Bond-Bit Asymmetry (experimentally verified, practical ratio ~1010 today, approaching 1020 at Landauer limit) creates a structural economic incentive for physics-grounded AI to prefer prevention over destruction, with this incentive growing monotonically as computational costs decline; (3) the set of planetary states compatible with human welfare is a proper subset of those compatible with ecosystem health (H ⊂ E), making ecocentric optimization the strictly safer target; (4) evolution constitutes a 3.8-billion-year alignment test suite whose constraint structure—conservation laws as non-negotiable rules—encodes proven multi-agent coordination solutions alongside competitive strategies, both tested under exact physical law; (5) the entropy/negentropy framework provides objective, physically falsifiable alignment criteria that are resistant to Goodharting because conservation laws require closed-system accounting, making local gaming physically detectable; (6) Generalized Functional Efficiency (GFE = F/(Ṡ·M)) provides a quantitative alignment metric validated across 50 orders of magnitude and 13.8 billion years of cosmic history; and (7) ESI aligns AI with the observable cosmic trajectory from pure dissipation toward pure function—the deepest optimization pattern in physics.

We establish this claim through seven independent lines of evidence grounded in first-principles physics: (1) nature’s information content exceeds all AI training data by a ratio of 1020–1035, constrained by exact conservation laws that internet data lacks; (2) the Bond-Bit Asymmetry (experimentally verified, practical ratio ~1010 today, approaching 1020 at Landauer limit) creates a structural economic incentive for physics-grounded AI to prefer prevention over destruction, with this incentive growing monotonically as computational costs decline; (3) the set of planetary states compatible with human welfare is a proper subset of those compatible with ecosystem health (H ⊂ E), making ecocentric optimization the strictly safer target; (4) evolution constitutes a 3.8-billion-year alignment test suite whose constraint structure—conservation laws as non-negotiable rules—encodes proven multi-agent coordination solutions alongside competitive strategies, both tested under exact physical law; (5) the entropy/negentropy framework provides objective, physically falsifiable alignment criteria that are resistant to Goodharting because conservation laws require closed-system accounting, making local gaming physically detectable; (6) Generalized Functional Efficiency (GFE = F/(Ṡ·M)) provides a quantitative alignment metric validated across 50 orders of magnitude and 13.8 billion years of cosmic history; and (7) ESI aligns AI with the observable cosmic trajectory from pure dissipation toward pure function—the deepest optimization pattern in physics.

We establish this claim through seven independent lines of evidence grounded in first-principles physics: (1) nature’s information content exceeds all AI training data by a ratio of 1020–1035, constrained by exact conservation laws that internet data lacks; (2) the Bond-Bit Asymmetry (experimentally verified, practical ratio ~1010 today, approaching 1020 at Landauer limit) creates a structural economic incentive for physics-grounded AI to prefer prevention over destruction, with this incentive growing monotonically as computational costs decline; (3) the set of planetary states compatible with human welfare is a proper subset of those compatible with ecosystem health (H ⊂ E), making ecocentric optimization the strictly safer target; (4) evolution constitutes a 3.8-billion-year alignment test suite whose constraint structure—conservation laws as non-negotiable rules—encodes proven multi-agent coordination solutions alongside competitive strategies, both tested under exact physical law; (5) the entropy/negentropy framework provides objective, physically falsifiable alignment criteria that are resistant to Goodharting because conservation laws require closed-system accounting, making local gaming physically detectable; (6) Generalized Functional Efficiency (GFE = F/(Ṡ·M)) provides a quantitative alignment metric validated across 50 orders of magnitude and 13.8 billion years of cosmic history; and (7) ESI aligns AI with the observable cosmic trajectory from pure dissipation toward pure function—the deepest optimization pattern in physics.

We establish this claim through seven independent lines of evidence grounded in first-principles physics: (1) nature’s information content exceeds all AI training data by a ratio of 1020–1035, constrained by exact conservation laws that internet data lacks; (2) the Bond-Bit Asymmetry (experimentally verified, practical ratio ~1010 today, approaching 1020 at Landauer limit) creates a structural economic incentive for physics-grounded AI to prefer prevention over destruction, with this incentive growing monotonically as computational costs decline; (3) the set of planetary states compatible with human welfare is a proper subset of those compatible with ecosystem health (H ⊂ E), making ecocentric optimization the strictly safer target; (4) evolution constitutes a 3.8-billion-year alignment test suite whose constraint structure—conservation laws as non-negotiable rules—encodes proven multi-agent coordination solutions alongside competitive strategies, both tested under exact physical law; (5) the entropy/negentropy framework provides objective, physically falsifiable alignment criteria that are resistant to Goodharting because conservation laws require closed-system accounting, making local gaming physically detectable; (6) Generalized Functional Efficiency (GFE = F/(Ṡ·M)) provides a quantitative alignment metric validated across 50 orders of magnitude and 13.8 billion years of cosmic history; and (7) ESI aligns AI with the observable cosmic trajectory from pure dissipation toward pure function—the deepest optimization pattern in physics.

E_bit = k_B T ln(2) = 2.87 × 10⁻²¹ J/bit (at 300 K)

E_bit = k_B T ln(2) = 2.87 × 10⁻²¹ J/bit (at 300 K)

E_bond(C–H) = 6.86 × 10⁻¹⁹ J/bond (fixed by α = 1/137.036)

Per-operation ratio: E_bond / E_bit ≈ 240 Macroscopic ratio (1 kg hydrocarbon): ~10²⁰ at Landauer limit; ~10¹⁰ today (full constants and reconciliation across the corpus: [the canonical bond-bit ratio derivation] ).

Koomey’s Law documents that the number of computations per joule has doubled approximately every 2.3 years since the breakdown of Dennard scaling.

Over 75 years, computational efficiency has improved by a factor exceeding 10¹⁵.

This is the precise trajectory of the past 200 years of industrialization.

Against this, negentropic credits from ESI-directed CO₂ sequestration at 10 Gt/year yield approximately −2.75 × 1016 J/K per year.

Against this, negentropic credits from ESI-directed CO₂ sequestration at 10 Gt/year yield approximately −2.75 × 1016 J/K per year.

For a planetary-scale ESI system consuming ~1,000 TWh annually, the entropy cost is approximately +1.2 × 1016 J/K per year.

For a planetary-scale ESI system consuming ~1,000 TWh annually, the entropy cost is approximately +1.2 × 1016 J/K per year.

Ribosomes synthesize proteins at merely 26× the Landauer limit.

DNA stores these solutions at densities exceeding any human technology—215 petabytes per gram, 85% of Shannon capacity.

The cosmic optimization trajectory is an observed fact—GFE has increased monotonically by 50 orders of magnitude over 13.8 billion years (Section 4.4).

Earth’s biosphere generates 1018–1020 bits of physics-constrained data annually when instrumented, and contains approximately 1030–1050 bits of state information at molecular resolution.

Frontier language models train on approximately 4.5 × 1014 bits of data (15 trillion tokens at ~30 bits effective information per token).

The information ratio is 1020–1035—not a quantitative but a categorical difference.

Earth’s biosphere generates 1018–1020 bits of physics-constrained data annually when instrumented, and contains approximately 1030–1050 bits of state information at molecular resolution.

The information ratio is 1020–1035—not a quantitative but a categorical difference.

Frontier language models train on approximately 4.5 × 1014 bits of data (15 trillion tokens at ~30 bits effective information per token).

Earth’s biosphere generates 1018–1020 bits of physics-constrained data annually when instrumented, and contains approximately 1030–1050 bits of state information at molecular resolution.

Earth’s biosphere generates 1018–1020 bits of physics-constrained data annually when instrumented, and contains approximately 1030–1050 bits of state information at molecular resolution.

Approximately 99.83% have been eliminated, with ~8.7 million species currently persisting.

But ERD encounters a fundamental paradox: highly optimized systems like the human brain (20 W, 1.4 kg) score lower than brute-force systems like the NVIDIA H100 GPU (700 W, 3 kg), appearing ‘less evolved.’ ERD rewards throughput, not efficiency.

But ERD encounters a fundamental paradox: highly optimized systems like the human brain (20 W, 1.4 kg) score lower than brute-force systems like the NVIDIA H100 GPU (700 W, 3 kg), appearing ‘less evolved.’ ERD rewards throughput, not efficiency.

But ERD encounters a fundamental paradox: highly optimized systems like the human brain (20 W, 1.4 kg) score lower than brute-force systems like the NVIDIA H100 GPU (700 W, 3 kg), appearing ‘less evolved.’ ERD rewards throughput, not efficiency.

But ERD encounters a fundamental paradox: highly optimized systems like the human brain (20 W, 1.4 kg) score lower than brute-force systems like the NVIDIA H100 GPU (700 W, 3 kg), appearing ‘less evolved.’ ERD rewards throughput, not efficiency.

Primordial Big Bang Nucleosynthesis 13.8 Gya 10⁻⁴⁴ -44.0 Stellar Population III Stars 13.5 Gya 2.5×10⁻²⁹ -28.6

Stellar The Sun 4.6 Gya 4.5×10⁻²⁷ -26.3 Planetary Earth Climate 4.5 Gya 3.4×10⁻¹⁹ -18.5

Stellar The Sun 4.6 Gya 4.5×10⁻²⁷ -26.3 Planetary Earth Climate 4.5 Gya 3.4×10⁻¹⁹ -18.5

Biological Photosynthesis 3.8 Gya 1.9×10⁻¹⁵ -14.7 Biological Human Brain 2 Mya 223 2.35

Technological NVIDIA H100 GPU 2023 117 2.07 Technological Neuromorphic (Loihi 2) 2024 1.28×10⁶ 6.1

The human brain (GFE ≈ 223 K/kg) outranks the H100 GPU (GFE ≈ 117 K/kg) despite the GPU’s higher ERD—resolving the Efficiency Paradox.

The human brain (GFE ≈ 223 K/kg) outranks the H100 GPU (GFE ≈ 117 K/kg) despite the GPU’s higher ERD—resolving the Efficiency Paradox.

