Horizon Kinetics — 2nd Quarter 2026 Market Commentary

Solving an insolvable problem — beware the planted axiom

• The commentary opens on method: an explanation must not rest on a "planted axiom" — an unsupported assertion that makes an argument seem logical at the start while everything after rests on a false foundation. The worked example is the breezy sector call ("pulled back from its risk-on highs… consensus estimates have bottomed out"), which tells you only what the price did and what others think.
• Client questions overflowed a single sitting, so the quarter is organized by (i) divide and conquer — a separate roundtable on the AI data-center hyperscalers; (ii) a central theme — compounding and time; and (iii) alarm-bell questions that "pretty much require a response."

The Mag 7 at 150x free cash flow — and $108 billion of stock comp

• The quoted bull case is a Mag 7 P/E of 24.8, "cheaper than the 29.6 of 12 months ago." Horizon Kinetics reprices the same group on free cash flow: two members have negative FCF and one trades at 700x; generously calling the two negatives 100x so all seven can be averaged, the group trades at 150x run-rate free cash flow.
• Stock compensation was deliberately not added back — "if a company didn't utilize stock options, you'd think employees would demand cash instead, so it really is an operating expense" — and on a run-rate basis across the Mag 7 it amounts to $108 BILLION.
• The earlier roundtable on the five AI data-center hyperscalers (four of them Mag 7 members) featured analyst and private fund manager Fredrik Tjernstrom, who writes the firm's short-sale report The Devil's Advocate, with James Davolos. Tjernstrom's May fundholder letter was "essentially a sell recommendation" on those AI data-center companies.

The client questions that set the agenda

• Explicitly asked and answered here: the "stunning" MIAX decline vs prediction markets like Kalshi; the perpetual-futures approval and its implications for the other exchanges; why the mineral royalty positions were liquidated; the short- and long-term crypto outlook and whether the bitcoin thesis is still the 2015 thesis; gold; and updates on LandBridge, PrairieSky and WaterBridge.
• Explicitly deferred: TPL data-center project development detail and the initial new Spin-Off strategy positions — clients are told to contact their relationship manager. Two "alarm-bell" questions (why an index critic now issues ETFs; why an IPO abstainer now participates) open the discussion instead.

Indexation: not the enemy — but the ETF became the price-maker

• No antagonism to indexation as Bogle practiced it in the 1970s; if an index offers a usable attribute the firm will use it (see AMLP). What changed is scale: once indexed dollars exceeded actively managed dollars, each new ETF dollar had to buy more of the very same index shares, so ETFs "crossed over from passively participating in price discovery, to having a hand in making the prices." How can the S&P 500 be a benchmark when its own ETFs set it?
• Inclusion rules had to accommodate trillions, creating an artificial bifurcation of winners (constituents) and losers (too small, too illiquid) — the excluded get less flow, higher cost of capital, less research and suppressed valuations. And the largest constituents take a disproportionate share of new inflows, making the index steadily less representative of the economy.
• The GDP-according-to-the-S&P sidebar: 3% Energy, 5% Consumer Staples, 6% Consumer Discretionary — and 49% Information Technology.

The Bogle origin story — the real one

• Bogle did not set out to democratize investing: hired by Wellington, he wanted to be a superior active manager, publicly disdained buy-and-hold, led the firm into Nifty Fifty growth names at twice the market's P/E, became CEO in 1970 at 38 and was fired in 1974 in the bear market.
• Vanguard was the negotiated consolation — a mutualized fund administrator whose charter barred him from managing money. A fund with no investment management — passive replication of the S&P 500 — was the way around the non-compete, with a structural cost advantage no profit-seeking competitor could match.
• Murray Stahl debated Bogle at a Grant's Interest Rate Observer event in 2017 and agreed with ~95% of it; the shared ground is that compounded earnings growth requires an unbroken long-horizon chain — "staying the course." The difference is the vehicle: a whole-market index for Main Street, versus identifying the rare business models that can sustain high-order compounding. Peter Doyle has recently described the same creation story alongside other institutionalized modern-portfolio-theory conventions (e.g. standard deviation as the measure of risk).

