Structural, or Just Winter?

Two popular reasons to be bearish on Bitcoin in 2026 — and whether they hold up

TON618 Capital Research · As of July 19, 2026 · Download PDF

The bottom line, up front

Two arguments are making the rounds as reasons Bitcoin's problems are permanent — not just the usual bad patch of a crypto "winter." We took both seriously and tested them against the data. Here is where each one lands:

Argument 1 — "The gamblers have left crypto for good." The idea: the speculative money that used to buy Bitcoin now goes to prediction markets (Kalshi, Polymarket) and fast-money stock-market bets instead. Our verdict: mostly rejected. It's true this month — trading money is visibly elsewhere right now. But "for good" is the part that fails. Speculative money has never had a permanent home; it chases whatever is going up, and it has "left Bitcoin forever" before every single recovery in Bitcoin's history. On top of that, the prediction-market boom is smaller than it looks (most of it is sports betting), and — the irony — much of it actually runs on crypto. This is a passing rotation, not a permanent change.

Argument 2 — "Bitcoin's power is worth far more running AI, so the energy is leaving mining for good." The idea: a megawatt of electricity earns 10–20× more serving AI than mining Bitcoin, so miners are converting, and the foundation under Bitcoin is being pulled away. Our verdict: true, but about the wrong thing. The "10–20×" is real today, and mining companies are converting to AI — that part is confirmed, and it's a genuine, permanent shift for the mining stocks. But it says almost nothing about the price of Bitcoin itself. Bitcoin is designed to automatically make mining easier and more profitable whenever miners leave, so the network heals itself. The argument confuses "bad news for a dozen mining companies" with "bad news for Bitcoin." Those are not the same thing.

So is the bearishness permanent? No — not on this evidence. But rejecting the two headline arguments doesn't mean nothing changed. Three real, lasting shifts are underway (Section 5), just not the ones the pessimists named. Winter is real; the case that it's permanent is not.

The rest of this note shows the work.

1. First, the bear case at full strength

We're not going to knock down a weak version of these arguments. Here is the strongest evidence for each.

The money really is elsewhere right now. Kalshi, the regulated US prediction market, traded roughly $29–33 billion of contracts in June 2026 alone (two data providers put it at $32.7bn and $29.2bn; there's no official figure). Eighteen months earlier that number was about $226 million — a jump of more than a hundred-fold. Polymarket, the crypto-based prediction market, did $4.3 billion in June, up more than seven-fold in a year — during a year when Bitcoin was falling. Meanwhile the US stock-options market just had its sixth record year running, and more than half of all trading in S&P 500 options is now in "zero-day" contracts — bets that expire the same day, the purest form of short-term speculation.

And Bitcoin's own activity is shrinking. Reported Bitcoin trading volume roughly halved between October 2025 and June 2026. June 2026 was the worst month on record for withdrawals from US Bitcoin ETFs — investors pulled a net $4.5 billion. And miners are earning far less: mining revenue per unit of computing power is down about 46% from a year ago. On today's snapshot, the speculative dollar clearly is somewhere other than Bitcoin. That much is simply fact.

Miners really are converting to AI. Since mid-2024, Bitcoin miners have signed roughly $96 billion of disclosed contracts (over $100bn including some undisclosed deals) to host AI computing instead of — or alongside — mining, covering about 3.9 gigawatts of power (enough for a few million homes). Core Scientific says it intends to convert "every megawatt in our portfolio" to AI hosting. Galaxy Digital quit mining entirely. Cipher literally renamed itself and ripped out working mining machines to make room for Amazon. If the companies closest to the electricity are voting with their megawatts, the bear says, the market has spoken.

Both arguments start from true facts. The question is what those facts mean.

2. Argument 1: the money chases the price — it always has

For "the gamblers left for good" to be true, the speculative money would have to have permanently moved house. The evidence says it doesn't have a house at all — it follows the price wherever the price goes.