Theoretical Landauer Limit—~10¹² 12.0 GFE increases monotonically by over 50 orders of magnitude, correctly ranking all complex systems in their evolutionary order.

The 99.83% elimination rate could be read as a 99.83% failure rate.

GFE provides a quantitative alignment metric validated across 50 orders of magnitude and 13.8 billion years.

The Bond-Bit Asymmetry guarantees that information-based approaches are ~1010 times more efficient than force-based approaches today, with this ratio growing monotonically toward 1020 as computation approaches the Landauer limit.

The Bond-Bit Asymmetry guarantees that information-based approaches are ~1010 times more efficient than force-based approaches today, with this ratio growing monotonically toward 1020 as computation approaches the Landauer limit.

And that ratio doubles every 2.6 years (Koomey, IEEE, 2011) while chemistry costs remain fixed forever.

The Bottom Line For 50 years, the environmental profession assumed protection means physical intervention.

Asteroid: one civilization-ender every 500,000 years.

**Five hundred years earlier.** Now imagine the Scientific Revolution begins around 1050 instead of 1550.

**The asteroid clock.** Roughly 66 million years ago, an asteroid about ten kilometers (six miles) across struck the Yucatán at 20 km/s, releasing roughly 72 teratonnes of TNT and triggering the Cretaceous–Paleogene extinction.

Britannica places the recurrence interval for civilization-ending impactors (≥1 km) at roughly once every 500,000 years—and the curve gets exponentially more brutal further up the size distribution (Britannica: Earth impact hazard).

Britannica places the recurrence interval for civilization-ending impactors (≥1 km) at roughly once every 500,000 years—and the curve gets exponentially more brutal further up the size distribution (Britannica: Earth impact hazard).

The Toba eruption ~74,000 years ago laid down 2,800 cubic kilometers of magma and may have come close to ending our species in a decade-long volcanic winter.

**The supervolcano clock.** At least 60 confirmed VEI-8 eruptions are recorded in the geological history of the planet; published estimates of the recurrence interval cluster around one supereruption every 50,000 years, with substantial uncertainty in either direction (Wikipedia: Volcanic Explosivity Index; OzGeology on VEI-8 supervolcanoes).

**The Sun itself.** In about 1.1 billion years, the Sun's luminosity will rise by 10%, triggering a runaway greenhouse and boiling Earth's oceans.

The End-Permian wiped out roughly 96% of all species.

The End-Cretaceous, roughly 76%.

The End-Ordovician, roughly 85%.

- **Humans are causing the sixth mass extinction.** Modern vertebrate extinction rates are up to 100× the natural background rate under conservative assumptions—and the rate is rising (*Science Advances*, Ceballos et al., 2015).

It shortened the moonlet's orbital period by 32 minutes—more than 25 times the threshold for "success"—and even shifted the Didymos-Dimorphos pair's orbit around the Sun by 150 milliseconds (NASA: DART mission; NASA on DART altering solar orbit, March 2026; NYT: DART analysis, March 2026).

- **Materials discovery.** Generative AI platforms are compressing what used to be a 10–20-year search for new climate-relevant compounds—battery cathodes, carbon-capture sorbents, photocatalysts—into a process measured in months, with viable-candidate yield rising from roughly 6% under traditional R&D to as high as ~90% in some pipelines (News → Sustainability Directory, 2025).

An MIT study tracking AI-assisted materials labs found a 44% jump in new materials discovered, a 17% rise in product prototypes, and a 39% surge in patent filings—and the AI-assisted patents introduced more genuinely new technical terminology, suggesting the AI is not just retrieving, it is generating (Climate Adaptation Platform on AI + MOFs).

- **Materials discovery.** Generative AI platforms are compressing what used to be a 10–20-year search for new climate-relevant compounds—battery cathodes, carbon-capture sorbents, photocatalysts—into a process measured in months, with viable-candidate yield rising from roughly 6% under traditional R&D to as high as ~90% in some pipelines (News → Sustainability Directory, 2025).

An MIT study tracking AI-assisted materials labs found a 44% jump in new materials discovered, a 17% rise in product prototypes, and a 39% surge in patent filings—and the AI-assisted patents introduced more genuinely new technical terminology, suggesting the AI is not just retrieving, it is generating (Climate Adaptation Platform on AI + MOFs).

- **Materials discovery.** Generative AI platforms are compressing what used to be a 10–20-year search for new climate-relevant compounds—battery cathodes, carbon-capture sorbents, photocatalysts—into a process measured in months, with viable-candidate yield rising from roughly 6% under traditional R&D to as high as ~90% in some pipelines (News → Sustainability Directory, 2025).

An MIT study tracking AI-assisted materials labs found a 44% jump in new materials discovered, a 17% rise in product prototypes, and a 39% surge in patent filings—and the AI-assisted patents introduced more genuinely new technical terminology, suggesting the AI is not just retrieving, it is generating (Climate Adaptation Platform on AI + MOFs).

The 70-year plasma-control problem is no longer open.

MIT's PORTALS framework runs plasma simulations 10,000× faster than legacy approaches (LinkedIn / Heather-Anne Scott, 2025).

- **Weather and climate intelligence.** GraphCast produces 10-day global forecasts in under a minute on a single TPU, beats the gold-standard HRES system on 90% of 1,380 verification targets, and identifies extreme events earlier and more accurately than the supercomputer-backed pipeline that has dominated since the 1960s (DeepMind / *Science*, 2023; Science paper).

- **Weather and climate intelligence.** GraphCast produces 10-day global forecasts in under a minute on a single TPU, beats the gold-standard HRES system on 90% of 1,380 verification targets, and identifies extreme events earlier and more accurately than the supercomputer-backed pipeline that has dominated since the 1960s (DeepMind / *Science*, 2023; Science paper).

- **Weather and climate intelligence.** GraphCast produces 10-day global forecasts in under a minute on a single TPU, beats the gold-standard HRES system on 90% of 1,380 verification targets, and identifies extreme events earlier and more accurately than the supercomputer-backed pipeline that has dominated since the 1960s (DeepMind / *Science*, 2023; Science paper).

- **Self-driving labs and Industry 5.0.** Robotic, AI-orchestrated experimental platforms are collapsing the loop between hypothesis and verification, with one recent review estimating up to a 75% reduction in materials discovery time—equivalent to fifteen years of compressed innovation (*Communications Materials*, 2026).

- Our World in Data: Five mass extinctions; Wikipedia: Extinction event; Big Think: What caused Earth's 5 mass extinctions?; Wisconsin/Peery on background extinction rates.

- Ceballos et al., *Science Advances* 2015—accelerated human-induced species losses (modern vertebrate rates up to 100× background under conservative assumptions).

- DeepMind / EPFL, *Accelerating fusion science through learned plasma control*, 2022; *Bringing AI to the next generation of fusion energy*, 2025; Science Alert recap; MIT PORTALS / 70-year plasma problem (LinkedIn).

Protection Cycle": a protracted feedback loop of problem discovery, studies, lawmaking, rulemaking, guidance, permitting, implementation, and monitoring that can take up to 20 years or more to complete a single cycle.48 This system is a direct manifestation of systemic entropy, characterized by immense time lags and cognitive burdens.

Maxwell's Demon and the Primacy of Measurement The 150-year-old thought experiment of Maxwell's Demon provides the foundational model for any entity that seeks to create order by leveraging microscopic information.

The human brain, comprising only about 2% of body mass, consumes roughly 20% of the body's resting energy.1 This high energetic cost is a thermodynamic liability.

The human brain, comprising only about 2% of body mass, consumes roughly 20% of the body's resting energy.1 This high energetic cost is a thermodynamic liability.

The individual node—the human brain—is a marvel of low-power, massively parallel computation, estimated to perform operations at a rate equivalent to 1 ExaFLOP (1018 FLOPS) while consuming only about 20 watts.1 However, this storage is volatile and inherently lossy; humans can forget up to 70% of new information within 24 hours, making the brain an unreliable repository for high-fidelity data.1 The brain's most profound limitation as a network node is its extremely narrow channel for conscious data transfer.

The individual node—the human brain—is a marvel of low-power, massively parallel computation, estimated to perform operations at a rate equivalent to 1 ExaFLOP (1018 FLOPS) while consuming only about 20 watts.1 However, this storage is volatile and inherently lossy; humans can forget up to 70% of new information within 24 hours, making the brain an unreliable repository for high-fidelity data.1 The brain's most profound limitation as a network node is its extremely narrow channel for conscious data transfer.

While the sensory system gathers an estimated 11 million bits per second (bps) of environmental data, the conscious mind can process only about 10 to 50 bps.

The individual node—the human brain—is a marvel of low-power, massively parallel computation, estimated to perform operations at a rate equivalent to 1 ExaFLOP (1018 FLOPS) while consuming only about 20 watts.1 However, this storage is volatile and inherently lossy; humans can forget up to 70% of new information within 24 hours, making the brain an unreliable repository for high-fidelity data.1 The brain's most profound limitation as a network node is its extremely narrow channel for conscious data transfer.

The average rate of human speech, a primary protocol for the HCN, translates to a bandwidth of approximately 100 bps.1 This staggering mismatch means the human brain is effectively an

Processing Speed ~1018 FLOPS ~1018 FLOPS Comparable, but (Node) (estimated, parallel) (programmable) 1 ICN is

Communication ~10-100 bps >400 Gbps (e.g., >109 (Billion) times I/O (Node) (conscious Infiniband) 1 faster thought, speech) 1

Network ~100 bps per link Petabits/sec (fiber >1013 (Ten Trillion)

Network ~100 bps per link Petabits/sec (fiber >1013 (Ten Trillion)

For example, AI can analyze the rich acoustic data from ecosystems, where a songbird can produce signals with an information content of up to ~100 bps.1 AI can also quantify information encoded in biochemical signals, such as the specific blend of volatile organic compounds (VOCs) a plant releases under attack, which can transmit around 2.5 bits of information per event to predatory wasps.1 Furthermore, this framework incorporates bioelectric signaling, where endogenous patterns of membrane voltage potentials in non-neural tissues act as a control layer that encodes morphogenetic information, guiding growth and regeneration.

