Bitcoin broke below 60,000 dollars overnight, hitting its lowest level since October 2024 and triggering close to a billion dollars in liquidations across crypto. The selloff was driven by spot ETF outflows, a strengthening dollar, and traders now pricing in a Fed rate hike by October instead of cuts. Meanwhile, Meta's Llama 4 is quietly becoming the default open-source model for enterprises, with over 650 million downloads. Strive added another 759 BTC. Metaplanet bought a Japanese securities firm to build Bitcoin-linked bond products. And Micron's blowout earnings sent AI stocks ripping while crypto bled. Let's get into it.
Bitcoin touched 59,000 dollars on Wednesday, the lowest it's been since October 2024. That's the third dip below 60K this year, and it caps an eighth straight month of decline from the 2025 high. The mechanics here are pretty ugly. US spot Bitcoin ETF outflows jumped from 114 million to 469 million dollars in a single day. Seven straight weeks of negative flows have dragged total ETF assets under management down to 77.5 billion dollars, from 113 billion at the end of 2025. When issuers see redemptions, they sell physical Bitcoin. That's mechanical selling pressure into a market that already has weak demand. On-chain data tells the same story. Large wallets holding between 10 and 10,000 BTC dumped about 45,000 coins in just eight days. Bitcoin supply held at a loss hit a record high of 10.83 million BTC. And the Fear and Greed Index dropped to 12, deep in extreme fear territory. The macro backdrop isn't helping. The dollar surged, capital is rotating into AI stocks, and traders are now betting the Fed hikes rates later this year rather than cutting. Micron posted blowout earnings and ripped 16%, but crypto didn't follow the AI rally. That divergence matters. For most of this cycle, Bitcoin traded as a high-beta tech proxy. Right now, it's getting sold while semiconductor stocks rally. Not all bad news though. 21Shares released its mid-year report and called this a normal post-halving pullback, not a capitulation. Bitcoin is still above the aggregate investor cost basis of around 54,000. Long-term holders control a record 14.8 million coins. And 21Shares is sticking with a 100,000 dollar year-end target as their base case. There's also a 525 million dollar buy wall sitting between 60,500 and 65,000 dollars, intersecting with a major liquidation zone. That's the battleground. If that level breaks cleanly, the next stop is the 58,000 dollar cluster where most of the long leverage sits.
While crypto sells off, the AI story keeps compounding. Meta's Llama 4 has crossed 650 million downloads and is now running on roughly 400,000 servers worldwide. Production deployments have tripled since January. The pitch to enterprises is simple: fine-tune it, host it on your own infrastructure, and pay a fraction of what you'd pay OpenAI or Anthropic for API calls. Over 340 startups since January 2025 list Llama as their primary foundation model. Legal tech, healthcare admin, supply chain, financial compliance. Perplexity and others are showing you can hit near-frontier capabilities by fine-tuning open weights on proprietary data. The paradox here is delicious. AWS, Azure, and Google Cloud all offer Llama 4 as a managed service. So Meta's free open-source model has become a revenue line for its biggest cloud competitors. Meta spends billions developing this stuff and gives it away, betting that a thriving open ecosystem ultimately benefits Meta's core advertising business more than licensing fees ever would. The new wrinkle is regulatory. The Trump administration is pressuring Meta to voluntarily submit Llama models for government security reviews through the Center for AI Standards and Innovation. OpenAI, Anthropic, Google, xAI, and Microsoft are already participating. Meta is resisting because a 30-day pre-release review window directly conflicts with the whole point of open-source: you publish weights, and the world has them instantly. If a two-tier market emerges, where federal contractors and regulated industries prefer vendors with government clearance, Meta could lose ground in healthcare, finance, and defense. Meta's response has been to layer enterprise features on top. The latest Llama 4 Agents update adds end-to-end encryption, role-based access controls, tamper-proof audit trails, and compliance tooling aligned with GDPR, the EU AI Act, and China's 2026 AI mandates. Translation: they're trying to make the open model enterprise-ready without giving up the open part.
