Bitcoin is fighting to hold $62,000 after a soft inflation print pulled it back from a deeper test of $60,000. Strategy just bought another 1,550 BTC for $101 million, one week after that strange little 32-coin sale that spooked the market. Microsoft is rolling out so-called Autopilots, autonomous agents that act without prompting, and lining up massive enterprise deals with Atos, KPMG, and the NHS. The Bitcoin DeFi project Botanix is shutting down with a brutal post-mortem: users just didn't care. And Japan's three megabanks are quietly preparing a joint stablecoin for next March. Let's get into it.
Bitcoin spent the last 24 hours doing what Bitcoin does in a jittery macro tape: testing nerves. It slid toward $62,000 local lows, traders openly modeled a path to $50,000, and then the May core CPI came in at 0.2%, slightly cooler than feared, and BTC bounced back above $62,000.
But don't confuse a bounce with a bottom. A few things are worth paying attention to here. First, US spot Bitcoin ETFs have quietly bled back down to net asset levels last seen right after Trump won the election in November 2024. All of that post-election institutional enthusiasm? On paper, gone. Cointelegraph is citing data showing institutions and ETFs dumping nearly 2,000 BTC a day, roughly 450% of daily new supply. That's a lot of distribution.
Second, Bitwise is making the case that Bitcoin is acting as a canary in the coal mine, leading a broader risk-off move while global liquidity is still elevated. Bitcoin and gold fell together this week, which tells you traders are bracing for a Warsh-led Fed that stays hawkish, not for a pivot.
And then there's Strategy. Michael Saylor announced another 1,550 BTC bought at an average of $65,332, funded through the ATM equity program, lifting the stash to 845,256 BTC. MSTR popped 5% premarket. But there's now a real debate on X over whether that purchase was actually dilutive, because Strategy's own BTC Yield metric ticked down from 13.0% to 12.8%. Saylor is pushing back, but the fact that the argument exists at all tells you the easy narrative phase of the Strategy trade is over. From here, every buy gets scrutinized on whether it actually creates value per share, or just adds coins.
The honest read: Bitcoin is in a grind, not a collapse. $60,000 is the line that matters. Lose it convincingly, and the $50,000 charts start mattering a lot more than anyone wants.
Microsoft is doing something this week that I think gets underrated because it's wrapped in enterprise jargon. They are quietly trying to own the agentic AI layer of corporate work, and the deal flow is staggering.
Start with NHS England. They're rolling out Microsoft 365 Copilot to 505,000 clinicians and staff, scaling from a 30,000-user pilot that reportedly saved each user about 43 minutes a day on admin. Then Atos: 56,000 employees across 54 countries, all getting Copilot, plus Microsoft's new Agent 365 governance layer to manage up to roughly 19,000 AI agents across the organization. KPMG is going further, extending to over 276,000 professionals globally, managed through their own internal Workbench on top of Agent 365.
That's nearly 850,000 knowledge workers, just in the announcements from this week, being onboarded onto a single vendor's agent stack.
And at Build 2026, Microsoft introduced what they're calling Autopilots, a step beyond Copilots. The first one is called Scout, powered by something called OpenClaw, and the pitch is that it runs in the background on your behalf, across Teams, Outlook, OneDrive, SharePoint, with its own Entra identity for attribution. They also announced Execution Containers, MXC, to sandbox what these agents can actually touch on a Windows machine.
Two things to take from this. First, the enterprise AI war is essentially over for the next few years, and Microsoft won the distribution layer. OpenAI builds models, Anthropic builds models, but the agents that actually log into your company's SharePoint and rewrite your quarterly report? Those are going to be Microsoft-branded for a long time.
Second, and this matters for anyone thinking about security: we are about to have tens of millions of autonomous software identities inside corporations, with their own credentials, acting on behalf of humans. The governance story Microsoft is telling, Agent 365, Entra, MXC containers, is real, but it's also the only story. If it breaks in a high-profile way, the entire agent rollout pauses. Watch for the first big Agent 365 breach. It's coming.