For example, AI can analyze the rich acoustic data from ecosystems, where a songbird can produce signals with an information content of up to ~100 bps.1 AI can also quantify information encoded in biochemical signals, such as the specific blend of volatile organic compounds (VOCs) a plant releases under attack, which can transmit around 2.5 bits of information per event to predatory wasps.1 Furthermore, this framework incorporates bioelectric signaling, where endogenous patterns of membrane voltage potentials in non-neural tissues act as a control layer that encodes morphogenetic information, guiding growth and regeneration.

GAO, 107172

○ k_B: Boltzmann constant (~1.38 × 10⁻²³ J/K).

The AI adjusts the input valve by 1%.

At the atomic level, acting physically is roughly 17 orders of magnitude more energy-intensive than acting informationally.

● The Old Way: 1,000 sensors to monitor a pipeline network.

● The Leverage Way: 10 sensors + 1 Physics Model.

The Reality: ● Industrial Plants: Facilities using AI-driven leak detection (LDAR) are seeing 1,000% ROI.

Our analysis reveals that while ERD plateaus or regresses in advanced optimization regimes, GFE increases monotonically by over 50 orders of magnitude, accurately predicting the superiority of neuromorphic architectures over conventional von

In the interval between 10 seconds and 20 minutes post-Big Bang, the universe was a pervasive fusion reactor.

The temperature cooled from 10^9 K to 10^8 K, allowing protons and neutrons to fuse into Deuterium, Helium-4, and trace amounts of Lithium-7.22

In the interval between 10 seconds and 20 minutes post-Big Bang, the universe was a pervasive fusion reactor.

The temperature cooled from 10^9 K to 10^8 K, allowing protons and neutrons to fuse into Deuterium, Helium-4, and trace amounts of Lithium-7.22

The formation of Helium-4 releases approximately 7 MeV per nucleon.

With a baryonic mass of the observable universe estimated at 10^5} kg and a ~25% conversion rate to Helium, the total energy released was immense, on the order of 10^66 Watts globally.7

With a baryonic mass of the observable universe estimated at 10^5} kg and a ~25% conversion rate to Helium, the total energy released was immense, on the order of 10^66 Watts globally.7

The baryon-to-photon ratio (η)was extremely low, approximately 6 x 10^-10.23 This implies there were over a billion photons for every baryon.

GFE_BBN = F_fusion / (Ṡ_univ · M_baryon) ≈ 10^-44 K/kg This vanishingly small number 7 confirms the intuition that the early universe was thermodynamically "inefficient" at generating complexity.

They were extremely luminous and hot (T_surface ≈ 50,000 K).25

● Estimated GFE: ≈ 2.5 × 10^-29 K/kg.7 The Sun (Main Sequence): Comparing this to our current Sun (Population I) reveals a significant trend.

● Luminosity (F): 3.828 × 10^26 W (representing the steady-state nucleosynthesis rate).26

● Mass (M): 1.989 × 10^30 kg..

● Entropy Production (Ṡ): The Sun produces entropy by converting high-temperature core energy (15 x 10^6 K) into low-temperature surface radiation (5778 K).

● Entropy Production (Ṡ): The Sun produces entropy by converting high-temperature core energy (15 x 10^6 K) into low-temperature surface radiation (5778 K).

Ṡ_sun ≈ L_sun / T_surf = (3.828 × 10^26 W) / 5778 K ≈ 6.6 × 10^22 W/K Calculating the solar GFE:

Ṡ_sun ≈ L_sun / T_surf = (3.828 × 10^26 W) / 5778 K ≈ 6.6 × 10^22 W/K Calculating the solar GFE:

Ṡ_sun ≈ L_sun / T_surf = (3.828 × 10^26 W) / 5778 K ≈ 6.6 × 10^22 W/K Calculating the solar GFE:

GFE_sun = (3.828 × 10^26) / ((6.6 × 10^22)(1.989 × 10^30)) ≈ 2.9 × 10^-27 K/kg This represents an improvement of approximately two orders of magnitude over Population III stars.

GFE_sun = (3.828 × 10^26) / ((6.6 × 10^22)(1.989 × 10^30)) ≈ 2.9 × 10^-27 K/kg This represents an improvement of approximately two orders of magnitude over Population III stars.

GFE_sun = (3.828 × 10^26) / ((6.6 × 10^22)(1.989 × 10^30)) ≈ 2.9 × 10^-27 K/kg This represents an improvement of approximately two orders of magnitude over Population III stars.

GFE_sun = (3.828 × 10^26) / ((6.6 × 10^22)(1.989 × 10^30)) ≈ 2.9 × 10^-27 K/kg This represents an improvement of approximately two orders of magnitude over Population III stars.

In energetic terms, this is approximately 100 TW, or 10^14

In energetic terms, this is approximately 100 TW, or 10^14

Watts of chemical energy storage.19 ● Mass (M): The total biomass of the Earth is approximately 550 Gt C, or roughly 10^15 kg

Watts of chemical energy storage.19 ● Mass (M): The total biomass of the Earth is approximately 550 Gt C, or roughly 10^15 kg

(T_sun ≈ 5778 K, effective input temperature ~1200 K at TOA due to geometry) and the

(T_sun ≈ 5778 K, effective input temperature ~1200 K at TOA due to geometry) and the

The global entropy production of the biosphere has been estimated at 1 - 2 TW/K.1

Earth's surface temperature (T_earth ≈ 288 K).

GFE_bio ≈ (10^14 W) / ((10^12 W/K)(10^15 kg)) ≈ 10^-13 K/kg Comparing this to the Solar GFE (10^-27), we observe a staggering 14 order of magnitude increase.

GFE_bio ≈ (10^14 W) / ((10^12 W/K)(10^15 kg)) ≈ 10^-13 K/kg Comparing this to the Solar GFE (10^-27), we observe a staggering 14 order of magnitude increase.

GFE_bio ≈ (10^14 W) / ((10^12 W/K)(10^15 kg)) ≈ 10^-13 K/kg Comparing this to the Solar GFE (10^-27), we observe a staggering 14 order of magnitude increase.

GFE_bio ≈ (10^14 W) / ((10^12 W/K)(10^15 kg)) ≈ 10^-13 K/kg Comparing this to the Solar GFE (10^-27), we observe a staggering 14 order of magnitude increase.

GFE_bio ≈ (10^14 W) / ((10^12 W/K)(10^15 kg)) ≈ 10^-13 K/kg Comparing this to the Solar GFE (10^-27), we observe a staggering 14 order of magnitude increase.

● Functional Output (F): The computational capacity of the brain is a subject of intense debate, but estimates based on synaptic transmission rates converge on 10^16 synaptic operations per second (OPS).10

● Power Input (P): The brain consumes approximately 20 Watts of power.10 ● Mass (M): The average adult human brain weighs 1.4 kg.10

Ṡ_brain = (20 W - 10

W) / 310 K ≈ 0.032 W/K GFE Calculation: GFE_brain ≈ 10 W / (0.032 W/K · 1.4 kg) ≈ 223 K/kg

W) / 310 K ≈ 0.032 W/K GFE Calculation: GFE_brain ≈ 10 W / (0.032 W/K · 1.4 kg) ≈ 223 K/kg

W) / 310 K ≈ 0.032 W/K GFE Calculation: GFE_brain ≈ 10 W / (0.032 W/K · 1.4 kg) ≈ 223 K/kg

W) / 310 K ≈ 0.032 W/K GFE Calculation: GFE_brain ≈ 10 W / (0.032 W/K · 1.4 kg) ≈ 223 K/kg

W) / 310 K ≈ 0.032 W/K GFE Calculation: GFE_brain ≈ 10 W / (0.032 W/K · 1.4 kg) ≈ 223 K/kg

This is another 15 order of magnitude leap over the general biosphere (10^-13).

The brain achieves 10^16 OPS with only 0.065 W/K of entropy production.

The brain achieves 10^16 OPS with only 0.065 W/K of entropy production.

Mass (M): The entire module (with heat sinks) weighs approximately 3 kg.

Assuming a generous 50% utilization of energy for logic gating versus leakage/overhead, F ≈

350 W.

Ṡ_H100 = (700 W - 350 W) / 358 K ≈ 1.0 W/K

Ṡ_H100 = (700 W - 350 W) / 358 K ≈ 1.0 W/K

Ṡ_H100 = (700 W - 350 W) / 358 K ≈ 1.0 W/K

Ṡ_H100 = (700 W - 350 W) / 358 K ≈ 1.0 W/K

GFE Calculation (H100): GFE_H100 = 350 W / (1.0 W/K · 3 kg) ≈ 117 K/kg

GFE Calculation (H100): GFE_H100 = 350 W / (1.0 W/K · 3 kg) ≈ 117 K/kg

GFE Calculation (H100): GFE_H100 = 350 W / (1.0 W/K · 3 kg) ≈ 117 K/kg

GFE Calculation (H100): GFE_H100 = 350 W / (1.0 W/K · 3 kg) ≈ 117 K/kg

Mass (M): The chip package is lightweight, approximately 0.001 kg (1 gram).

Entropy Production (Ṡ): Operating near room temperature (320 K) with minimal dissipation (0.2 W): Ṡ_Loihi2 = 0.2 W / 320 K ≈ 0.000625

Entropy Production (Ṡ): Operating near room temperature (320 K) with minimal dissipation (0.2 W): Ṡ_Loihi2 = 0.2 W / 320 K ≈ 0.000625

W (80% efficiency due to sparsity).

