Chained to the Chips

Bitcoin hit $55,106 on Saturday, its lowest since September 2024, a day after Nvidia lost 14%. Five ways AI can move Bitcoin, scored against this month's data, and the one that did the work.

A burning Bitcoin coin and a black AI chip, bound by a glowing cable, plunge through a night sky trailing shattered candlestick shards above a city of data centers.

Bitcoin was sold for a decade as the asset that does not care what Silicon Valley does. This month it moved with Nvidia.

On Friday the Philadelphia chip index lost 10.7%, and Bitcoin broke below $60,000 for the first time since July. Then the stock exchanges closed, and it kept falling. In a podcast episode released two days earlier, BlackRock's head of digital assets, Robbie Mitchnick, reportedly said Bitcoin had traded almost inversely to the AI trade since June. Over the 30 trading days to Friday, the correlation between Bitcoin's daily returns and the Nasdaq's was 0.70 (my calculation).

This piece asks a narrower question than "why did Bitcoin fall?" Can AI move Bitcoin, and through what? I count five channels. This month's evidence supports one strongly, two partly, and two not this month. Nothing here is investment advice.

The dip in one table

ItemValue
All-time high$126,296 on Coinbase, October 6, 2025
November low$55,106 on Coinbase, November 21 around 06:00 UTC ($55,130 on Binance; Reuters: $55,251.60)
Last time this lowSeptember 2024
Price at writing$60,151 (CoinGecko, 23:41 UTC November 21)
November so farAbout −22% from the October close of $77,424 (−29% at the low)
From the peakAbout −52% (−56% at the low, 411 days after the peak)
US spot Bitcoin ETFsAbout −$3.47 billion net in November through the 19th; BlackRock's IBIT about −$2.94 billion of it
Futures liquidatedAbout $2.4 billion across crypto in the 24 hours to 04:16 UTC November 21
Fear & Greed index10 to 13, "Extreme Fear", since November 14

Sources: Coinbase and Binance exchange data, CoinGecko, Reuters, SoSoValue data via The Block, CoinGlass data via CoinDesk and alternative.me, all read November 21. The percentages are my arithmetic. Venues disagree on the low by a few hundred dollars, and ETF totals depend on the cutoff: Friday's flows were not final at writing.

How it fell

This is not the cycle's first crash. Bitcoin already fell to $60,001 on February 6 and to $57,718 on July 1 (Coinbase). November's low sits only about 4.5% under July's. What November added was speed.

  • October 29. Binance said a security incident hit about $6.3 billion of client assets, which CNBC called "the largest crypto hack on record". Bitcoin closed 4.3% lower.
  • November 5. Two days after Democrats won both houses of Congress, AI stocks broke on fears of a slower data-center buildout. The Nasdaq fell 5.5% and the chip index about 12%. Bitcoin dropped to just under $76,000, and more than $1 billion of crypto positions were liquidated (CoinDesk). "The midterms appear to have triggered this sell-off," GSR's Carlos Guzman told CoinDesk.
  • November 11. MSCI confirmed that Strategy, the Bitcoin-treasury company, leaves the MSCI USA Index after the close on November 30.
  • November 12. Jensen Huang resigned as Nvidia's CEO "effective immediately", per Nvidia's 8-K. The filing gives no reason.
  • November 12–14. A long-liquidation cascade: nearly $700 million of Bitcoin longs on Thursday and $629 million on Friday (CoinGlass via CNBC). Bitcoin lost 10% over November 13–14.
  • November 19. Anthropic began trading on Nasdaq. The Nasdaq fell 2.2%. US spot Bitcoin ETFs lost $582.2 million that day, $541.1 million of it from IBIT (The Block).
  • November 20. The Nasdaq fell 5.28%, Nvidia 14.1% and the chip index 10.73% (Seoul Economic Daily, matching the Yahoo Finance closes). Bitcoin went under $60,000.
  • November 21. The low around 06:00 UTC, then a 9% bounce. Reuters had it back across $60,000 at about 13:30 UTC.