We can actually measure this. We built a model of the money flowing into and out of US Bitcoin ETFs versus Bitcoin's price, using every trading day since these ETFs launched in January 2024 (617 days). The result is clear: the price moves first, and the money follows. When we line up price changes and money flows, the strongest relationship by far is yesterday's price move predicting today's money flow — not the other way around. In plain terms, these investors don't lead the price; they chase it. Money like that leaves when prices fall and comes back when prices rise. That is the definition of a cycle, not a permanent exit — and we measured it rather than assumed it. (One extra finding: the price impact of that money has actually gotten stronger in 2026, so when the flows do return, they'll push the price up harder per dollar, not softer.)

"Bitcoin is finished" has a perfect track record — of being wrong. Every past crypto winter produced confident obituaries from serious people:

• Warren Buffett: Bitcoin is "a mirage, basically" (March 2014)
• USA Today: "Bitcoin is headed to the ash heap" (January 2015, with Bitcoin near $208)
• Economist Nouriel Roubini, to the US Senate: "the mother or father of all scams and bubbles" (October 2018)
• The European Central Bank: an "artificially induced last gasp before the road to irrelevance" (November 2022)
• CNBC's Jim Cramer: "I wouldn't touch crypto in a million years" (December 2022)

Each of these came within one to four years of a new all-time high. The 2022 obituaries were the fastest to be proven wrong — Bitcoin reclaimed its record roughly fifteen months later. The closest match to today's specific "money moved to another casino" claim was 2023, when same-day stock options boomed while crypto sat at its lows — and that rotation then happily coexisted with the 2024–25 Bitcoin rally. These venues are competing for attention; they aren't draining a fixed pool of money from one to another.

How big is the threat, really? Honestly, in both directions:

• Prediction markets are still small next to crypto: Kalshi and Polymarket combined traded about $33–37 billion in June, versus something like $1 trillion of reported Bitcoin trading — roughly 3–4%. Not yet a serious rival for total crypto turnover.
• But they've already caught up to the flagship Bitcoin ETF: that same monthly figure now matches or beats the entire monthly dollar volume of IBIT, the largest US spot Bitcoin fund. Among regulated, everyday-investor venues, prediction markets reached parity with the biggest Bitcoin product in under two years. That's the part worth respecting.
• And the composition matters: roughly 85–87% of Kalshi's volume is sports betting (in 2025, 89% of its fee revenue came from sports). So most of the "prediction-market boom" is really a sports-betting boom in a financial-market wrapper — and sports-betting dollars were never going to buy Bitcoin anyway. The part that directly competes with Bitcoin for a speculator's attention — non-sports events — is real and growing, but it's a slice of the headline number, not the whole thing.

That's the scoreboard on Argument 1: the rotation is real, the scale is smaller than the headlines, and nothing about it looks permanent. It's worth watching — but it's weather, not climate.

3. The competitor is running on crypto's own rails

The "money left crypto" story has one final twist: the biggest prediction market is crypto. Polymarket runs on a blockchain and settles every trade in a US-dollar stablecoin — every dollar wagered there is a crypto-based dollar. Kalshi, though dollar-based, lets users fund accounts with Bitcoin and stablecoins and gives crypto deposits a higher limit than bank transfers. And the money behind these venues is blending the two worlds: the parent company of the New York Stock Exchange put up to $2 billion into Polymarket; Robinhood routes its prediction-market product through Kalshi.

So to the extent gamblers are drifting toward prediction markets, many of them are being quietly funneled into stablecoins and crypto accounts — not away from crypto. (The honest counter-example: Polymarket's new US-regulated version deliberately settles in regular dollars through traditional brokers, stepping off the crypto rails. Which model wins is still undecided.)

This sets up a bigger, slower question — and here we're explicit that it's a hypothesis, not a finding. There are two different layers to keep separate:

• The spending layer — everyday and machine-to-machine payments — has clearly been won by stablecoins, not Bitcoin. When people say "AI agents will pay each other in crypto," they're right that it's crypto, but it's stablecoins (digital dollars), not Bitcoin. Anyone claiming Bitcoin wins because AI runs on crypto is pointing at the wrong layer.
• The savings layer is the open question. A stablecoin is still a US dollar — it carries all the dollar's convenience and all the dollar's inflation. The hypothesis worth watching runs like this: crypto rails get people into digital dollars first; then, if inflation and low interest rates quietly eat away at those dollar balances, some of that money graduates to a harder store of value — Bitcoin — sitting one tap away on the same app. We've seen this sequence in countries with weak currencies: people flee to digital dollars first, and to harder assets second. The global version would need that same pressure to reach the dollar itself.