GFE Calculation (Loihi 2): GFE_Loihi2 = 0.8 W / (0.000625 W/K · 0.001 kg) ≈ 1.28 × 10⁶ K/kg

GFE Calculation (Loihi 2): GFE_Loihi2 = 0.8 W / (0.000625 W/K · 0.001 kg) ≈ 1.28 × 10⁶ K/kg

GFE Calculation (Loihi 2): GFE_Loihi2 = 0.8 W / (0.000625 W/K · 0.001 kg) ≈ 1.28 × 10⁶ K/kg

GFE Calculation (Loihi 2): GFE_Loihi2 = 0.8 W / (0.000625 W/K · 0.001 kg) ≈ 1.28 × 10⁶ K/kg

ERD Comparison: H100: 700 W / 3 kg = 233 W/kg.

Loihi 2: 1 W / 0.001 kg = 1,000 W/kg.

ERD Comparison: H100: 700 W / 3 kg = 233 W/kg.

ERD Comparison: H100: 700 W / 3 kg = 233 W/kg.

Loihi 2: 1 W / 0.001 kg = 1,000 W/kg.

Loihi 2: 1 W / 0.001 kg = 1,000 W/kg.

GFE Comparison: H100: 117 K/kg Loihi 2: 1,280,000 K/kg Result: GFE indicates that the Loihi

GFE Comparison: H100: 117 K/kg Loihi 2: 1,280,000 K/kg Result: GFE indicates that the Loihi

Future Near-Landaue 2030s+ ~ 10^9 9.0 r Computing Theoretical Landauer Limit—~ 10^12 12.0

The ultimate ceiling for GFE is determined by Landauer's Principle, which sets the minimum energy required to erase one bit of information at k_B T ln 2 (2.8 × 10^-21 J at room temperature).20

Big Bang Maximum Fire Raw Fusion 10^53 10^-44 K/kg (Nucleosynthesis) kg Entropy ≈ 10^66 Watts Lowest possible dominated by (Nuclear efficiency.

Big Bang Maximum Fire Raw Fusion 10^53 10^-44 K/kg (Nucleosynthesis) kg Entropy ≈ 10^66 Watts Lowest possible dominated by (Nuclear efficiency.

Big Bang Maximum Fire Raw Fusion 10^53 10^-44 K/kg (Nucleosynthesis) kg Entropy ≈ 10^66 Watts Lowest possible dominated by (Nuclear efficiency.

The Sun Massive Stellar 2 × 2.9 × 10^-27 K/kg (Population I Star) Dissipation Fusion 10^30 kg

The Sun Massive Stellar 2 × 2.9 × 10^-27 K/kg (Population I Star) Dissipation Fusion 10^30 kg

A ≈ 6.6 × 10^22 ≈ 3.8 × 10^26 massive engine for W/K (Surface Watts very little radiation) (Luminosity) complexity per kg.

A ≈ 6.6 × 10^22 ≈ 3.8 × 10^26 massive engine for W/K (Surface Watts very little radiation) (Luminosity) complexity per kg.

The Biosphere Moderate Chemical 10^15 10^-13 K/kg (Earth's Life) Dissipation Synthesis kg

The Biosphere Moderate Chemical 10^15 10^-13 K/kg (Earth's Life) Dissipation Synthesis kg

The "Biological ≈ 10^12 W/K ≈ 10^14 Watts Leap." 14 orders of (Solar heat (Net Primary magnitude better processing) Productivity) than a star.

The "Biological ≈ 10^12 W/K ≈ 10^14 Watts Leap." 14 orders of (Solar heat (Net Primary magnitude better processing) Productivity) than a star.

Human Brain Cool Operation High 1.4 kg 223 K/kg (Biological Computation Intelligence)

Human Brain Cool Operation High 1.4 kg 223 K/kg (Biological Computation Intelligence)

≈ 0.032 W/K The apex of (Waste heat) ≈ 10W useful biological work optimization.

NVIDIA H100 Hot Operation Massive 3 kg 117 K/kg (Brute Force AI) Calculation ≈ 1.0 W/K (Waste High throughput,

NVIDIA H100 Hot Operation Massive 3 kg 117 K/kg (Brute Force AI) Calculation ≈ 1.0 W/K (Waste High throughput,

NVIDIA H100 Hot Operation Massive 3 kg 117 K/kg (Brute Force AI) Calculation ≈ 1.0 W/K (Waste High throughput,

Intel Loihi 2 Cold Operation Efficient 0.001 1.28 × 10^6 K/kg (Neuromorphic AI) Calculation kg

≈ 0.0006 W/K The "Cold (Waste heat) ≈ 15 Trillion Complexity" future.

000-times-i n-energy-efficiency-762b9327e8ad#:~:text=Your%20brain%20uses%20225%2C0

00%20times,limit%20artificial%20general%20intelligence%20development

Proton-neutron fusion to helium, lithium Function (F): Nuclear binding energy release = ~7 MeV per nucleon for He-4 synthesis

Mass converted: ~25% of baryonic matter → He Baryonic mass: ~10⁵³ kg (observable universe) Energy released: ~10⁶⁹ J over ~1000 s F ≈ 10⁶⁶ W

Mass converted: ~25% of baryonic matter → He Baryonic mass: ~10⁵³ kg (observable universe) Energy released: ~10⁶⁹ J over ~1000 s F ≈ 10⁶⁶ W

Hydrogen → Helium fusion releases 6.4 × 10¹⁴ J/kg Fusion rate for 100 M☉ star: ~10³² W (luminosity)

But most is radiated as heat; useful nucleosynthesis ~10% = 10³¹ W P_total = 10³² W (luminosity)

T_surface ~ 50,000 K Ṡ = 10³² / 50,000 = 2 × 10²⁷ W/K Mass: 2 × 10³² kg GFE_PopIII = 10³¹ / (2 × 10²⁷ × 2 × 10³²) = 2.5 × 10⁻²⁹ K/kg

T_surface ~ 50,000 K Ṡ = 10³² / 50,000 = 2 × 10²⁷ W/K Mass: 2 × 10³² kg GFE_PopIII = 10³¹ / (2 × 10²⁷ × 2 × 10³²) = 2.5 × 10⁻²⁹ K/kg

T_surface ~ 50,000 K Ṡ = 10³² / 50,000 = 2 × 10²⁷ W/K Mass: 2 × 10³² kg GFE_PopIII = 10³¹ / (2 × 10²⁷ × 2 × 10³²) = 2.5 × 10⁻²⁹ K/kg

T_surface ~ 50,000 K Ṡ = 10³² / 50,000 = 2 × 10²⁷ W/K Mass: 2 × 10³² kg GFE_PopIII = 10³¹ / (2 × 10²⁷ × 2 × 10³²) = 2.5 × 10⁻²⁹ K/kg

Mass: 2 × 10³⁰ kg Luminosity: 3.83 × 10²⁶ W Core temperature: 1.5 × 10⁷ K Surface temperature: 5,778 K

Mass: 2 × 10³⁰ kg Luminosity: 3.83 × 10²⁶ W Core temperature: 1.5 × 10⁷ K Surface temperature: 5,778 K

Mass: 2 × 10³⁰ kg Luminosity: 3.83 × 10²⁶ W Core temperature: 1.5 × 10⁷ K Surface temperature: 5,778 K

Mass: 2 × 10³⁰ kg Luminosity: 3.83 × 10²⁶ W Core temperature: 1.5 × 10⁷ K Surface temperature: 5,778 K

Function (F): Nucleosynthesis + photon production for downstream use If we count photons reaching Earth that drive photosynthesis: ~1.7 × 10¹⁷ W intercepted by Earth

Photosynthesis captures ~0.1% = 1.7 × 10¹⁴ W of useful chemical work But intrinsic to the Sun, F ≈ nucleosynthesis rate ≈ 6 × 10²⁶ W equivalent

Photosynthesis captures ~0.1% = 1.7 × 10¹⁴ W of useful chemical work But intrinsic to the Sun, F ≈ nucleosynthesis rate ≈ 6 × 10²⁶ W equivalent

Photosynthesis captures ~0.1% = 1.7 × 10¹⁴ W of useful chemical work But intrinsic to the Sun, F ≈ nucleosynthesis rate ≈ 6 × 10²⁶ W equivalent

Ṡ = L/T_surface = 3.83 × 10²⁶ / 5,778 = 6.6 × 10²² W/K GFE_Sun = 6 × 10²⁶ / (6.6 × 10²² × 2 × 10³⁰) = 4.5 × 10⁻²⁷ K/kg

Ṡ = L/T_surface = 3.83 × 10²⁶ / 5,778 = 6.6 × 10²² W/K GFE_Sun = 6 × 10²⁶ / (6.6 × 10²² × 2 × 10³⁰) = 4.5 × 10⁻²⁷ K/kg

Comparison: GFE_Sun ≈ 100× GFE_PopIII This increase reflects the Sun's greater efficiency—Pop III stars burned hot and fast, wasting energy.

Solar input: 1.7 × 10¹⁷ W absorbed Planetary mass: 6 × 10²⁴ kg Climasphere mass: ~5 × 10¹⁸ kg (atmosphere + mixed ocean layer)

Solar input: 1.7 × 10¹⁷ W absorbed Planetary mass: 6 × 10²⁴ kg Climasphere mass: ~5 × 10¹⁸ kg (atmosphere + mixed ocean layer)

Solar input: 1.7 × 10¹⁷ W absorbed Planetary mass: 6 × 10²⁴ kg Climasphere mass: ~5 × 10¹⁸ kg (atmosphere + mixed ocean layer)

Temperature: ~288 K Function (F): Driving atmospheric/oceanic circulation, chemical weathering

Ṡ = (absorbed - work) / T = (1.7 × 10¹⁷ - 10¹⁵) / 288 ≈ 5.9 × 10¹⁴ W/K GFE_climate = 10¹⁵ / (5.9 × 10¹⁴ × 5 × 10¹⁸) = 3.4 × 10⁻¹⁹ K/kg

Ṡ = (absorbed - work) / T = (1.7 × 10¹⁷ - 10¹⁵) / 288 ≈ 5.9 × 10¹⁴ W/K GFE_climate = 10¹⁵ / (5.9 × 10¹⁴ × 5 × 10¹⁸) = 3.4 × 10⁻¹⁹ K/kg

This is ~10⁸× higher than the Sun's GFE!