Speaking of enterprise AI, the customer service space just had a busy week. Talkdesk, Zoom, Verint, and eGain all rolled out new agentic AI tooling, and the common thread is that they're trying to make AI agents deployable without months of prompt engineering. Talkdesk launched Agent Builder. You describe what you want an AI agent to do in plain language, the system ingests your SOPs and policies, and it produces a production-ready agent. Pre-deployment validation checks for gaps and inconsistencies. They're claiming a compression from weeks of iteration down to hours. Zoom announced Agent Architect and an Agent Performance Suite for its Virtual Agent product. Same idea: generate voice or digital agents from a prompt, then simulate customer scenarios before deployment and track resolution rates, containment, and cost per resolution after launch. Verint unveiled something called Agent Factory, which is explicitly designed to orchestrate hybrid workforces of human and AI agents together. Prebuilt CX agents, central prompt governance, bring-your-own-model support, and human handoff routing. eGain dropped an AI Agent for Zoom Contact Center with case-based reasoning, full audit trails, and policy-grounded answers. The pattern across all four is striking. We're moving past the demo phase where AI agents were impressive but unreliable. The new wave is built around governance, auditability, simulation testing, and human-in-the-loop approval. That's what enterprises actually need to deploy this in regulated environments. There's also a legal angle worth watching. A US arbitration giant just rolled out what they're calling a legal layer for agentic commerce. As AI agents start executing transactions on behalf of users, someone has to answer the question: what happens when something goes wrong? And House Democrats are now pressing the SEC about AI agent investment advisers making consequential decisions for retail investors. The deployment is racing ahead of the rulebook.
Bitcoin treasury companies are evolving past the simple buy-and-hold playbook, and two stories this week show what that looks like. First, Metaplanet in Japan. They're acquiring Siiibo Securities, a licensed financial instruments firm, for 2.1 billion yen, roughly 13 million dollars. The deal closes in July and Siiibo gets renamed Metaplanet Securities. This is the opening move in something they're calling Project Nova: building a Bitcoin-centric financial ecosystem inside Japan's regulated framework. With Siiibo's Type I license, Metaplanet can offer Bitcoin-linked bonds, tokenized securities, and yield products to Japanese investors. They already hold around 40,000 BTC, making them the largest publicly disclosed corporate holder in Asia. The strategy is straightforward: stop just sitting on Bitcoin, start productizing it. Separately, Metaplanet raised 255 million dollars from global institutions with a structure they're calling first-of-its-kind. Fixed-strike warrants with a moving NAV clause that only allows exercise when the stock trades above 1.01 times mNAV. The point is to make sure any new share issuance is accretive to Bitcoin per share, not dilutive. They're targeting 210,000 BTC total. Then there's Strive, the company that formed from the Semler Scientific merger. They bought another 759 BTC for about 50 million dollars, average price around 65,850, bringing total holdings past 19,800 BTC. What's interesting is how they're funding it. They issue a perpetual preferred stock called SATA that pays a 13% Bitcoin-linked dividend. In one week of SATA payouts, the mechanism generated enough cash to buy around 600 BTC. They're using structured finance instead of convertible notes, which is a different model than Strategy. And Strategy itself is feeling the squeeze. Saylor spent 1.5 billion in May repurchasing convertible notes, which reduced debt but also drained cash that backstops their preferred stock dividends. The treasury company model is getting tested. When Bitcoin rallied, every structure looked clever. With BTC below 60,000, the financial engineering is meeting reality. Companies with average cost bases above current spot prices are sitting on paper losses, and the question is whether their funding mechanisms can survive a prolonged drawdown without forcing dilution.
Here's the question worth asking today: if Bitcoin is selling off while AI stocks rip on Micron earnings, what does that say about the digital gold thesis? Either this is a temporary correlation break and BTC recouples on the next macro shift, or institutional flows just told us something uncomfortable about what Bitcoin actually is in 2026.