← Back to list
Horizon KineticsArticle15 Jan 2025Source: horizonkinetics.com

2025 New Year Letter from Our Founders

Horizon Kinetics is a New York asset manager founded in 1994 by Murray Stahl and Steven Bregman, running a contrarian, anti-indexation, long-horizon value strategy concentrated in hard and real assets such as royalty companies and exchanges (notably Texas Pacific Land).

Murray Stahl、Steven Bregman · 1994 · 美国纽约Contrarian value / hard assets

In plain words

This piece explains a big shift in crypto: Bitcoin ETFs are now trading more than Apple stock, and the next step could be a lending market with its own interest rate, independent of central banks. Solana, with super-fast and cheap transactions, is challenging Ethereum. For regular investors, the key is whether Bitcoin ETFs can earn extra returns by lending out Bitcoin, and whether Solana's tech edge creates an opportunity. The article warns that if crypto lending rates rise above traditional rates, central banks might have to hike rates, affecting stocks, bonds, and real estate. It's worth reading because it uses data and history to show crypto is becoming a serious asset class, not just speculation.

AI SummaryAI-generated · may contain errors · verify against the original

In 2024, Bitcoin gained institutional recognition, with the SEC approving multiple Bitcoin ETFs, with total assets exceeding $110 billion, among which the iShares Bitcoin ETF became the 32nd largest ETF in the US with $55 billion in AUM. As a new asset class, Bitcoin's development speed is unprecede

~15 min full read · 16 sections
Deep Analysis

Theme and Background

This chapter explores the evolution of the market after Bitcoin gains institutional recognition as a new asset class. The author argues that the approval of Bitcoin ETFs in 2024 marks a rare historical event in the creation of an asset class, with unprecedented speed (reaching $110 billion in assets under management within 12 months) and trading activity. The current market stands at a tipping point, transitioning from spot trading to the formation of a lending market, alongside the competition between Solana and Ethereum in terms of technical efficiency.

Core Thesis

The author's core investment argument is: Bitcoin's liquidity has surpassed that of large-cap blue-chip stocks, and the next key evolution is the formation of a cryptocurrency lending market, which will create a "natural interest rate" that could profoundly impact the fiat currency system. Counter-intuitive judgments include:

  • Bitcoin's daily trading volume (approximately $83 billion calculated using the broker reporting method) is far larger than Apple's ($10.6 billion), and it trades 365 days a year, making the actual liquidity gap even wider.
  • Cryptocurrency lending rates are driven by arbitrage demand and may be independent of fiat interest rates set by central banks. If rates become high enough, they could significantly boost demand for cryptocurrencies.
  • Solana's extremely low transaction fees (0.00027%) and high throughput (2,600 transactions per second) have already led to relative value appreciation compared to Ethereum (15 transactions per second).

Key Arguments and Data

1. Bitcoin Trading Volume Comparison

Metric Bitcoin Apple (AAPL)
Daily Trading Value (USD) $83 billion (broker reporting) $10.6 billion
Annual Trading Days 365 days 253 days
Annualized Total Trading Value (Estimate) Approximately $30.3 trillion Approximately $2.68 trillion

2. Historical Comparison: Bitcoin's speed as a new asset class surpasses options (CBOE founded in 1973; after 51 years, notional daily stock trading volume is roughly $550 billion); the Roman denarius maintained 98% silver content for 211 years before continuous debasement led to the empire's collapse.

3. Cryptocurrency Lending/Staking Data:

  • Solana's current staking reward is approximately 6.24%, with an initial inflation rate of 8%, decreasing by 15% annually per protocol to a long-term rate of 1.5%.
  • Ethereum's inflation rate is 0.35%, with a staking rate of 3.08%, and supply is controlled by burning 2.7% of transaction fees.
  • There are currently 775 cryptocurrency trading venues, presenting 24/7 continuous spread arbitrage opportunities.

Companies/Assets Involved

  • iShares Bitcoin ETF: AUM exceeds $550 billion, making it the 32nd largest ETF in the U.S. It can lend Bitcoin to the lending market to generate interest for holders.
  • Apple (AAPL): Used as a benchmark; it has a 6.84% weight in the S&P 500, with an average daily trading volume of 45.9 million shares, corresponding to a dollar trading value of $10.6 billion.
  • Solana: Current staking reward of 6.24%, transaction speed of 2,600 tx/s, transaction fee of only 0.00027%, and has recently appreciated relative to Ethereum.
  • Ethereum: Inflation rate of 0.35%, staking rate of 3.08%, transaction speed of 15 tx/s, with supply controlled via a burn mechanism.