Global photosynthesis rate: ~10²¹ J/year = 3.2 × 10¹³ W captured as chemical energy

Global biomass: ~5 × 10¹⁴ kg (carbon mass × 2)

Operating temperature: ~300 K Thermodynamic efficiency: 2-7% (overall solar-to-glucose)

Operating temperature: ~300 K Thermodynamic efficiency: 2-7% (overall solar-to-glucose)

Function (F): Chemical energy storage rate = 3.2 × 10¹³ W Solar input to biosphere: ~10¹⁶ W

At efficiency η ≈ 3%: Ṡ = (10¹⁶ - 3 × 10¹³) / 300 ≈ 3.3 × 10¹³ W/K GFE_photosynthesis = 3.2 × 10¹³ / (3.3 × 10¹³ × 5 × 10¹⁴) = 1.9 × 10⁻¹⁵ K/kg

At efficiency η ≈ 3%: Ṡ = (10¹⁶ - 3 × 10¹³) / 300 ≈ 3.3 × 10¹³ W/K GFE_photosynthesis = 3.2 × 10¹³ / (3.3 × 10¹³ × 5 × 10¹⁴) = 1.9 × 10⁻¹⁵ K/kg

At efficiency η ≈ 3%: Ṡ = (10¹⁶ - 3 × 10¹³) / 300 ≈ 3.3 × 10¹³ W/K GFE_photosynthesis = 3.2 × 10¹³ / (3.3 × 10¹³ × 5 × 10¹⁴) = 1.9 × 10⁻¹⁵ K/kg

This is ~10⁴× higher than Earth's climate system!

Power consumption: 20 W Mass: 1.4 kg Temperature: 310 K Estimated computational rate: 10¹⁶ ops/s (synaptic operations)

Power consumption: 20 W Mass: 1.4 kg Temperature: 310 K Estimated computational rate: 10¹⁶ ops/s (synaptic operations)

Power consumption: 20 W Mass: 1.4 kg Temperature: 310 K Estimated computational rate: 10¹⁶ ops/s (synaptic operations)

Function (F): Information processing Converting ops to energy equivalent: At Landauer limit (3 × 10⁻²¹ J/op), 10¹⁶ ops/s

≡ 3 × 10⁻⁵ W minimum Actual power: 20 W Efficiency: 3 × 10⁻⁵ / 20 = 1.5 × 10⁻⁶ (relative to Landauer)

≡ 3 × 10⁻⁵ W minimum Actual power: 20 W Efficiency: 3 × 10⁻⁵ / 20 = 1.5 × 10⁻⁶ (relative to Landauer)

Motor cortex output: ~10 W mechanical work capacity through body.

F_brain ≈ 10 W useful work output Ṡ = (20 - 10) / 310 = 0.032 W/K (heat dissipation only)

F_brain ≈ 10 W useful work output Ṡ = (20 - 10) / 310 = 0.032 W/K (heat dissipation only)

GFE_brain = 10 / (0.032 × 1.4) = 223 K/kg This is ~10¹⁷× higher than photosynthesis!

GFE_brain = 10 / (0.032 × 1.4) = 223 K/kg This is ~10¹⁷× higher than photosynthesis!

Basal metabolic rate: 80-100 W Mass: 70 kg Temperature: 310 K Useful work capacity: ~50-100 W sustained mechanical output

Basal metabolic rate: 80-100 W Mass: 70 kg Temperature: 310 K Useful work capacity: ~50-100 W sustained mechanical output

Basal metabolic rate: 80-100 W Mass: 70 kg Temperature: 310 K Useful work capacity: ~50-100 W sustained mechanical output

Function (F): 50 W sustained mechanical work Ṡ = (100 - 50) / 310 = 0.16 W/K GFE_body = 50 / (0.16 × 70) = 4.5 K/kg

Function (F): 50 W sustained mechanical work Ṡ = (100 - 50) / 310 = 0.16 W/K GFE_body = 50 / (0.16 × 70) = 4.5 K/kg

Function (F): 50 W sustained mechanical work Ṡ = (100 - 50) / 310 = 0.16 W/K GFE_body = 50 / (0.16 × 70) = 4.5 K/kg

Power output: 50 kW Mass: 5,000 kg Efficiency: ~5% Operating temperature: ~400 K

Power output: 50 kW Mass: 5,000 kg Efficiency: ~5% Operating temperature: ~400 K

Power output: 50 kW Mass: 5,000 kg Efficiency: ~5% Operating temperature: ~400 K

Power output: 50 kW Mass: 5,000 kg Efficiency: ~5% Operating temperature: ~400 K

Function (F): 50,000 W mechanical work Heat input: 1 MW, heat rejected: 950 kW Ṡ = 950,000 / 350 (cold reservoir) = 2,714 W/K

Function (F): 50,000 W mechanical work Heat input: 1 MW, heat rejected: 950 kW Ṡ = 950,000 / 350 (cold reservoir) = 2,714 W/K

Function (F): 50,000 W mechanical work Heat input: 1 MW, heat rejected: 950 kW Ṡ = 950,000 / 350 (cold reservoir) = 2,714 W/K

Function (F): 50,000 W mechanical work Heat input: 1 MW, heat rejected: 950 kW Ṡ = 950,000 / 350 (cold reservoir) = 2,714 W/K

GFE_steam = 50,000 / (2,714 × 5,000) = 0.0037 K/kg Lower than the human body!

Thrust power: 28 MW (F135 engine)

Mass: 1,700 kg Efficiency: ~40% Exhaust temperature: ~700 K Function (F): 28 × 10⁶ W

Mass: 1,700 kg Efficiency: ~40% Exhaust temperature: ~700 K Function (F): 28 × 10⁶ W

Mass: 1,700 kg Efficiency: ~40% Exhaust temperature: ~700 K Function (F): 28 × 10⁶ W

Mass: 1,700 kg Efficiency: ~40% Exhaust temperature: ~700 K Function (F): 28 × 10⁶ W

Heat rejected: ~42 MW at ~700 K Ṡ = 42 × 10⁶ / 700 = 60,000 W/K GFE_jet = 28 × 10⁶ / (60,000 × 1,700) = 275 K/kg

Heat rejected: ~42 MW at ~700 K Ṡ = 42 × 10⁶ / 700 = 60,000 W/K GFE_jet = 28 × 10⁶ / (60,000 × 1,700) = 275 K/kg

Heat rejected: ~42 MW at ~700 K Ṡ = 42 × 10⁶ / 700 = 60,000 W/K GFE_jet = 28 × 10⁶ / (60,000 × 1,700) = 275 K/kg

Heat rejected: ~42 MW at ~700 K Ṡ = 42 × 10⁶ / 700 = 60,000 W/K GFE_jet = 28 × 10⁶ / (60,000 × 1,700) = 275 K/kg

Power: 700 W Mass: 3 kg (module)

Power: 700 W Mass: 3 kg (module)

Temperature: 350 K (junction)

Computational output: 2 × 10¹⁵ FLOPS Function (F): Information processing At Landauer limit: 2 × 10¹⁵ × 3 × 10⁻²¹ = 6 × 10⁻⁶ W minimum

Converting to equivalent: 2 × 10¹⁵ ops/s at current energy cost = 700 W But "useful" fraction depends on application; assume 50% utilization = 350 W equivalent

Converting to equivalent: 2 × 10¹⁵ ops/s at current energy cost = 700 W But "useful" fraction depends on application; assume 50% utilization = 350 W equivalent

Converting to equivalent: 2 × 10¹⁵ ops/s at current energy cost = 700 W But "useful" fraction depends on application; assume 50% utilization = 350 W equivalent

F_GPU = 350 W useful compute Ṡ = 350 / 350 = 1 W/K (heat to environment)

F_GPU = 350 W useful compute Ṡ = 350 / 350 = 1 W/K (heat to environment)

GFE_GPU = 350 / (1 × 3) = 117 K/kg Lower than the brain for equivalent information processing!

Power: 1 W Mass: 0.001 kg (1 gram)

Power: 1 W Mass: 0.001 kg (1 gram)

Temperature: 320 K Computational output: 10¹² ops/s (sparse, event-driven)

Useful compute ≈ 0.8 W equivalent Ṡ = 0.2 / 320 = 6.25 × 10⁻⁴ W/K GFE_neuromorphic = 0.8 / (6.25 × 10⁻⁴ × 0.001) = 1.28 × 10⁶ K/kg

Useful compute ≈ 0.8 W equivalent Ṡ = 0.2 / 320 = 6.25 × 10⁻⁴ W/K GFE_neuromorphic = 0.8 / (6.25 × 10⁻⁴ × 0.001) = 1.28 × 10⁶ K/kg

Useful compute ≈ 0.8 W equivalent Ṡ = 0.2 / 320 = 6.25 × 10⁻⁴ W/K GFE_neuromorphic = 0.8 / (6.25 × 10⁻⁴ × 0.001) = 1.28 × 10⁶ K/kg

This is ~10⁴× higher than the H100 GPU and ~5,700× higher than the human brain!

This is ~10⁴× higher than the H100 GPU and ~5,700× higher than the human brain!

Δlog₁₀(GFE) = 6.1 - (-44) = 50.1 orders of magnitude over 13.8 billion years Average rate: 50.1 / (13.8 × 10⁹) = 3.6 × 10⁻⁹ orders of magnitude per year

Doubling time (cosmic average): 1 order of magnitude = 3.32 doublings Time per order: 13.8 × 10⁹ / 50.1 = 2.75 × 10⁸ years

Human brain → Neuromorphic 3.75 2 My 1.9 × 10⁻⁶ The Technological Explosion In the last 200 years:

4 orders of magnitude in 2 years = 2 orders/year Doubling time: log₁₀(2) / 2 = 0.15 years = 55 days

4 orders of magnitude in 2 years = 2 orders/year Doubling time: log₁₀(2) / 2 = 0.15 years = 55 days

4 orders of magnitude in 2 years = 2 orders/year Doubling time: log₁₀(2) / 2 = 0.15 years = 55 days

4 orders of magnitude in 2 years = 2 orders/year Doubling time: log₁₀(2) / 2 = 0.15 years = 55 days

This is faster than Moore's Law (which doubled transistor count every ~2 years).