We hold this loosely, and we state the counterarguments with equal weight: the dollar is still the strongest of the major currencies; most stablecoin demand today is a flight toward dollars, not away from them; new products now pay interest on dollar balances, which makes a zero-yield asset like Bitcoin a harder sell; and measured movement from stablecoins into Bitcoin is, so far, thin. What the crypto-rails buildout changes isn't the odds of that scenario — it's the friction if it ever arrives. The largest pool of digital dollars ever assembled would be sitting one click from the exit.

4. Argument 2: the energy point is true — but it's about mining stocks, not Bitcoin

Now for the energy argument — the more interesting of the two. It takes three steps to unpack: whether the energy is actually leaving mining (4a), whether the 10–20× price gap will last (4b), and what the miners' own conversions really tell us (4c).

4a. Bitcoin heals itself when miners leave

The argument runs: electricity is worth more in AI → so it leaves mining → so Bitcoin is in trouble. The problem is that the middle step mostly hasn't happened, and even if it did, the last step doesn't follow.

It mostly hasn't happened. The total computing power securing Bitcoin is near its all-time high (recent readings around 1,043 EH/s, only modestly below the October-2025 peak) — even though the reward for mining has roughly halved. If energy were really fleeing mining, that number would be falling. It isn't. What actually squeezed miners' pay isn't an exodus — it's competition among miners themselves. (And the slight dip from last year's peak has happened while Bitcoin's price was falling — that's ordinary miners giving up in a downturn, not miners defecting to AI.)

And even if miners did leave, Bitcoin is built to shrug it off. Roughly every two weeks, Bitcoin automatically re-tunes how hard it is to mine, based on how many miners are competing. If a big chunk of miners left for AI, mining would automatically get easier and more profitable for everyone who stayed — until it was no longer worth leaving. The system self-corrects by design. So miners leaving hurts the miners' profits; it does not starve Bitcoin. Bitcoin's price is set by how much people want to own Bitcoin — not by the electricity bill of the people who happen to produce it. Treating "mining companies are struggling" as "Bitcoin is struggling" is the core mistake in the argument.

4b. The "10–20×" is today's price, not tomorrow's

The 10–20× gap is real right now. A modern mining fleet earns roughly $67–99 of revenue per megawatt-hour of power; AI hosting currently earns a large multiple of that. But "right now" is doing a lot of work in that sentence.

The AI premium is a shortage price, and shortages fade. Renting a top AI chip (the Nvidia H100) has already gotten 64–75% cheaper than at its peak, because supply is catching up on the same roughly two-year cycle that makes each generation of chip obsolete. What stays scarce isn't the AI computing — it's the electricity hookup itself. Getting new large-scale power connected to the US grid now takes about five years, and the big gas turbines that generate it are sold out through 2030. And here's the punchline the bears miss: the miners own that scarce thing. So "miners converting to AI" isn't miners fleeing a sinking ship — it's landlords who happen to own the one scarce resource (connected power) renting it to the highest bidder of the moment. The 10–20× is the price of standing at the front of a five-year queue. Queues eventually clear; the gap will narrow.

4c. The miners' flexibility is real — but it's in the pipeline, not the contracts

We built the entire sector's contract book from public filings — ten companies, 19 AI hosting deals, about 3.9 gigawatts under contract, ~$96 billion of value (the full dataset is published with this note). It tells a two-sided story.

The part of the bull case that does not work as usually told: these AI contracts are long and locked. Almost nothing comes up for renewal before 2031, and more than half of the contracted power is tied up until 2041 or later. Worse, each new deal has been longer than the last (12 years → 15 → 20), and the renewal options at the end mostly belong to the customers, not the miners. So these signed contracts are not a source of year-to-year flexibility — they're the opposite: the miners deliberately gave up flexibility in exchange for guaranteed income and the loans that income secures. And once a site is converted to AI, it doesn't come back to mining.