Botanix is shutting down. If you haven't heard of it, that's kind of the point. It was a Polychain and Placeholder-backed Bitcoin Layer 2, EVM-compatible, launched on mainnet around July 2025. Users have until July 9 to withdraw, after which the Federation sweeps remaining BTC and any non-BTC tokens are simply gone.
The post-mortem from the team is unusually honest. Quote: it did not work, at least not in this market and not in this timeline. They concluded demand for Bitcoin-native DeFi just wasn't there. Bitcoin holders, it turns out, mostly want to hold Bitcoin. Shocking.
Now contrast that with what's actually getting traction on Bitcoin this week. Second launched Bark on mainnet, their implementation of the Ark protocol. No Lightning-style channel management, no pre-allocated liquidity, off-chain transaction trees that share on-chain UTXOs. They shipped a full Rust SDK, a wallet daemon called Barkd, integrations with Noah, Arke, Satsigner, BTCPay Server. A real self-custody payments stack, not a yield casino.
Meanwhile BitGo and Voltage expanded their Lightning partnership, putting institutional-grade Lightning behind a simple API for banks, exchanges, and custodians. Lightning capacity hit a record 5,606 BTC, about $490 million, last December. The network is reportedly processing over 8 million transactions a month at a 99% success rate in well-configured setups. Businesses moving from on-chain to Lightning are reporting 80%-plus cost reductions.
The pattern is pretty clear. The Bitcoin builders chasing Ethereum-style DeFi keep dying quietly. The ones building actual payment rails, Ark, Lightning, Taproot Assets, RGB on Lightning through projects like Bitlight, keep shipping and keep growing. The market is telling you what Bitcoin wants to be, and it isn't Solana with a Bitcoin logo. It's money that moves.
While everyone argues about chatbots, healthcare AI is shipping into actual hospitals at a pace that's worth noting.
This week alone: Philips got FDA 510(k) clearance for Elevate Plus on its EPIQ Elite and Affiniti ultrasound systems, with automated abdominal measurements running at about 93% accuracy versus manual readings. Subtle Medical got FDA clearance for SubtleHD CT, their 11th cleared product, pushing AI image enhancement from MRI and PET into CT. Clarius got clearance for an Ejection Fraction AI that calculates left ventricular function in real time from standard cardiac views, particularly useful in emergency and point-of-care settings.
And in Taiwan, Foxconn, yes, the iPhone assembler, partnered with NVIDIA to deploy AI across hospitals under a $1.5 billion Healthy Taiwan fund. Their platform, CoDoctor, covers cardiovascular, oncology, and ophthalmology. There's a nursing robot called Nurabot reportedly saving nurses 2 to 3 hours a day. Digital twins via NVIDIA Omniverse cut deployment time by about 40%. Taiwan now has 85 cleared medical AI products in routine use across major hospitals, serving over 14 million patient encounters a year.
But here's the part that should make you sit up. There's reporting from On Healthcare Tech that elements inside HHS, CMS, and FDA, aligned with the MAHA crowd, are quietly laying groundwork for autonomous AI that can diagnose and prescribe with no human in the loop. A Utah pilot is already running chatbot-driven prescription refills, with plans to remove human oversight, and the state Medical Licensing Board is demanding an immediate suspension.
The legal problem is enormous. No state medical board, no FDA pathway, currently licenses fully autonomous AI to practice medicine. Corporate Practice of Medicine laws in many states bar non-physician entities from controlling clinical judgment. If the AI prescribes and something goes wrong, who carries the malpractice insurance? The developer? The hospital? The model weights?
The genuinely useful stuff, image enhancement, ejection fraction measurement, nursing logistics, is shipping with clear FDA pathways and obvious accountability. The autonomous-doctor push is a different category of risk entirely. Expect two years of lawsuits, lobbying, and at least one very public failure before any of it gets resolved.
One observation to end on. This week, Microsoft announced agents that act for you without asking. A US Layer 2 announced that nobody wanted its yield products. And Bitcoin held $60,000 while the smartest institutional buyers in the world quietly distributed. The technologies that win don't ask permission and don't need narratives. They just get used. Watch what people actually do with their time and their money, not what they post about.