Reuters summed up the backdrop as "weeks of concern about the profitability of massive AI infrastructure spending, rising borrowing costs, regulatory risks following Democrats' win in the US midterm elections and the resignation of Nvidia CEO Jensen Huang." Add a Fed that held in October with two votes for a hike; futures price a quarter-point December 9 hike at about 95%, with the rest on a half-point move (CME FedWatch as reported by 24/7 Wall St; other readings range from about 89% to 100%). The 10-year Treasury yield peaked at 5.15% on November 13.

Gold fell too, about 10% in November on the December futures contract, which looks more like a liquidity squeeze than a crypto-only selloff.

Five ways AI can move Bitcoin

1. The same holders, the same sell button

This is the channel with the evidence. Computed for this article from Yahoo Finance daily closes, Bitcoin's daily returns correlated with the Nasdaq Composite at 0.70 over the last 30 trading days, with Nvidia at 0.63 and with the chip index at 0.58. Over 90 trading days the Nasdaq figure is 0.53. Bitcoin's daily close is at midnight UTC and the Nasdaq's at 4 p.m. in New York; that timing mismatch may bias the figures slightly. Cryptoticker's own calculation found 0.67 over a similar window.

Those figures don't fit the "almost inverse" remark. It may have described an earlier stretch: by Yahoo's closes, Bitcoin rose only about 7.6% from June 30 to November 18, not the roughly 30% attributed to Mitchnick.

Why would an asset with no AI in it trade with AI stocks? The likeliest reason: the people who own it now also own the AI trade. Since US spot ETFs launched in January 2024, Bitcoin has been a ticker in brokerage accounts and model portfolios, inside the same risk budget as Nvidia. When a risk manager cuts exposure, the most volatile liquid line often goes first. Grayscale's Zach Pandl reportedly named pressure on Nvidia and other AI-linked stocks as a possible source of that de-risking. Katie Stockton of Fairlead Strategies put it plainly to CNBC: "It's still an area that people might go to sell when they think that risk is higher in other areas."

The weekend makes it worse. Bitcoin is the only large risk market open on Saturday, so it absorbs the next leg of a Friday panic alone. That is when this low printed.

Verdict: strong. AI moves Bitcoin through the balance sheets that own both.

2. Competing for the same speculative dollar

The second story is that AI is now where speculative money wants to be, and crypto is where it comes from. "The price of the bitcoin dropped 22.7% in the 30 days before Anthropic's IPO and fell 22.3% in the month leading up to SpaceX's June 12 listing," CNBC calculated, and hedged in the same piece: "that can't be said with precise certainty. Crypto also has had other pressures."

The listing day looked like a funding trade. Anthropic raised $81.3 billion at $83 a share, plus a $26 billion concurrent placement (Renaissance Capital; reported totals vary with what they include). "Liquidity events can elevate correlations across asset classes quickly," Matthew Sheffield, CIO of Sharplink, told CNBC. "The first 15 min of [Thursday's] Anthropic IPO put that on full display. Tech stocks, value stocks, oil, gold, crypto all sold off as traders sold assets to buy shares in the secondary market."

CoinShares was measured: "There has been plenty of discussion about AI drawing capital away from crypto, and there is probably something in that. But the relationship works both ways." Two IPOs preceded by the same drop is a pattern of two. No source I found traces money out of crypto and into AI shares.

Verdict: partly. Plausible and widely argued, not measured.

3. The miners became AI landlords

This channel is slower, and it is the one that changes Bitcoin itself. A Bitcoin miner is a company that owns cheap power, land, cooling and a grid connection. That is exactly what an AI lab needs.

CoinShares estimated in September that AI hosting earns these companies about $1.5 million of profit per megawatt, against about $500,000 from mining, and that at least 35 EH/s of hashing power is already scheduled to leave the listed miners. "A BTC recovery is unlikely to reverse the AI transition," it said (via The Block). IREN signed a $9.7 billion AI cloud contract with Microsoft in November 2025. Riot's latest quarterly report names the tenant of its 910-megawatt Corsicana campus: Anthropic, on a 15-year lease worth about $28 billion.