The part that does work — and is actually stronger: the flexibility lives in the ~20 gigawatts of power the miners have secured but not yet committed to anyone — roughly five times what's currently under contract. Every year since 2024, the industry has converted a fresh batch of that pipeline into new deals (TeraWulf now states a target of 250–500 MW of new contracts per year as company policy). That's the real option: each year's new power gets pointed at whichever use pays best that year. And at the margin, the mining-vs-AI choice really is made opportunistically — Cipher, for instance, plans to keep mining one site "through expiration of the [power contract] in July 2027, while evaluating a potential conversion" to AI. So the accurate version of the claim isn't "staggered contracts let miners switch back and forth" — it's "a huge uncontracted pipeline plus a yearly signing rhythm lets the industry aim each year's new power at the best-paying use." Stated that way, it holds up.

What still argues against it, plainly: the customer list is dangerously short. One company (CoreWeave) is the entire AI business of three of these miners; a handful of AI buyers, not "the market," decide whether all that pipeline ever converts. Three of the ten miners hadn't signed a single AI tenant as of June (Riot spent about 18 months with 600 MW earmarked and no taker) — though CleanSpark's $6.6 billion deal on July 14 shows the demand can still show up. If AI spending stalls, that 20-gigawatt pipeline is an option on nothing.

5. What did actually change

Grading one argument "mostly rejected" and the other "true, but about the wrong thing" does not mean nothing is different this cycle. Three things really did change — they're just not the three the pessimists named:

1. Bitcoin's typical buyer grew up. With ETFs now the main on-ramp, the marginal dollar buying Bitcoin belongs to a financial adviser or institution, not a retail gambler. As we showed, that money chases the price — which cuts both ways. It guarantees the fast money comes back when prices rise, but it may also permanently soften Bitcoin's wild swings in both directions: shallower crashes, and shallower manias. That's a real, lasting change — just not a bearish one. Its direction is genuinely ambiguous.

1. Bitcoin now has a bigger rival for attention: AI. For the first time in Bitcoin's institutional era, there's a competing story in the market that is larger than Bitcoin itself — the AI investment boom. When investors are worried about money losing value, Bitcoin gets the bid; when the AI trade is roaring, attention and capital flow there instead. That tug-of-war is real and slow-moving, and it's a genuine dependency — but it's a cycle measured in years, not a permanent verdict.

1. The mining industry itself is permanently changing — and here the bears are simply right. Mining companies are turning themselves into AI-era power landlords, the conversions don't reverse, and "pure Bitcoin miner" is a shrinking category of stock. But note what that sentence is about: it's a fact about ten companies' business models, not about the asset their machines secure. The bears found a real, permanent shift — and filed it under the wrong name.

6. How we'd know we're wrong

We'd rather be checkable than clever. Here are the specific things that would prove us wrong, and we intend to revisit them:

• If Bitcoin ETF money stops chasing the price — if it keeps flowing out even as prices rise — our "it's just a cycle" read in Section 2 breaks.
• If the computing power securing Bitcoin actually falls while Bitcoin's price holds firm — that would mean energy is genuinely leaving mining faster than the network can heal, the first real crack in Section 4. (The mild 2026 dip doesn't count: it happened while the price was falling — ordinary capitulation, not defection.)
• If the miners' 20-gigawatt uncontracted pipeline gets used up without a fresh pipeline replacing it, their year-to-year flexibility disappears and we'll say so.
• If prediction-market activity (the non-sports part) keeps growing even while Bitcoin is rising — that would suggest a real, lasting substitution rather than a passing rotation.
• If AI chip-rental prices stop falling and durably reverse — that would revive the bear's "10–20× forever" claim.
• The inflation tell for Section 3: persistently negative real interest rates plus a still-growing pile of stablecoins would be the setup; actual measured movement from stablecoins into Bitcoin would be the confirmation. Nobody publishes that cleanly yet — it's a data series worth building.

Where this lands

Strip it down, and each argument fails in its own way. The liquidity argument mistakes a habit for a homecoming: speculative money chases past winners — we built the tool that measures it doing so — and the venue it supposedly defected to turns out to be largely a sports book that happens to run on crypto. The energy argument is sharper. It found a real, permanent shift — and filed it under the wrong name: the megawatts and the business models are migrating, but Bitcoin's own machinery heals behind them, and a hard decade for the miners is not a verdict on the asset their machines secure.