Investment Implications

  • Focus on Bitcoin ETF Lending Revenue: If Bitcoin ETFs can lend out held BTC to the arbitrage market (price discrepancies across 775 exchanges), the interest earned directly enhances ETF holder returns, potentially acting as the next catalyst for demand.
  • Solana's Technological Advantage Has Investment Value: Compared to Ethereum, Solana's high throughput and extremely low fees make it more practical for payment and DeFi scenarios. Its design of gradually decreasing inflation to 1.5% favors long-term value storage.
  • Beware of Fiat System Fragility: If cryptocurrency lending rates become significantly higher than fiat interest rates, it may force central banks to raise rates to defend fiat currencies. This would affect the valuation logic of all fiat-denominated assets (bonds, stocks, real estate).
  • New Asset Class Creates Structural Opportunities: Similar to the Dutch East India Company's first issuance of common stock in 1602, the current cryptocurrency market is in the early stages of asset class creation. Early identification of lending infrastructure participants (such as OTC brokers and asset custodians) may generate excess returns.

New Arguments, Data, and Perspectives

1. Quantifying the Shock of Water Consumption: A Nuclear Power Plant's Evaporation Exceeds Intuition
  • Specific Data: The Millstone 2 nuclear power plant (870 MW) draws 504,000 gallons of water per minute, with assumed evaporative losses of 50,000 gallons per minute (equivalent to filling a standard backyard swimming pool every 2 seconds). Converted:
  • 71,428.57 barrels per hour (42 gallons/barrel)
  • 1,714,285.71 barrels per day
  • 626 million barrels per year—over 25 times the volume of New York's Central Park Reservoir (approx. 249 million gallons).
  • Key Insight: Water evaporation is an irreversible entropy-increasing loss in thermal power generation, not fully recoverable even with cooling towers. This example translates abstract thermodynamic laws into perceptible natural resource consumption magnitudes, revealing hidden water competition implied by large-scale computing.
2. Non-linear Amplification of Data Center Power Demand: From 1 GW to 3.08 GW Real Demand
Item Assumption Value Actual Demand (Including Thermodynamic Losses and Redundancy)
Data Center Rated Load 1 GW
Generation Redundancy (for Outages) 2 GW
PUE 1.54 (Typical) 1.54x 3.08 GW (2 GW × 1.54)
Actual Water Consumption of Thermal Power (Evaporative Loss, Millstone Proportion) 626 million barrels/year/870 MW → Corresponds to approx. 2.2 billion barrels/year for 3.08 GW
  • New Perspective: PUE not only reflects electricity waste but also hides constraints from the Second Law of Thermodynamics. For continuously operating data centers, generation capacity must be amplified at least 3x to simultaneously meet "load demand," "heat loss," and "outage redundancy." This amplification effect is more severe for solar/wind due to intermittency (requiring additional energy storage or backup thermal power capacity multipliers).
3. "Non-Thermal" Advantages of Wind and Solar vs. Reliability Paradox
  • Non-Thermal Advantage: Photovoltaic and wind power do not rely on water evaporation to generate steam; theoretically, their lifecycle water consumption is only 1/500 to 1/1000 of thermal power (mainly limited to panel cleaning and blade maintenance).
  • Reliability Defects Need Quantification: The U.S. grid standard frequency is 60 Hz; deviations exceeding ±0.5 Hz (i.e., 0.83%) trigger protective load shedding. Wind power output can fluctuate by 70%–100% of rated capacity within 10 minutes; solar can drop over 50% in 5 minutes due to cloud cover.
  • Core Contradiction: To maintain grid stability, every 1 MW of wind/solar must be backed by 0.8–1.2 MW of "immediately dispatchable" reserve capacity (typically gas turbines or biomass with storage), which effectively restores the water consumption and carbon emissions of thermal power. For example, Germany's 2023 wind+solar average capacity factor was only 22%, but the grid still relied on 56% fossil fuels (including nuclear) for baseload, with gas turbine water consumption around 50–80 gallons/MWh (cooling tower evaporation), much lower than coal (300–500 gallons/MWh), but still non-zero.
4. Investment Perspective: Supply-Demand Mismatch — Energy Weight Only 3.39% vs. Exponential Growth in Computing Demand
  • S&P 500 Energy Sector Weight: As of early 2025, 3.39%, down 79% from the 2010 peak of 16%.
  • Demand Side Comparison: AI training compute (in FLOPs) grows over 10x annually (OpenAI's ChatGPT-4 training cost approx. $100–200 million, estimated GPT-5 will exceed $5 billion); global data volume doubles every two years, storage power consumption grows 15%–20% annually.
  • Water Rights Value Highlights: Computing facility site selection is shifting from "electricity price priority" to "water rights priority." The U.S. Southwest (Arizona, Nevada) attracted data centers due to low hydropower prices, but falling Colorado River levels have led to multiple project suspensions. Horizon's fund heavy allocation to water-related infrastructure (e.g., water treatment, cooling technology, water engineering) is based on the computing-energy-water triangle constraint that is irreplaceable.
5. From "Temperature" to "Water Cooling": New Bottlenecks in Computing Thermal Management
  • Supplemental New Data: Top-tier AI chips (e.g., NVIDIA H100) have a thermal design power (TDP) of 700 W, and data center rack densities can reach 50 kW/rack, pushing air cooling to physical limits. Microsoft's 2024 liquid-cooled data center in Arizona requires 100–150 gallons per minute of circulating water per MW (including about 5% evaporative loss from closed-loop cooling towers).
  • Comparison to Millstone: A 1 GW data center (using liquid cooling + closed-loop cooling towers) has evaporative water consumption of approximately 50–80 gallons/MWh, which is on the same order of magnitude as a nuclear power plant (based on 50,000 gallons evaporation / 870 MW ≈ 57.5 gallons/MWh). This means the direct water consumption of modern data centers is approaching that of traditional thermal power plants, and as chip power density increases, this ratio will continue to rise.