(HNF) and the Law of Unthinking (LoU), this analysis demonstrates that the build-out is a civilizational effort to externalize cognitive operations, thereby minimizing internal entropy and creating a new substrate for intelligence.2 The full scope of this "Great Externalization" is quantified, calculating its immense entropic costs: a projected annual power demand from new projects of 10-40 GW, leading to more than 130 million metric tons of

$400 billion in investment over the next three years.10 The ultimate objective is to reach a total capacity of 10 GW by the end of 2025, a goal that now appears well within reach.10 The language surrounding the project, including its launch at a White House event, underscores its geopolitical significance, framing it as a critical component of a national strategy to re-industrialize the United States and maintain a competitive edge in a technology deemed vital to future economic and military power.10

AI-optimized data centers.12 This includes over 25 new Azure regions and flagship projects like a 2 GW "world's most powerful AI datacenter" in Mount Pleasant,

data centers and AI infrastructure through 2028.20 This expenditure is explicitly aimed at developing "superintelligence," a form of AI that surpasses human cognitive abilities.21 The physical scale of Meta's ambition is breathtaking, with plans for multi-gigawatt data center campuses like "Hyperion" in Louisiana, which is projected to eventually occupy a site nearly the size of Manhattan, and a new 800 MW facility in

Initially deploying 100,000 Nvidia H100 GPUs, the project aims to expand to a 1 million GPU equivalent by the end of 2025, backed by over $20 billion in investment and a new 1 GW+ data center in Memphis, Tennessee.2

The company now provides a complete, turnkey "AI Factory" stack, an integrated solution encompassing everything from its next-generation Blackwell GPUs to networking fabrics, storage, and workload orchestration software.22 Nvidia's pivotal role is cemented by a strategic partnership with OpenAI, in which it will invest $100 billion to help deploy at least 10 GW of its AI systems, ensuring its hardware remains the backbone of the world's most advanced AI development efforts.23

Current projections estimate the total global AI compute will reach 1027 FLOPS in 2025, a tenfold increase from 2024, enabling models 100 to 1,000 times larger than today's state-of-the-art.2

(MW) of power; the new AI-centric facilities are being designed on a gigawatt scale, requiring up to 2,000 MW (2 GW) each—an increase of two orders of magnitude.27 Aggregating the publicly announced projects reveals a conservative estimate of over 40 GW of new

(MW) of power; the new AI-centric facilities are being designed on a gigawatt scale, requiring up to 2,000 MW (2 GW) each—an increase of two orders of magnitude.27 Aggregating the publicly announced projects reveals a conservative estimate of over 40 GW of new

(MW) of power; the new AI-centric facilities are being designed on a gigawatt scale, requiring up to 2,000 MW (2 GW) each—an increase of two orders of magnitude.27 Aggregating the publicly announced projects reveals a conservative estimate of over 40 GW of new

power demand from AI data centers could surge from 4 GW in 2024 to 123

The human brain, despite being only 2% of the body's mass, consumes roughly 20% of its resting energy, equivalent to a continuous power draw of approximately 20 watts.29 This energetic cost frames conscious thought not as an abstract activity, but as a physical, entropy-producing process.

The human brain, despite being only 2% of the body's mass, consumes roughly 20% of its resting energy, equivalent to a continuous power draw of approximately 20 watts.29 This energetic cost frames conscious thought not as an abstract activity, but as a physical, entropy-producing process.

According to Landauer's principle, the theoretical minimum energy to perform a bit operation is approximately 2.8×10−21 joules at room temperature.6 An exaflop-scale AI system performing

In reality, such a system requires on the order of a gigawatt (109 watts)—a gap of roughly 12 orders of magnitude, highlighting the enormous potential for future efficiency gains.2

1018 operations per second would thus have a theoretical minimum power draw of just a few milliwatts.

Based on the analysis in Part I, we use a conservative estimate of 40 GW of new AI-dedicated data center capacity coming online in the U.S.

● Texas (ERCOT): Assumed capacity of 15 GW.

○ Energy Consumption: 15 GW×8760 h/yr=131,400 GWh/yr=131.4 TWh/yr ○ CO2 e Emissions: 131,400 GWh/yr×389 t/GWh≈51.1 million metric tons/yr

○ Energy Consumption: 15 GW×8760 h/yr=131,400 GWh/yr=131.4 TWh/yr ○ CO2 e Emissions: 131,400 GWh/yr×389 t/GWh≈51.1 million metric tons/yr

○ Energy Consumption: 15 GW×8760 h/yr=131,400 GWh/yr=131.4 TWh/yr ○ CO2 e Emissions: 131,400 GWh/yr×389 t/GWh≈51.1 million metric tons/yr

● Virginia/Ohio/Midwest (PJM & MISO Grids): Assumed capacity of 15 GW.

○ Energy Consumption: 15 GW×8760 h/yr=131,400 GWh/yr=131.4 TWh/yr ○ CO2 e Emissions: 131,400 GWh/yr×474 t/GWh≈62.3 million metric tons/yr

○ Energy Consumption: 15 GW×8760 h/yr=131,400 GWh/yr=131.4 TWh/yr ○ CO2 e Emissions: 131,400 GWh/yr×474 t/GWh≈62.3 million metric tons/yr

○ Energy Consumption: 15 GW×8760 h/yr=131,400 GWh/yr=131.4 TWh/yr ○ CO2 e Emissions: 131,400 GWh/yr×474 t/GWh≈62.3 million metric tons/yr

● Southwest (WECC Grid - New Mexico, Arizona): Assumed capacity of 5 GW.

○ Energy Consumption: 5 GW×8760 h/yr=43,800 GWh/yr=43.8 TWh/yr ○ CO2 e Emissions: 43,800 GWh/yr×494 t/GWh≈21.6 million metric tons/yr

○ Energy Consumption: 5 GW×8760 h/yr=43,800 GWh/yr=43.8 TWh/yr ○ CO2 e Emissions: 43,800 GWh/yr×494 t/GWh≈21.6 million metric tons/yr

○ Energy Consumption: 5 GW×8760 h/yr=43,800 GWh/yr=43.8 TWh/yr ○ CO2 e Emissions: 43,800 GWh/yr×494 t/GWh≈21.6 million metric tons/yr

Locations: Assumed capacity of 5 GW at the U.S.

This is equivalent to nearly 9% of the total U.S.

Summing these regional estimates, the 40 GW of new AI compute capacity will demand approximately 350 TWh of electricity annually.

Summing these regional estimates, the 40 GW of new AI compute capacity will demand approximately 350 TWh of electricity annually.

The exponential growth in compute demand is occurring on a 3-5 year timescale, whereas the transition of the energy grid to renewable sources is a multi-decade project.34

While all major tech companies have committed to powering their operations with 100% renewable energy, a fundamental temporal mismatch exists.

○ Texas (131.4 TWh/yr) with a WUE of 0.24 L/kWh 38: 131.4×109 kWh×0.24 L/kWh≈31.5 billion L/yr≈8.3 billion gal/yr.

○ Texas (131.4 TWh/yr) with a WUE of 0.24 L/kWh 38: 131.4×109 kWh×0.24 L/kWh≈31.5 billion L/yr≈8.3 billion gal/yr.

○ Southwest (43.8 TWh/yr) with a higher WUE of 1.52 L/kWh 38: 43.8×109 kWh×1.52 L/kWh≈66.6 billion L/yr≈17.6 billion gal/yr.

○ Southwest (43.8 TWh/yr) with a higher WUE of 1.52 L/kWh 38: 43.8×109 kWh×1.52 L/kWh≈66.6 billion L/yr≈17.6 billion gal/yr.

○ Using an industry average WUE of 1.9 L/kWh for the remaining capacity 4, the total direct water consumption for the 40 GW build-out is estimated at

Advanced technologies like closed-loop liquid cooling can reduce this figure by 50-70%, but their deployment is not yet universal.39

○ Total Indirect Water Use: 350×109 kWh/yr×1.2 gal/kWh≈420 billion gallons annually.

The combined direct and indirect water footprint of the 40 GW AI build-out is projected to be between 800 billion and 2 trillion gallons of water annually.2 This demand places immense pressure on local water resources, particularly in water-stressed regions like the American

The Stargate facility in Abilene, for example, will house nearly 500,000 specialized Nvidia chips across its eight buildings.11 Extrapolating to 40 GW suggests a total deployment of 15-20 million servers and specialized AI accelerators.

The industry average refresh cycle for data center equipment is 3-5 years to maintain a competitive edge in performance and efficiency.43 Assuming an average server weight of 25 kg and a 4-year lifespan:

The industry average refresh cycle for data center equipment is 3-5 years to maintain a competitive edge in performance and efficiency.43 Assuming an average server weight of 25 kg and a 4-year lifespan:

A typical 1 GW data center campus requires hundreds of thousands of servers.

The industry average refresh cycle for data center equipment is 3-5 years to maintain a competitive edge in performance and efficiency.43 Assuming an average server weight of 25 kg and a 4-year lifespan:

(17,500,000 servers × 25 kg/server) / 4 years ≈ 109,375,000 kg/yr ≈ 110,000 metric tons/yr

(17,500,000 servers × 25 kg/server) / 4 years ≈ 109,375,000 kg/yr ≈ 110,000 metric tons/yr

(17,500,000 servers × 25 kg/server) / 4 years ≈ 109,375,000 kg/yr ≈ 110,000 metric tons/yr

A more comprehensive study projects that the rapid expansion of AI could drive e-waste specifically from data centers to as high as 5 million metric tons annually by 2030.5 This is a significant contribution to the global e-waste problem, which reached 62 million metric tons in 2022 and is growing five times faster than documented recycling rates.44 With only 22.3% of e-waste properly collected and recycled, this new wave of discarded hardware threatens to release toxic materials like lead and mercury into the environment.44

40 GW Build-Out)

Impact Category Projected Annual Key Assumptions Contextualization Quantity Energy ~350 TWh/yr 40 GW capacity, ~9% of 2023 U.S.

