Bitcoin just gave back every gain it made in May, slipping below 77,000 after Trump told Iran the clock is ticking. Oil ripped, yields climbed, and 563 million in crypto longs got liquidated. Strategy bought the dip anyway, scooping up nearly 25,000 BTC for 2 billion last week. Anthropic is making an aggressive push into legal AI with 12 plugins and 20-plus connectors. Google killed the Chromebook and replaced it with a Gemini-native laptop called Googlebook. And Bitcoin Depot, once North America's largest Bitcoin ATM operator, filed for Chapter 11. Let's get into it.
So Bitcoin is back under 77,000 and the mood has flipped fast. The trigger was geopolitical. Trump warned Iran the clock is ticking, oil shot back up to around 110 dollars a barrel, and risk assets got hit across the board. Ether took it even harder than Bitcoin, with Tom Lee at Fundstrat pointing out the inverse correlation between crude prices and ETH selling pressure.
The liquidations were brutal. 563 million in leveraged longs wiped out in a day, with Ether and Bitcoin taking the biggest hits. Some analysts are now openly debating whether we're staring at a classic sell-in-May setup, with downside targets as low as 65,000 or even a revisit to the mid-30s in a worst case. The bears point to collapsing US bond markets and Japanese investors dumping nearly 30 billion in US Treasuries in Q1, the largest quarterly net sale since 2022. That puts upward pressure on yields, which historically pressures Bitcoin.
The bulls counter that the buyer base is now far more institutional than in 2018 or 2022. And the evidence is right there: Strategy added 24,869 BTC last week for 2.01 billion, bringing total holdings to 843,738 coins. Roughly 97% of that buy was funded by STRC preferred share sales. Long-term holders aren't selling, exchange balances are near six-year lows. The structural bid is real. But short-term holders are underwater, and that makes the market vulnerable to exactly these kinds of macro shocks.
Oh, and one more thing worth flagging: Citi published a note saying Bitcoin faces an outsized quantum computing threat as breakthroughs accelerate. Not a today problem, but the timeline is compressing. Worth keeping on the radar.
Anthropic had a big week, and the most interesting move isn't another model release. It's a frontal assault on the legal industry. They dropped 12 role-specific Claude plugins covering things like M&A due diligence, contract review, employment handbook drafting, and bar exam prep. Alongside that, more than 20 MCP connectors wiring Claude directly into DocuSign, LexisNexis, Westlaw, Everlaw, iManage, and the rest of the legal software stack.
The whole strategy is built around grounding. Instead of letting Claude hallucinate citations, the connectors pull live verified case files and primary law into context. Descrybe announced a complementary integration that plugs 300 million structured US primary law records directly into Claude, with citator functionality and source verification. Pricing is aggressive: 25 dollars a month standalone, 50 for the full platform. That's a fraction of what Westlaw or LexisNexis charge for their AI tools.
Anthropic is also doing something interesting on the access-to-justice side. They highlighted that roughly 80% of civil litigants in the US appear in court without lawyers. So they're partnering with legal aid organizations like the Free Law Project and Courtroom5 to give those users free access to the connectors. It's a productivity play for big law and a market expansion play for everyone else.
The business backdrop is wild. Dario Amodei said Q1 revenue and usage grew 80x year over year on an annualized basis, blowing past their 10x internal target. The company is reportedly valued around 380 billion now, with the next round potentially targeting north of 900 billion. An IPO might not be far off.
And separately, Anthropic announced a 200 million dollar four-year partnership with the Gates Foundation focused on global health, disease forecasting, and education in low and middle-income countries. Public goods, evaluation benchmarks, health datasets. Whether that pencils out as philanthropy or strategic positioning to shape AI governance standards, probably both.
Public Bitcoin miners just reported, and the results tell you a lot about where this industry is going. Short answer: nobody is purely mining anymore.
Marathon, now just MARA, posted Q1 revenue of 174.6 million, down 18% year over year, missing estimates. They reported a 1.3 billion dollar net loss, but almost all of it, roughly 1 billion, was unrealized mark-to-market losses on their Bitcoin holdings. Operationally, hash rate hit a record 72.2 exahash, up 33%. They mined 2,247 BTC. Cost per Bitcoin jumped to over 40,000 dollars thanks to rising network difficulty. The big strategic move: acquiring Long Ridge Energy, adding about 505 megawatts of capacity and 1,600 acres, with a 100 megawatt AI buildout planned for 2027 and online by mid-2028. Power cost roughly 15 dollars per megawatt-hour. They also cut headcount 15%.
CleanSpark told a similar story. Revenue 136 million, down 25%. Net loss 378 million, of which 263 million was non-cash Bitcoin mark-to-market. They mined 1,799 BTC, kept margins above 40%, brought power costs down to 5.2 cents per kilowatt-hour. They sit on 13,561 Bitcoin worth around 925 million, plus 260 million in cash and 400 million in untapped Bitcoin-backed credit. ERCOT-approved capacity now at 585 megawatts.
Both companies are pivoting hard into AI and high-performance computing. The pitch is: mining funds the infrastructure, AI tenants monetize the spare capacity. Whether that thesis holds depends on whether hyperscalers actually want to rent from miners or just build their own. The macro setup is tough: lower Bitcoin price, higher difficulty, and a market that's stopped giving these companies a premium just for being Bitcoin proxies. The ones that survive will be the ones with cheap power and real diversification.
Metaplanet reported Q1 and the numbers are a Rorschach test. Headline net loss: 725 million dollars. But strip out the non-cash Bitcoin valuation markdown of 737 million, and the underlying business actually looks great. Revenue up 251% year over year. Operating profit up 283%. They didn't sell a single Bitcoin. They added 5,075 BTC at an average price around 79,000, bringing total holdings to 40,177.
That makes Metaplanet the third-largest publicly listed Bitcoin holder globally and roughly 87% of all Bitcoin held by listed Japanese companies. Their long-term target is 210,000 BTC by the end of 2027. That's audacious. They funded the latest accumulation with 12.2 billion yen in share issuance in February, 40.7 billion in March, plus a 500 million dollar Bitcoin-backed credit facility with about 302 million drawn.
CEO Simon Gerovich also met with Japanese lawmaker Junichi Kanda to push for regulatory clarity, and Metaplanet has proposed Japan's first perpetual preferred shares linked to Bitcoin, pending approval. They're essentially building the Saylor playbook adapted for Japanese capital markets, with the added wrinkle of investing in stablecoin issuer JPYC and launching a US-based asset management subsidiary.
Here's the thing to watch though. Cost basis on the full stack is around 104,000 per BTC. With Bitcoin at 76,000, that's roughly a 32% unrealized paper loss. The strategy works as long as they can keep raising capital at or above NAV and as long as Bitcoin trends up over their accumulation horizon. If mNAV compresses or Bitcoin chops sideways for 18 months, the dilution math gets ugly fast.
Meanwhile, Capital B added 15 million in Bitcoin to its treasury, one of only four corporate treasury announcements in May. The pace of new entrants is slowing. The treasury trade isn't dead, but the easy money phase is clearly over.
One thing to chew on: Iran is reportedly building a platform called Hormuz Safe, offering ship insurance for Strait of Hormuz transits with premiums settled in Bitcoin. Projected revenue above 10 billion dollars. If that's real, it's the most concrete test yet of Bitcoin as neutral money in a sanctioned, geopolitically contested corridor. Doesn't matter what you think of the regime. Watch whether the network actually gets used that way.