Why an active manager now issues ETFs

• Practical access first: some clients prefer or require an ETF wrapper, so it is a tool for reaching a strategy — "we wouldn't bother" without differentiated exposure.
• The Inflation Beneficiaries ETF is the exhibit: its S&P-overlapping holdings are a 0.57% index weight, so most investors have no countervailing exposure to commodity inflation, one of the few systemic risks that can't be diversified away. At minimum a completion fund; the firm thinks it is a better mousetrap than "mining and chemicals companies" that pass for inflation hedges.
• Second function: reaching otherwise inaccessible markets. In Japan, ETF exposure is limited to globally-revenued mega-caps, so the firm runs two local strategies beneath that umbrella — the Japan Owner Operator ETF and the Japan Special Opportunity strategy, which buys listed captive subsidiaries the government wants made independent or taken private, at Graham-and-Dodd discounts, and is being validated by takeouts at substantial premiums. (That one has no ETF yet — "in this case, we wish we had an ETF.") A Blockchain Development ETF was established off the exchange/tokenization thesis.

Gratitude for indexation — the ETF Divide and the Equity Yield Curve

• "Thank goodness for the market-structure-distorting impact of ETFs." Company analysts have been decimated over 20 years; the rejection of smaller-caps, odd business models and companies that shrink liquidity via buybacks means those names are not efficiently priced. Two named sources of the free lunch: the Equity Yield Curve (value realization too far away or too indeterminate for a relative-return industry) and the ETF Divide (discounts for non-economic reasons). The counter-risk is named too: NFL risk — No Free Lunch risk.
• The unnamed example: one of the fastest-growing Asian airports, a wholly owned subsidiary of a several-billion-dollar parent in a different industry (half of whose shares are held by its parent), available at a discount to book and a single-digit cash earnings multiple when comparable listed airports trade at ~25x EBITDA and run ~30% operating margins under long-term government-granted jurisdictions. An IPO of the subsidiary "may well feature in the foreseeable future."

AutoNation and Penske — buybacks as the disqualifier

• AutoNation: 13.7% annual per-share earnings growth since 2000, at 11x net income, having repurchased 58% of its shares in five years. Penske: 14% 15-year EPS growth at 11x, with >50% insider ownership plus 20% held by Mitsui, so an $11 billion market cap is a ~$3 billion float; a 2.5%-of-total buyback is 7.5% of traded shares.
• Both are smaller than the two smallest S&P 500 constituents, which carry 0.1% and 0.0% weights. Shrinking your float is "entirely the wrong direction relative to the expanding trading liquidity needs of ETF organizers" — which is precisely the source of the discount.
• Post-writing note: Penske Corporation and Mitsui & Co. offered to buy the remaining 17% of Penske Automotive they don't own, now in independent special-committee review.

IPOs and us — the stance never changed

• The base rate is still bad: across 9,000-plus IPOs over 46 years to 2021 (Jay Ritter's data), 56% traded below their IPO price three years later, 60% for buyers at the end of day one, and worse at five years. Most IPOs "have been engineered to be sold to the public."
• But the decision is "definition-agnostic" — business and valuation analysis, not the label. The Asian-airport case would be the mirror image of the typical IPO: an IPO used as a value realization catalyst inside an existing holding.

LandBridge — the IPO that proves the point

• July 2024, anchored by Horizon Kinetics, sourced through the firm's long association with TPL and adjacent Permian property transactions; management wanted holders who understood the business and would stay long term.
• Over 300,000 Delaware Basin surface acres, currently earning royalty-like fees on water piped across or stored in subsurface pore space. The hydrogeology does the growing: an ancient inland sea means aging and deeper wells lift more brackish water — 4 barrels of water per barrel of oil today, an estimated 6-to-1 by 2030 (~9% annualized volume growth) — before any oil-output expansion by the majors and with no active growth effort.
• Ten-year water contracts carry inflation escalators, so 12%+ revenue growth requires no capital spending; new pore-space contracts should reprice above the current ~$0.11/barrel. The company introduced "powered land" — surface contiguity that can capture data centers, roads, power lines, wind, solar, carbon capture and water, each a recurring high-margin royalty-like stream — and its aquifer lets it supply source water at ~$1/barrel.