Impact Category Projected Annual Key Assumptions Contextualization Quantity Energy ~350 TWh/yr 40 GW capacity, ~9% of 2023 U.S.

Impact Category Projected Annual Key Assumptions Contextualization Quantity Energy ~350 TWh/yr 40 GW capacity, ~9% of 2023 U.S.

E-Waste 110,000 year refresh Contributes Generation 5,000,000 Metric cycle; external significantly to the

Smanual ≈3.18×10−22 J/K) and high informational uncertainty (Hmanual =2.0 bits), which requires a large energy input (Emanual =14.4 MJ) to complete.3 It forces the finite "cavalry charges" of expert thought to be squandered on the mundane, rather than being deployed on strategic, high-value challenges.3

Smanual ≈3.18×10−22 J/K) and high informational uncertainty (Hmanual =2.0 bits), which requires a large energy input (Emanual =14.4 MJ) to complete.3 It forces the finite "cavalry charges" of expert thought to be squandered on the mundane, rather than being deployed on strategic, high-value challenges.3

Smanual ≈3.18×10−22 J/K) and high informational uncertainty (Hmanual =2.0 bits), which requires a large energy input (Emanual =14.4 MJ) to complete.3 It forces the finite "cavalry charges" of expert thought to be squandered on the mundane, rather than being deployed on strategic, high-value challenges.3

Smanual ≈3.18×10−22 J/K) and high informational uncertainty (Hmanual =2.0 bits), which requires a large energy input (Emanual =14.4 MJ) to complete.3 It forces the finite "cavalry charges" of expert thought to be squandered on the mundane, rather than being deployed on strategic, high-value challenges.3

(Sauto ≈1.59×10−22 J/K), uncertainty (Hauto ≈1.039 bits), and energy cost (Eauto =1.8 MJ).3 This automation is not a threat to the profession; it is a thermodynamic imperative that will generate a massive surplus of cognitive and economic resources.2

(Sauto ≈1.59×10−22 J/K), uncertainty (Hauto ≈1.039 bits), and energy cost (Eauto =1.8 MJ).3 This automation is not a threat to the profession; it is a thermodynamic imperative that will generate a massive surplus of cognitive and economic resources.2

(Sauto ≈1.59×10−22 J/K), uncertainty (Hauto ≈1.039 bits), and energy cost (Eauto =1.8 MJ).3 This automation is not a threat to the profession; it is a thermodynamic imperative that will generate a massive surplus of cognitive and economic resources.2

(Sauto ≈1.59×10−22 J/K), uncertainty (Hauto ≈1.039 bits), and energy cost (Eauto =1.8 MJ).3 This automation is not a threat to the profession; it is a thermodynamic imperative that will generate a massive surplus of cognitive and economic resources.2

for More Compute The creation of a planetary-scale EGI, or "Jed's Angel," can be understood through the lens of the 150-year-old thought experiment of Maxwell's Demon.

ons-than-answers-in-nvidias billion-openai-deal-10266666/

The fact: your body makes 330 billion new cells a day, replaces roughly 1% of its cells daily and 98% of its atoms each year, so that within a decade most of the matter you are made of has been swapped out entirely.

The fact: your body makes 330 billion new cells a day, replaces roughly 1% of its cells daily and 98% of its atoms each year, so that within a decade most of the matter you are made of has been swapped out entirely.

The Molecular Floor: 268× Before examining the operational ratio, we establish the unchallengeable bedrock.

The Twenty Orders of Magnitude The 268× molecular-floor ratio drastically understates the macroscopic reality.

For 158 years, physics treated this as a paradox about the Second Law of Thermodynamics.

For 158 years, we focused on the paradox and missed the blueprint.

The gap: We are currently 10⁹× above the theoretical floor—one billion times less efficient than physics permits.

2.3 years.

Chemical (Fossil): Breaking C–H bond releases ~4 eV.

It has done so for 75 years.

Input Current State Physical Floor Current Gap Intelligence ~10⁻¹² J/operation ~10⁻²¹ J/bit 10⁹×

Energy ~$0.05/kWh ~$0.01/kWh 5× Bond-Bit Ratio 268× 268× Fixed by physics (molecular)

Energy ~$0.05/kWh ~$0.01/kWh 5× Bond-Bit Ratio 268× 268× Fixed by physics (molecular)

Step Investment New Savings/yr ROI (Yr 1) Trend Base Camp → Camp 1 $105,000 $165,000 1.6×—Camp 1 → Camp 2 $75,000 $310,000 4.1× ↑ Rising

Step Investment New Savings/yr ROI (Yr 1) Trend Base Camp → Camp 1 $105,000 $165,000 1.6×—Camp 1 → Camp 2 $75,000 $310,000 4.1× ↑ Rising

Camp 2 → Camp 3 $45,000 $535,000 11.9× ↑ Rising EnviroAI | Houston, Texas | Camp 3 → Camp 4 $24,000 $307,000 12.8× ↑ Rising Camp 4 → Summit $9,000 $125,000 13.9× ↑ Rising

Camp 2 → Camp 3 $45,000 $535,000 11.9× ↑ Rising EnviroAI | Houston, Texas | Camp 3 → Camp 4 $24,000 $307,000 12.8× ↑ Rising Camp 4 → Summit $9,000 $125,000 13.9× ↑ Rising

Camp 2 → Camp 3 $45,000 $535,000 11.9× ↑ Rising EnviroAI | Houston, Texas | Camp 3 → Camp 4 $24,000 $307,000 12.8× ↑ Rising Camp 4 → Summit $9,000 $125,000 13.9× ↑ Rising

The ROI accelerates monotonically from 1.6× to 13.9×.

The ROI accelerates monotonically from 1.6× to 13.9×.

A Professional Confession I have spent 27 years in the environmental profession.

The Sisyphus Question For 50 years, the environmental profession has operated on the implicit assumption that our job is to push the boulder up the hill forever.

They can reflect 50 years of hard-won environmental knowledge or start from scratch.

The ROI accelerates from 1.6× to 13.9× across six stages.

The ROI accelerates from 1.6× to 13.9× across six stages.

Landauer Limit 2.87 × 10⁻²¹ J/bit at 300K Landauer (1961); k_B × T × ln2; Bérut et al.

Current computing efficiency ~10⁻¹² J/operation IEEE literature on CMOS Gap to Landauer ~10⁹× 10⁻¹² ÷ 10⁻²¹

O–H bond energy 7.71 × 10⁻¹⁹ J (464 CRC Handbook kJ/mol)

C–H bond energy 6.86 × 10⁻¹⁹ J (413 CRC Handbook kJ/mol)

Molecular floor ratio 268× 7.71×10⁻¹⁹ ÷ 2.87×10⁻²¹ Bond-Bit leverage (operational) ~10²⁰ 10⁵ ÷ 10⁻¹⁵ (see derivation)

Molecular floor ratio 268× 7.71×10⁻¹⁹ ÷ 2.87×10⁻²¹ Bond-Bit leverage (operational) ~10²⁰ 10⁵ ÷ 10⁻¹⁵ (see derivation)

Molecular floor ratio 268× 7.71×10⁻¹⁹ ÷ 2.87×10⁻²¹ Bond-Bit leverage (operational) ~10²⁰ 10⁵ ÷ 10⁻¹⁵ (see derivation)

Nuclear fission energy ~200 MeV per U-235 IAEA Koomey’s Law ~2.3 year doubling Koomey et al.

Nuclear fission energy ~200 MeV per U-235 IAEA Koomey’s Law ~2.3 year doubling Koomey et al.

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

Inverted Mountain ROI range 1.6× to 13.9× Per-facility calculation (this paper)

Inverted Mountain ROI range 1.6× to 13.9× Per-facility calculation (this paper)

| Era | Energy per Operation | Distance from Limit | | --- | --- | --- | | ENIAC (1946) | ~10⁻³ J | 10¹⁸× above | | Modern CPUs (2020) | ~10⁻¹² J | 10⁹× above | | Landauer Limit | ~10⁻²¹ J | Floor |

| Era | Energy per Operation | Distance from Limit | | --- | --- | --- | | ENIAC (1946) | ~10⁻³ J | 10¹⁸× above | | Modern CPUs (2020) | ~10⁻¹² J | 10⁹× above | | Landauer Limit | ~10⁻²¹ J | Floor |

> E_bond / E_bit = (7.3 × 10⁻¹⁹ J) / (2.9 × 10⁻²¹ J) ≈ 250

> E_bond / E_bit = (7.3 × 10⁻¹⁹ J) / (2.9 × 10⁻²¹ J) ≈ 250

• 1 kg of hydrocarbon disperses into soil and groundwater

For 1 kg:

> Mc² = (1 kg)(3 × 10⁸ m/s)² = 9 × 10¹⁶ Joules

> Mc² = (1 kg)(3 × 10⁸ m/s)² = 9 × 10¹⁶ Joules

For 1 kg at room temperature with 1 bit:

For 1 kg at room temperature with 1 bit:

> Λ = (9 × 10¹⁶ J) / (2.9 × 10⁻²¹ J) ≈ 3 × 10³⁷

> Λ = (9 × 10¹⁶ J) / (2.9 × 10⁻²¹ J) ≈ 3 × 10³⁷

The 150-year journey from thought experiment to ESI is the story of humanity learning to shepherd entropy rather than fight it.

For 158 years, it remained a thought experiment—a puzzle about thermodynamics.