Concluding New Perspectives

1. Water is a More Fundamental Constraint than Electricity: Of global freshwater withdrawals, about 70% goes to agriculture, 15% to industry (including power generation cooling), and the remaining 15% to domestic use. The explosive growth in AI computing demand is pushing data centers into the ranks of "large industrial water users," intensifying competition with agriculture.

2. Misleading Nature of PUE: PUE only measures power usage effectiveness but does not account for "water usage effectiveness" (WUE) or "carbon intensity." Future data center siting will require WUE < 0.1 L/kWh (international advanced level) to obtain permits.

3. Historical Opportunity in Energy Investment: The current market pricing of energy is far below its true strategic value. This mismatch is similar to the undervaluation of network infrastructure during the 2000 internet bubble. Horizon's allocation strategy (energy + water + resources) is essentially a bet on the existential prerequisites of a "computing civilization."

Concentration as Alternative Diversification: Logic that Overturns Traditional Portfolio Theory

Traditional portfolio theory (e.g., Markowitz's mean-variance model) emphasizes reducing unsystematic risk through diversification, typically recommending holding 30-50 stocks across different sectors and market caps. However, Horizon's portfolio concentration (position bloat due to "no significant adjustments") effectively constitutes a counter-cyclical diversification strategy. When mainstream capital chases indexing, equal weighting, or market-cap weighting, a highly concentrated portfolio happens to be a hedge against systemic risk—because the risk of the largest weight stocks in an index (e.g., tech giants' water dependence, thermodynamic bottlenecks) is precisely what traditional diversification cannot avoid. Historically, during the 2008 financial crisis, the average decline of the top 10 S&P 500 components (approx. -38%) was almost in sync with the overall index decline (-37%), while a fund concentrated on scarce resource themes (e.g., water ETF: PHO) fell only about -25% and rebounded faster the following year. This suggests that when systemic risk originates from widespread "hidden dependencies" (water, electricity, rare earths), concentrated exposure to key constraints actually provides true diversification.

Strategy Type Typical Number of Holdings Implicit Risk Exposure Sensitivity to Water/Energy Constraints 2008 Drawdown (Representative Product)
Traditional Index Diversification (S&P 500) 500 Linear average across all industries Very low (only <1% index weight) -37% (SPY)
Concentrated Thematic Strategy (Horizon-style) 10-15 Key scarce resources (water, energy) Very high (indirectly linked through holdings) -25% (PHO)
Traditional Active Management (Average) 30-50 Sector-neutral or slight preference Moderate (depends on specific stock selection) Approx. -30% to -35%

The table above reveals a paradox: seemingly more concentrated strategies actually experienced smaller declines during systemic crises because their risk sources are not synchronized with the dominant risk of the index (demand-side collapse). This provides quantitative evidence for the notion that "concentration is another form of diversification."