So miner stocks now carry two risks, Bitcoin's price and the AI buildout's financing, and November hit both. From October 30 to November 20, computed for this article from Yahoo closes, Bitcoin fell 23.8%. IREN fell 50.6%, CleanSpark 36.2% and MARA 24.8%. Riot, which announced the Corsicana lease on November 4, rose 26.7%. One reading: the market paid for a powered-land lease, where the tenant builds the data halls, and sold AI exposure the miner must finance itself. MSCI's November list deletes IREN from the MSCI USA Index too.

The network itself held up. Hashrate averaged about 1,030 EH/s over 30 days (Hashrate Index, November 16). Difficulty fell 9.13% on November 14, the largest cut since June, which looked like miners giving up. Luxor traced it elsewhere: "Texas curtailment can account for roughly three-quarters of the drop." The 35 EH/s leaving for AI is about 3.5% of the network (my arithmetic).

The stress is in the margins. Hashprice, a miner's daily revenue per unit of hashing power, was $30.45 per PH/s on November 16, at cycle lows. CoinShares put the average cash cost of mining one bitcoin at about $75,500 in the second quarter, well above Saturday's price. And Glassnode says the miners sold: "MARA and Riot reported sales of about 28K BTC this month, while Strategy kept buying."

Verdict: partly, and structural. This month it split miner stocks: some fell far harder than the coin, some held up better.

4. Machines in the order book

The scary version is AI trading agents stampeding together. The visible amplifier was ordinary leverage: exchange risk engines closing positions that could no longer meet margin, about $2.4 billion in a day. Those engines run on rules, not models. The widely shared claim that 65% to 80% of crypto volume is "AI-driven" has no published method behind it that I could find, and none of those figures separate a market-making script from an LLM.

Regulators describe a risk, not an event. Asked on November 11 whether AI will raise correlation across markets, the Bank of England's Jonathan Hall said: "at this point we don't know, but there is a risk that it will" (as reported by NCFA Canada). A lab study points to the mechanism. An ICLR 2026 workshop paper found that when retrieval-augmented LLM trading agents read overlapping information, the mean pairwise correlation of their actions rose from 0.04 to 0.41, in crypto backtests.

As for skill: in Nof1's Alpha Arena season one, in late 2025, six frontier models each traded $10,000 of crypto for about two weeks, and four lost money. An independent analysis by Magnus Ross found a random strategy matched the best model in about 10% of runs. "I am skeptical that there is any evidence from this iteration of the benchmark that any of these LLMs can trade skilfully," he wrote.

The tokens that sell the AI-agent story fell harder than Bitcoin. Over the 30 days to Saturday, Virtuals lost 42.4%, AIXBT 37.2% and elizaOS 43.7%, against Bitcoin's 26.2% (CoinGecko). They traded like leveraged bets on the AI narrative.

Verdict: no evidence. Herding is real in the lab, and the risk grows as agents read the same sources.

5. Agents as customers

The bullish AI story says software agents will need money that software can hold, and that money will be crypto. Some of it is. x402, the open protocol for agents paying per request, has settled more than $42 million across 12 chains, but only $274,000 in October 2026. More than 99% of its value on Base over the past 30 days settled in USDC, a dollar stablecoin (Fireblocks, November 3). Block added Lightning to x402 in September, so an agent can pay in bitcoin. I found no figure for how many do.

Verdict: no. Measured agent payments are tiny.

The scorecard

ChannelVerdict this month
Shared holders and risk appetiteStrong. 30-day correlation with the Nasdaq 0.70; Bitcoin fell on both AI crash days
Capital pulled into AIPartly. A 22% drop before both the SpaceX and Anthropic listings; no flow data found links them
Miners as AI landlordsPartly, and structural. IREN −50.6% and Riot +26.7% against Bitcoin's −23.8%; hashrate held near 1,030 EH/s
AI trading agentsNo evidence. Leverage and rule-based liquidations amplified the fall
Agents as customersNo. About $274,000 of x402 payments in October; on Base, almost all USDC

What the chain says

On-chain data reads like capitulation by recent buyers. Glassnode puts the realized price, the average price at which every coin last moved, at $53.7K; Saturday's low came within about 3% of it. Recent buyers' average cost is $74.1K, so they are deep in the red. "The crash brought conditions seen at past local bottoms: realized losses at capitulation levels, leverage flushed and the largest bids since August below price," Glassnode wrote on November 17, after the mid-month cascade. The same note added: "Broad demand is still missing, with stablecoins shrinking, miners selling and money leaving the ETFs."