PrairieSky — acres per share, and the ownership rule

• Twelve years post-IPO: acreage tripled from 5 million to 18 million acres while acres per share doubled — roughly 6% a year of per-share land accretion on top of revenue and earnings growth.
• The 750-mile land position across three Canadian provinces is large and largely unexploited, so cash flow need not be reinvested into buying more royalty contracts — it funds dividends, buybacks and land acquisition.
• Governance detail the firm highlights: executives must, within three years of appointment, buy enough stock with cash to be worth 2x–5x salary, separate from stock-based compensation — "a strategy toward ensuring itself of a very long stretch of financial compounding possibilities."

Why land at all — the perpetuity argument

• Land is a perpetuity, the longest-lived asset, and can be repurposed to higher uses — "can't do that with a tractor or a data-center server." Supply in acres per person always shrinks as population grows, and even remote Texas and Canadian acreage is exposed to oil, gas, and now compute — water to cool the power plants feeding a multi-thousand-acre 10-gigawatt campus that could power 2x Chicago's electricity demand.
• It is a rare public-market investor who buys land, though it is common among the extremely wealthy (Ted Turner, who died in May, owned 2 million acres). Land is not even listed as a sector in the equity indexes — which was the point of the original Texas Pacific Land recommendation over 30 years ago: buy the land, not the oil, and let buybacks compound the per-share acres.

The Repetitive Language Model of compounding — TPL's drawdowns

• A client asked "what is the matter with TPL?" because the shares were 30% off their high — and, as a fund manager himself, added "yeah, but not for this long." The record: in the past eight years TPL has fallen 40% to 53% on four separate occasions, and is nonetheless up 4-fold from its high of eight years ago. Fifteen years ago it fell almost 75% in about a year.
• The smooth early portion of a long-term chart is "a mirage, a misimpression wrought by time and compounding" — a fractal: magnify any earlier window and the volatility is identical. The second chart in the "is this TPL or is this TPL?" exercise turns out to be Intercontinental Exchange, which fell 66% over a year and change. Different industries, different cap classes; what they share is a high rate of compounding over a long frame.
• Why we can't see it: brains evolved for the veldt are good at spotting a charging lion, bad at geometric functions and at time frames below the sensory threshold.

The cost of breaking the chain

• The sell-at-the-high table: sell a many-times-cost holding, pay the tax, reinvest 85¢ on the dollar into a company that genuinely grows 25% faster — and you don't break even until year seven. "The more interesting question is whether you will have held the second stock long enough to find out."
• Peter Doyle: "The hardest temptation to resist is to trade something good for something average." Murray Stahl, about a decade ago: "Although it might seem hard to believe, patient inactivity is hard, demanding, and exhausting work."

A math model of compounding — water glass, snowball, Amazon, bitcoin

• Waiter!!! — nanobots double every 8 minutes and fill a glass in 64 minutes. At the halfway point in TIME the glass looks 6% full and Table 2 storms out; Table 5 waits to time step 7 of 8 and leaves with the glass 50% full — the remaining half arrives in the final 12% of the time. An exponential 2^t function: watch time, not the water level.
• Snowball Fight — a cartoon snowball on a 20° slope reaches ~3 feet and 200+ pounds in 6.8 seconds. At half the time it has 15% of final mass; at 78% of the time, 50%; a quarter of the final weight accrues in the last 10% of the time. A cubic (t^3) volume function, not exponential.
• Amazon (1,512 weekly closes since May 1997) and bitcoin (~6,300 daily prices since 2009) reproduce the same shape from real prices: at 50% of elapsed time bitcoin was at 6% of the recent level, Amazon 5%; at the 75% time mark bitcoin was 47%. "A remarkable consonance" between a mass-market retailer-slash-hyperscaler and a still-debated digital asset.