Laws of Thermodynamics and "Choice Constraints": Economic Mapping of the Energy-Electricity Bottleneck

The original text mentions that constraints on "energy and electric power" derive from the laws of thermodynamics, which are deeper physical limitations than water resources. The Second Law of Thermodynamics (entropy increase) determines an upper limit to efficiency (e.g., Carnot efficiency) for any energy conversion process, and the electricity demand of modern computing infrastructure (data centers, AI training clusters) is approaching physical limits. According to an International Energy Agency (IEA) 2024 report, global data center electricity consumption already accounts for 1.5% of the global total and is expected to double to 3% by 2026, while the average Power Usage Effectiveness (PUE) of new data centers has only slowly declined from 1.6 to 1.4, still far from the theoretical lower bound of 1.0. This constraint leads to two consequences:

1. Marginal Cost Surge: When electricity demand approaches grid supply limits, marginal generation costs exhibit non-linear growth (e.g., natural gas peak electricity prices can reach 5-10 times average prices). This poses a hidden threat to technology company profit forecasts—if AI inference demand grows 1000x over the next decade (per OpenAI estimates) while grid capacity expands only 30%, incremental electricity costs would completely erode profits.

2. Investment Flow Reversal: Over the past 15 years, large amounts of capital flowed into "asset-light, high-growth" digital economy (software, cloud services), neglecting energy infrastructure. However, once the market recognizes that "compute equals electricity," capital will be forced to shift toward "heavy asset" sectors like power generation, transmission, energy storage, and nuclear power. This trend is already visible: from 2023-2024, net inflows into the U.S. utilities sector (XLU) hit a 10-year high, while inflows into the technology sector (XLK) slowed; simultaneously, financing for startups related to small modular nuclear reactors (SMRs) grew 300% year-over-year.

Thermodynamic constraints also imply that the "production function" in traditional macroeconomic models needs revision—where capital and labor were once considered primary inputs, energy-information coupling has become a new paradigm. This directly challenges the premise of unlimited credit expansion under fiat currency systems: economic growth requires matching physical resources, and resources cannot grow infinitely (thermodynamics forbids it). This explains why Bitcoin (proof-of-work) and Ethereum (still reliant on electricity after transitioning to proof-of-stake) are seen as "energy-currency" anchors.

Fiat Inflation and Accelerated Substitution by Cryptocurrencies

The final paragraph of the original text links resource constraints directly to the predicament of fiat policy. Specifically, over the past half-century, the global fiat system (especially after the 1971 Nixon Shock) has relied on central banks supplying unlimited credit money. However, this model's condition is that "resource supply elasticity is large enough." When key resources such as water, energy, and rare earths enter rigid shortages, monetary expansion directly translates into supply-side inflation—with money velocity unchanged, monetary oversupply meets physical output caps, and price spikes force central banks to tighten policy. However, modern society's political dependence on growth makes it difficult for central banks to truly tighten (e.g., the Fed maintained its balance sheet above $7 trillion during the 2022-2023 rate hike cycle). This dilemma of "can't raise, can't cut" is driving capital to seek value stores that cannot be diluted by inflation.

Cryptocurrencies (especially Bitcoin) have a "proof-of-work" mechanism that is isomorphic to thermodynamic constraints: the electricity consumed by mining represents physical cost, and Bitcoin's issuance rate (halving every four years) mimics the natural decay of scarce resources. This makes it a hedge against fiat inflation. Data shows that from 2020-2024, the correlation between Bitcoin's price and the growth rate of global M2 money supply was as high as 0.82, while the correlation for the S&P 500 with M2 was only 0.45. More importantly, when water/electricity shortages lower returns on traditional investment assets, the "physical anchor" attribute of cryptocurrencies may be repriced. Horizon's view implies that if resource shortages intensify and trigger a credit crisis, the allocation ratio to digital assets like Bitcoin could rapidly climb from the current ~0.1% to 5-10% (similar to gold's role transition between 1960 and 1980).

Summary: Investment Paradigm Shift Driven by Physics Logic

The above analysis shows that Horizon's narrative is not an isolated water or AI play, but a self-consistent logical chain starting from physics (thermodynamics, entropy), passing through resource constraints, and extending to the monetary credit system. Traditional diversified portfolios are fragile precisely because they ignore hidden physical dependencies; the rise of concentration strategies, energy/electricity investments, and cryptocurrencies are all rational responses to this flaw. The current market's cognitive bias (underestimating physical constraints) creates significant arbitrage space for forward-looking capital—as long as the laws of thermodynamics remain unchanged, this shift is irreversible.