What's actually left standing is real, but it isn't the pessimists' story: a more grown-up, less wildly volatile Bitcoin; a formidable new rival for investor attention in AI; a mining industry remaking itself, one-way, into power landlords holding a giant option on the AI boom; and a payments world that has gone digital on stablecoins — leaving the real open question as whether savings eventually follow into Bitcoin if the dollar's own discipline slips.

Winter is real. The case that it's permanent is not — yet. The checklist above is where "yet" gets settled.

Related TON618 research. This note draws its energy-side arguments from three companion pieces and should be read alongside them: The Power Wall — why the AI premium settles in firm, grid-connected power rather than computer chips (the source of the "shortage vs. durable scarcity" test and the 64–75% AI-chip-rental decline used in Section 4b); the AI-Infrastructure / Neocloud series — the miner-to-AI conversion economics and financing behind Section 4c; and our in-house bubble-tracker, whose "AI boom vs. inflation-hedge" framing informs Section 5.

Companion data published with this note: miner_contract_ladder.json (22 filing-anchored contract entries across ten companies) and its simplified ladder_chart.json, prediction_market_volumes.json, hashprice.json and hashrate_monthly.json, and the Bitcoin-ETF flow model output (flow_leadlag.json). Every chart renders directly from these files. Methodology and per-figure caveats are held in the evidence files.

Disclosures

Information only. TON618 Capital. This report is for information purposes only. Nothing here is an offer to sell or a solicitation of an offer to buy any security, fund interest, or digital asset, and nothing here is personalized investment advice or a recommendation regarding any instrument. This note is explicitly a lens, not a call: it carries no price target and is deliberately kept off our signal ledger. Figures are as-of dated (market and on-chain data as of July 19, 2026 unless noted) and drawn from company filings, exchange, grid and first-party keyless data, or dated public reporting; facts and estimates are separated throughout, and where sources conflict both are shown.

Publisher's exclusion. All research is published solely as general, impersonal information of regular circulation. It is not tailored to the objectives or circumstances of any individual and is not issued in connection with compensation from any client. The Fund has no clients and distributes all research free of charge. On that basis it publishes in reliance on the publisher's exclusion from the definition of "investment adviser" under the Investment Advisers Act of 1940 (§202(a)(11)(D); cf. Lowe v. SEC, 472 U.S. 181 (1985)).

Registration & conflicts. TON618 Capital is not registered as an investment adviser or broker-dealer in any capacity. The Fund is a Bitcoin fund: it maintains a strategic Bitcoin allocation, and because this note bears on the durability of Bitcoin demand it should be read as touching on an asset in which the Fund has a material interest. As of the date of this note the Fund holds no position in Core Scientific, TeraWulf, IREN, Hut 8, Cipher Digital, Galaxy Digital, MARA, CleanSpark, Applied Digital, Riot, CoreWeave, or any other single equity named herein, and receives no compensation from any party in connection with this research.

Use of AI, and a specific counterparty conflict. Artificial intelligence is used in the creation of this research. A material and note-specific conflict is disclosed here: the AI tooling used in TON618's research process is made by Anthropic, and Anthropic is itself the counterparty to one of the contracts this note analyzes — TeraWulf's ~$19bn, 20-year Hawesville (Justified Data) lease cited in Section 4, the single largest new deal in the sector dataset. Readers should weigh that relationship when assessing the note's treatment of the miner-to-AI conversion thesis. All methodology and data integrity are reviewed and approved before publication by TON618 Capital's Chief Investment Officer, Keyth Beck; errors may nonetheless occur, and readers should verify independently.

CFA. This report was prepared to align with CFA Institute analytical standards (methodology only). CFA® and Chartered Financial Analyst® are registered trademarks owned by CFA Institute. That reference describes the analytical framework applied; it does not imply the report was prepared, reviewed, or authored by a CFA charterholder, and the report is not issued, reviewed, endorsed, certified, or approved by — nor affiliated with — CFA Institute.

Risk & feedback. Past performance and forward estimates are not indicative of future results. Digital assets and equities are volatile and may result in total loss of capital. Corrections and feedback are welcome — please direct them to CIO Keyth Beck at keyth@ton618capital.com. See full disclosures at the foot of the research library.