Why they reduced or sold the precious-metals royalty companies

• Start with why they bought: after 2008, zero rates and deficits drove gold from ~$1,900 (2011) down to under ~$1,100 by 2015 — below the cash cost of production for a meaningful share of the sector. That capital starvation was "the absolutely best period" for royalty companies, which locked in high double-digit financing rates on world-class mines for 20-plus years of output. It was "a business valuation selection, not a gold price wager."
• The gold call proved right anyway — a $5,400/oz high earlier this year, now ~$4,100 — but those prices are embedded as premiums in royalty valuations exactly as the depressed prices were embedded as discounts a decade ago. Gold now sits comfortably above all-in sustaining cost, miners have capital access, and the royalty companies can no longer sign contracts with the same earnings power or optionality.
• Structural gold backdrop still supportive: new mine supply has grown only 2.6% in total since 2018 (0.36%/yr), and a 24% increase in recycling has been more than offset by rising investment demand, primarily central banks. But "we focus on business analysis and fundamental valuation, not gold price forecasts." The model's minor drawbacks are named: no discretion over mine operations, and no pricing power — royalty companies are price takers.

Valuations have meaning — the 8.6% arithmetic

• Wheaton and Franco-Nevada trade at roughly 2x consensus NAV. Those NAVs start from current gold prices declining modestly, value non-producing royalties at zero (missing exploration optionality and new deals), and discount decades of cash flow at just 3%–5% — the inverse of which is a 20–33x P/E.
• The contrast: $100,000 a year for 20 years discounted at 15% is ~$625,000 — buying $2 million of future payments for about 30¢ on the dollar. Discounted at 3%, the same $2 million costs almost $1.5 million. "Not a bargain by any means" — and note the royalty companies themselves negotiate far higher discount rates in their own deals.
• The kill shot is multiple compression, not business quality: hold earnings growth at 15% a year for 15 years, but let 2x NAV become 1.5x NAV in five years, and the annualized return is 8.6%. "A valuation contraction within that time frame overwhelms the gradual accumulation of compounded growth."

The exchanges — what actually happened on May 15

• From May 15 to June 22 the major U.S. regulated exchanges fell almost as one, 22% to 38%, then recovered ~20% in the following five weeks. MIAX and CBOE are substantially higher than 12 months ago and ahead of the S&P; two are substantially lower.
• The businesses did not wobble: ICE's Q1 revenue, operating margin and per-share net income all rose; FY2025 revenue +6% produced double-digit EPS growth on a 33% after-tax margin, and the year's 14% EPS increase matches the 20-year 15% annualized rate. MIAX reported 40% higher revenue and 2x-plus free cash flow in the March 2026 quarter without full scale economies.
• The trigger was a Bloomberg story that ICE and CME had petitioned the CFTC to force Hyperliquid — a three-year-old decentralized exchange whose blockchain oil tokens went from marginal to billions of notional dollars a day when Brent rose from $71.32 to $126.69 — to register. Then the CFTC's May 29 approval of Kalshi's U.S. bitcoin perpetual, and CME's mid-June suit arguing a perpetual is a swap, not a future (no expiry, no delivery), decided by a commission currently staffed by its chair alone.

Perpetuals and prediction markets — threat or expansion?

• A futures contract exists to hedge a real exposure: the worked example is a South Korean distributor holding 500,000 barrels — $40 million of oil — for a three-week transit; a departure on May 31 would have meant a 20% lower selling price on arrival. That needs a fixed delivery date and an exchange enforcing collateral. ICE has had no futures contract failure in nearly three decades across 1.2 billion contracts traded this year.
• A perpetual has no expiry and no delivery, so it cannot hedge a commercial commitment — "useless for a commercial customer and, more broadly, for the economy." It is a phone-click wager, north of 50x leverage offshore; Kalshi's own example is 6x on bitcoin: $200 controls $1,000, +10% in bitcoin is +50%, −20% wipes you out. Market gamification, sold to retail — while 85–90% of ICE's business is institutional.
• Nothing stops incumbents from offering event contracts, and they are: ICE owns 17% of Polymarket (for the data, not the sports); CME runs 24/7 crypto futures, 24/7 mini Gold with 24/7 Oil coming, single-stock futures, and FanDuel/FutureSports partnerships; CBOE has proposed near-24/5 equities; the SEC approved Nasdaq token-form trading and settlement via DTCC; NYSE is exploring blockchain settlement. Hence: Kalshi and Polymarket are "an indicator of extraordinary next-generation, next-decade expansion possibilities for exchanges" — and hence the Blockchain Development ETF.

