The biggest business story of the week has nothing to do with crypto — on the surface. On Wednesday, July 22, AMD agreed to invest up to $5 billion in Anthropic as part of a strategic partnership under which the AI company will deploy up to 2 gigawatts of AMD Instinct MI450 GPUs on Helios rack-scale systems, with the first gigawatt scheduled for the first half of 2027, per CNBC and Bloomberg, citing the Wall Street Journal. The chip purchases over the deal's term are expected to run to tens of billions of dollars.
Here is the Bitcoin angle, and it is not forced: 2 gigawatts is roughly the power draw of a top-five national Bitcoin mining industry. Every one of these AI megadeals is a bid for the same inputs Bitcoin mining runs on — grid-connected land, transformers, cooling and cheap long-term power contracts. And the buyers on the other side of the table are, increasingly, the miners themselves. One detail in the deal's fine print makes the point: Anthropic's existing infrastructure partners already include TeraWulf — until recently known primarily as a Bitcoin miner, per Analytics Insight.
The conversion is no longer hypothetical — it's on the balance sheets
The 2026 numbers are stark. IREN — a company that mined Bitcoin at scale and holds zero BTC in treasury — signed a $9.7 billion deal with Microsoft covering 76,000 NVIDIA GB300 GPUs across 200 megawatts at its Childress, Texas campus, alongside a separate $3.4 billion five-year AI cloud contract with NVIDIA, per Altrady's industry review. Core Scientific is raising $3.3 billion in junk bonds to build six facilities leased to CoreWeave for 12 years — a pipeline expected to generate roughly $10 billion in revenue, per Blockchain Council. TeraWulf is in the Anthropic camp. Among the majors, only Marathon has stayed comparatively cautious, per crypto.news.
The economics explain everything. A megawatt pointed at hashing earns whatever the Bitcoin protocol pays — and with BTC near $66,000, hashprice remains pinned near multi-year lows, a squeeze we documented when the production-cost floor broke earlier this month. A megawatt leased to a hyperscaler earns a contracted, dollar-denominated yield for a decade, underwritten by the best credit in corporate America. One revenue stream is volatile and halves every four years; the other looks like a bond. Boards have stopped debating.
The hashrate vacuum is already visible
This is not a forecast — it is showing up in the chain's own telemetry. Network hashrate has retreated from its early-2026 peak near 918 EH/s, difficulty fell to its lowest level of 2026 after consecutive downward retargets in July, and at July's hashprice roughly one in five miners was operating at a loss — the backdrop we covered in our miner capitulation update. Industry analysts now openly describe a "hashrate vacuum": capacity exiting not because miners went bankrupt, but because their megawatts got a better offer. As ETHNews puts it, AI contracts are replacing hashrate.
What it means for Bitcoin: three effects, two of them underpriced
Effect one — less forced selling. Miners have historically been structural sellers, liquidating coins to fund power bills and fleet upgrades. Q1 2026 saw record miner treasury sales. Every miner that converts to AI hosting stops selling mined BTC — because it stops mining. At the margin, the AI pivot removes a persistent source of sell pressure. This effect is real, immediate, and mildly bullish.
Effect two — a cheaper network to attack, eventually. Hashrate is Bitcoin's security budget. A sustained, structural decline — driven not by price cycles but by a permanently better use for the underlying power — lowers the cost of mounting a majority attack over time. To be clear about the data: at hundreds of exahashes, Bitcoin remains extraordinarily expensive to attack, and difficulty adjustment is doing exactly its job. But the direction of travel matters, and 2026 is the first year the marginal megawatt is leaving for reasons other than a bear market.
Effect three — the miners that remain get paid more. Difficulty retargeting is mechanical: every exahash that leaves for AI raises the revenue per exahash of whoever stays. Low-cost operators with locked-in power — and no better AI offer for their sites — are quietly inheriting the margin their departing competitors abandoned. If hashprice mean-reverts while difficulty stays suppressed, the surviving fleet's economics improve without Bitcoin's price moving at all.
Why this time is different from 2021
Bitcoin has survived a hashrate exodus before. When China banned mining in mid-2021, roughly half the network went dark in weeks — and recovered to new highs within a year as machines were re-racked in Texas, Kazakhstan and elsewhere. That precedent is why many holders shrug at hashrate charts. But the mechanism was different in kind: 2021 was a relocation of capacity chasing the same revenue stream. 2026 is a repricing of the underlying asset — the megawatt — by a competing buyer with deeper pockets and contracted returns. Machines can move across borders; they cannot outbid a hyperscaler for the substation they are plugged into. When a Childress campus flips from ASICs to GB300 racks under a decade-long lease, that capacity does not come back at the next difficulty adjustment.
The financing markets have noticed the same asymmetry. Core Scientific's $3.3 billion raise is being done in the junk bond market — debt investors who would never underwrite hashprice risk are comfortable underwriting a 12-year CoreWeave lease. Capital markets will now fund the conversion of mining infrastructure at a scale they were never willing to fund mining itself. That one-way financing door is the strongest evidence that the pivot is structural.
A side effect for investors: mining stocks are no longer Bitcoin proxies
For years, retail investors bought public miners as leveraged BTC plays. That trade is quietly dying. IREN holds zero Bitcoin in treasury and now earns Microsoft- and NVIDIA-contracted revenue; Core Scientific's forward revenue is dominated by an AI tenant. A portfolio that holds miner equities for Bitcoin exposure increasingly owns data-center REIT economics with a crypto ticker. Investors who want Bitcoin exposure should look at the flows behind the spot ETFs — or the asset itself — rather than assuming the old correlation still holds.
The tell to watch
Watch the next several difficulty retargets against Bitcoin's price. In every previous cycle, price recovery pulled hashrate back within months. If BTC holds the $64,000–$66,000 range or better through Q3 and hashrate doesn't recover, that is the confirmation that the AI bid has permanently repriced miner megawatts — and that the hashrate vacuum is structural, not cyclical. The AMD–Anthropic deal did not cause that shift. It is simply the week's clearest evidence of how much capital is on the other side of the auction.
FAQ: AI, miners and the hashrate
What did AMD and Anthropic announce? On July 22, 2026, AMD agreed to invest up to $5 billion in Anthropic, which will deploy up to 2 gigawatts of AMD Instinct MI450 GPUs, with the first gigawatt targeted for H1 2027.
Why does an AI chip deal affect Bitcoin? AI data centers compete for the same grid-connected power and sites that Bitcoin miners use — and pay contracted dollar yields that currently beat mining economics, pulling capacity off the network.
Which miners have pivoted to AI? IREN ($9.7B Microsoft deal, $3.4B NVIDIA contract), Core Scientific ($3.3B raise for CoreWeave-leased facilities) and TeraWulf (an Anthropic infrastructure partner) lead the conversion; Marathon has been more cautious.
Is falling hashrate bad for my Bitcoin? Near term it reduces miner sell pressure and raises surviving miners' revenue per exahash; long term, a structural decline in the security budget is worth monitoring. The network remains extremely costly to attack today.
What signal confirms a structural shift? Hashrate failing to recover across several difficulty retargets while Bitcoin's price holds steady or rises — decoupling hashrate from price for the first time.