Open interest shrank in dollars, not in coins. Open interest on Binance's main Bitcoin perpetual went from 96,320 BTC on October 31 to 96,082 BTC on November 21, while its dollar value fell from $7.45 billion to $5.66 billion (Binance data). Glassnode noted traders adding shorts into the decline.

Past cycles give a calendar, not a depth:

CyclePeak to trough
2013–15−86.9% in 410 days
2017–18−84.3% in 363 days
2021–22−77.6% in 376 days
2025–26, so far−56.4% in 411 days

Computed for this article from Bitstamp and Coinbase prices. On timing, this bear market has run as long as the last three. On depth, it is far shallower. CoinShares rejects the copy-paste: "There are increasingly bearish calls for Bitcoin to revisit US$40,000, usually based on applying the drawdowns of previous cycles to the current one. We believe this is too simplistic." Its base case is a range of roughly $60,000 to $90,000.

If you build agents that trade or pay

This month has lessons for software, not just for markets.

  • Diversify what your agents read. The herding in the ICLR paper rose with overlapping information. Agents that retrieve the same headlines tend to make the same trades.
  • Keep the brakes outside the model. Position caps, loss limits, no leverage and a kill switch belong in deterministic code. Alpha Arena showed models losing money; your guardrail should not depend on the model's judgment.
  • Assume nobody is watching. The low printed around 06:00 UTC on a Saturday. Limits must hold with nobody on your team awake.
  • Treat correlation as a regime. 0.53 over 90 days, 0.70 over 30. A hedge sized on a calm window can fail on the day you need it. Test on the crash days.
  • Budget in stable units. A 29% drop inside one month can break an agent that holds Bitcoin as working capital. On Base, agent payments already settle almost entirely in USDC.
  • Expect adversarial AI on the other side. Chainalysis found that scams with on-chain links to AI vendors take $3.2 million per operation, against $719,000 for those without. An agent that pays must verify who it pays.

What to watch

  • Monday, November 23: the US stock open, and Strategy's weekly filing, which usually comes on Mondays.
  • Around November 26: the next difficulty adjustment. Early estimates run from about +10% to +15% as Texas power eased. A rise would support the view that curtailment, not exits, drove the last cut.
  • November 27: a Deribit options expiry worth about $6.7 billion (my calculation from Deribit's data).
  • November 30, at the close: Strategy and IREN leave the MSCI USA Index.
  • December 8–9: the Fed meeting, with a quarter-point hike priced at about 95%.
  • December: ERCOT's audit of data centers waiting to connect to the Texas grid, which miners and AI hosts share, is due (DLA Piper).
  • December 25: the largest open Deribit expiry, about $9.4 billion (my calculation).
  • Every bad Nvidia day: if channel one is right, the chip index leads and Bitcoin follows.

My read

This is opinion, labeled as such.

Bitcoin did not become an AI stock because of anything AI does to Bitcoin. It trades like one because the same accounts now hold both.

The deeper AI effect is in megawatts. Listed miners now measure their power against what an AI lab would pay for it. That ties part of Bitcoin's hashrate to AI demand, in an odd direction: an AI boom pulls power away from hashing, and a cooler buildout slows the exodus. In that narrow sense, this month's AI scare could prove good news for the network's hashrate, even as it hurt the price.

What I don't buy is either extreme story. I see no sign AI bots crashed Bitcoin, and AI agents are not about to rescue it. My guess for the next few months: until AI-capex fears settle or the holder base changes, Bitcoin falls on the bad Nvidia days and trades on its own story on the quiet ones.