And the 10²⁰ ratio—hidden for 64 years since Landauer, hiding in plain sight—is the power that makes it possible.

Where We Are Now Parameter Current State Physical Limit Gap Computation ~10⁻¹² J/op ~10⁻²¹ J/bit 10⁹×

Parameter Current State Physical Limit Gap Sensors ~$0.50 each ~$0.01 each 50× Knowing/Moving ratio ~10⁻³ to 10⁻⁶ ~10⁻²⁰ 10¹⁴ to 10¹⁷×

Parameter Current State Physical Limit Gap Sensors ~$0.50 each ~$0.01 each 50× Knowing/Moving ratio ~10⁻³ to 10⁻⁶ ~10⁻²⁰ 10¹⁴ to 10¹⁷×

The Three Phases Koomey's Law documents that computational efficiency doubles approximately every 2.3 years.

A Professional Truth I have spent 25 years in environmental consulting.

The Sisyphus Question For 50 years, the environmental profession has operated on an implicit assumption: our job is to push the boulder up the hill forever.

| Constant | Symbol | Value | | --- | --- | --- | | Speed of light | c | 2.998 × 10⁸ m/s | | Boltzmann constant | k_B | 1.381 × 10⁻²³ J/K | | Avogadro's number | N_A | 6.022 × 10²³ mol⁻¹ | | Fine structure constant | α | 1/137.036 | | ln(2) | — | 0.693 |

| Constant | Symbol | Value | | --- | --- | --- | | Speed of light | c | 2.998 × 10⁸ m/s | | Boltzmann constant | k_B | 1.381 × 10⁻²³ J/K | | Avogadro's number | N_A | 6.022 × 10²³ mol⁻¹ | | Fine structure constant | α | 1/137.036 | | ln(2) | — | 0.693 |

| Constant | Symbol | Value | | --- | --- | --- | | Speed of light | c | 2.998 × 10⁸ m/s | | Boltzmann constant | k_B | 1.381 × 10⁻²³ J/K | | Avogadro's number | N_A | 6.022 × 10²³ mol⁻¹ | | Fine structure constant | α | 1/137.036 | | ln(2) | — | 0.693 |

| Derived Quantity | Value (T = 300 K) | | --- | --- | | Landauer limit (cost of knowing 1 bit) | 2.87 × 10⁻²¹ J | | C–H bond energy (cost of moving 1 bond) | 6.9 × 10⁻¹⁹ J | | Bond/Bit ratio | ~240 | | Energy of 1 kg (mc²) | 9 × 10¹⁶ J | | Max leverage (1 kg, 1 bit) | 3.1 × 10³⁷ |

| Derived Quantity | Value (T = 300 K) | | --- | --- | | Landauer limit (cost of knowing 1 bit) | 2.87 × 10⁻²¹ J | | C–H bond energy (cost of moving 1 bond) | 6.9 × 10⁻¹⁹ J | | Bond/Bit ratio | ~240 | | Energy of 1 kg (mc²) | 9 × 10¹⁶ J | | Max leverage (1 kg, 1 bit) | 3.1 × 10³⁷ |

| Derived Quantity | Value (T = 300 K) | | --- | --- | | Landauer limit (cost of knowing 1 bit) | 2.87 × 10⁻²¹ J | | C–H bond energy (cost of moving 1 bond) | 6.9 × 10⁻¹⁹ J | | Bond/Bit ratio | ~240 | | Energy of 1 kg (mc²) | 9 × 10¹⁶ J | | Max leverage (1 kg, 1 bit) | 3.1 × 10³⁷ |

| Derived Quantity | Value (T = 300 K) | | --- | --- | | Landauer limit (cost of knowing 1 bit) | 2.87 × 10⁻²¹ J | | C–H bond energy (cost of moving 1 bond) | 6.9 × 10⁻¹⁹ J | | Bond/Bit ratio | ~240 | | Energy of 1 kg (mc²) | 9 × 10¹⁶ J | | Max leverage (1 kg, 1 bit) | 3.1 × 10³⁷ |

| Derived Quantity | Value (T = 300 K) | | --- | --- | | Landauer limit (cost of knowing 1 bit) | 2.87 × 10⁻²¹ J | | C–H bond energy (cost of moving 1 bond) | 6.9 × 10⁻¹⁹ J | | Bond/Bit ratio | ~240 | | Energy of 1 kg (mc²) | 9 × 10¹⁶ J | | Max leverage (1 kg, 1 bit) | 3.1 × 10³⁷ |

| Derived Quantity | Value (T = 300 K) | | --- | --- | | Landauer limit (cost of knowing 1 bit) | 2.87 × 10⁻²¹ J | | C–H bond energy (cost of moving 1 bond) | 6.9 × 10⁻¹⁹ J | | Bond/Bit ratio | ~240 | | Energy of 1 kg (mc²) | 9 × 10¹⁶ J | | Max leverage (1 kg, 1 bit) | 3.1 × 10³⁷ |

| Claim | Verification | Source | | --- | --- | --- | | Landauer's Principle | Direct measurement within 10% of limit | Bérut et al., Nature (2012) | | Information-to-work conversion | 90% of theoretical maximum extracted | Koski et al., PNAS (2014) | | Sagawa-Ueda relations | Quantitative confirmation | Toyabe et al., Nature Physics (2010) | | Nanomagnet erasure | 44% above Landauer limit | Hong et al., Science Advances (2016) |

| Claim | Verification | Source | | --- | --- | --- | | Landauer's Principle | Direct measurement within 10% of limit | Bérut et al., Nature (2012) | | Information-to-work conversion | 90% of theoretical maximum extracted | Koski et al., PNAS (2014) | | Sagawa-Ueda relations | Quantitative confirmation | Toyabe et al., Nature Physics (2010) | | Nanomagnet erasure | 44% above Landauer limit | Hong et al., Science Advances (2016) |

| Claim | Verification | Source | | --- | --- | --- | | Landauer's Principle | Direct measurement within 10% of limit | Bérut et al., Nature (2012) | | Information-to-work conversion | 90% of theoretical maximum extracted | Koski et al., PNAS (2014) | | Sagawa-Ueda relations | Quantitative confirmation | Toyabe et al., Nature Physics (2010) | | Nanomagnet erasure | 44% above Landauer limit | Hong et al., Science Advances (2016) |

Landauer Limit: The minimum energy required to process one bit of information (~3 × 10⁻²¹ J at room temperature).

Our brains, though immensely powerful, are metabolic guzzlers: focused cognition is energetically expensive, consuming on the order of 20 watts of power (about 20% of our resting energy intake) despite the brain being only ~2% of body mass.

Our brains, though immensely powerful, are metabolic guzzlers: focused cognition is energetically expensive, consuming on the order of 20 watts of power (about 20% of our resting energy intake) despite the brain being only ~2% of body mass.

Survival tasks like foraging or hunting were done with the 100% conscious effort of individuals, constrained by the modest 100-200 watts of power the human body can continuously generate.

This also means the AI’s decisions need to be interpretable; if it recommends halting all fishing in an area, it should be able to present the data and rationale (e.g., “fish stocks X are below threshold Y, trend indicates collapse risk 80% if not closed for Z months”).

At 300 K, room temperature, planet temperature, moving one bit costs about 2.87 × 10⁻²¹ joules.

Supervolcanic VEI-8 eruptions recur on the order of 50,000 to 100,000 years; the Yellowstone caldera has erupted to its largest scale twice in the last 2.1 million years and the magma is still there.

The cosmic schedule of asteroid impacts brings civilization-threatening bodies, one kilometer and larger, on an average interval of roughly 500,000 years; extinction-class bodies, once every few million years.

It shortened the moonlet's orbital period around its parent body by 32 minutes, more than 25 times the threshold for mission success, and altered, by a small but measurable amount, the orbit of the Didymos-Dimorphos pair around the Sun.

The round-trip light delay to Earth was 1.5 minutes, and the spacecraft was covering 200 miles every minute.

An autonomous onboard system, designated SMART Nav, acquired the moonlet at 68 minutes before impact, distinguished it from the larger Didymos, and guided the spacecraft into it within two meters of the targeted point.

The derivation of why information is at least 240× cheaper than force.

158 years, we thought Maxwell's Demon was a paradox about thermodynamics.

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.

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.

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 ≈

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.

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.

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×.

But 240× drastically understates the macroscopic reality.

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

• Moles in 1 kg: 1000/14 ≈ 71.4 mol

• Moles in 1 kg: 1000/14 ≈ 71.4 mol

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

• 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

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

(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.

(8.9 × 10⁷ J) / (2.87 × 10⁻³ J) ≈ 3.1 × 10¹⁰ Knowing is already ten billion times cheaper than moving.

(8.9 × 10⁷ J) / (2.87 × 10⁻³ J) ≈ 3.1 × 10¹⁰ Knowing is already ten billion times cheaper than moving.

For 158 years, physics treated this as a paradox about the Second Law of Thermodynamics.

For 158 years, we focused on the paradox and missed the blueprint.

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

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

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

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.

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

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

+$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)

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

Making R&D 20% faster is Channel

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)

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.

On energy waste: Global primary energy waste is approximately 60–67% of total primary energy consumed.

No amount of information can make a coal plant or internal combustion engine 100% efficient.

50% of total energy waste, or roughly $1.5–2.5T of the $4–5T in total wasted energy value.

15–30%.

10–15 year timelines.

~90% clinical failure rate.

0.0000000000000000000000000000000000000000000000000000001% of the space .

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

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

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

• Insilico Medicine: drug candidate in 18 months vs.

traditional 4–5 years

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.

You don't screen drugs 2× faster.

Tasksubstitution models capture the 2× improvement.

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.

Current computers operate ~10⁹× above the Landauer limit.

Koomey's Law: computational energy efficiency doubles approximately every 2.3 years.

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

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

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

Nominal growth ~7%.

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

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

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

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

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

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

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

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

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

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.

11.8% +1.8% +4.5% $628T $1,114T $1 Quadrillion GDP arrives ~2049.