What exchanges do — crisis as a business opportunity

• The longevity table runs from the NYSE (formed 1792) and Hong Kong (1891) through CME (1898) to MIAX (formed 2000, listed 2025). The NYSE predates the first U.S. gasoline automobile; these businesses adapted to and benefited from every technological, financial and political crisis since.
• The current example: Polymarket lets retail speculate on Strait-of-Hormuz events, facilitating nothing real, while Syria, Iraq, Saudi Arabia and the U.A.E. build alternate export routes — and each route needs new hedging contracts, because Brent futures is not one contract, it's hundreds, differing by journey time and spread relationships. When institutional customers find an unhedgeable exposure, CME and ICE compete to launch the contract.
• Result: ICE and CME valuations "haven't been this low in the 17 years since the Great Financial Crisis," against an S&P 500 at 26x 2026 estimates; CBOE hasn't been this inexpensive in three to five years.

Bitcoin, supply side — the cost of production floor

• "It's right on schedule" (Murray Stahl). The current decline is the fourth in a series that repeats every four years, tied to the halving that doubles production cost — a 19% annualized increase in cost of production, all else equal — and the iron law of supply and demand puts price at least at cost so long as anyone wants to own or use it.
• The firm built its own estimate: electricity (via data-center hosting) is ~60% of cost; a modeled network electricity-consumption schedule divided by known network hash power gives average fleet efficiency, and coins issued per day give cost per coin. At $0.05/kWh the all-in estimate today is ~$65,000, doubling at the April 15, 2028 halving to ~$130,000; at the typical ~75% post-halving premium over cost, ~$225,000. Using the newest rigs and a 25% miner ROIC, today's all-in is $117,000–$149,000.
• Supply itself is essentially fixed: ~96% of total supply is released, so the next 114 years average <0.04%/yr of supply inflation — against U.S. M2 up 5.5% in 12 months on ~2.7% real GDP growth.

Bitcoin, demand side — Metcalfe's Law and the t^6 power law

• Metcalfe's Law (network value grows with the square of users) has described Facebook/Meta, Tencent and internet usage; for bitcoin each new user adds payment, savings, liquidity, developer and merchant utility. Hashrate — up tenfold in five years — tracks the price remarkably closely, prompting the open question of which follows which.
• Combining the t^2 network-value function with the t^3 network-size function (the snowball's cubic volume) gives a power function a touch under t^6. Backtest: $1 invested in 2011 was projected to be $9,986 by December 16, 2019 vs an actual 9,169x; by April 5, 2025 the model said 99,878x vs an actual 110,726x.
• The important part is the slowdown: each 10x takes longer — 1.1 years, then 1.6, then 2.4, and nearly 8 years for the next one from April 2025. The 90% annualized returns of the 2016 era become more like 33%. "It's why we here at Horizon Kinetics Central are not perturbed by the periodic price declines in bitcoin." The stated conclusion for April 2028: $150,000–$250,000 on production economics, $270,438.05 on the power law.

Not rowed (passing mentions and illustrations)

• Named only as index-weight or sector illustrations: Costco, Walmart, General Mills, Procter & Gamble; "the Chevrons and ExxonMobils" as generic Permian operators. Named only as historical/academic references: Vanguard, Wellington Management, Grant's Interest Rate Observer, Fidelity (chart source), Tencent (Metcalfe's Law), DTCC, FanDuel, FutureSports, Mitsui & Co. (as the Penske bidder), SeaRates.
• The exchange-longevity table names ten non-U.S. exchanges (Japan Exchange Group, HKEX, LSEG, Deutsche Boerse, Philippine, Singapore, ASX, Bursa Malaysia, Euronext Athens, Johannesburg) purely as evidence of business persistence — no view is argued on any of them.
• The Asian airport subsidiary, its parent and its parent's parent are deliberately unnamed; no ticker has been inferred. The Japan Owner Operator ETF and Blockchain Development ETF are discussed but not given tickers in the text, so they are covered in the talking points rather than rowed.