Big day in markets. Strategy just made its first Bitcoin sale since 2022, and the market is not taking it well. Bitcoin is trading below $70,000, the lowest level in roughly two months, ETF outflows just hit their longest redemption streak since launch, and a $150 million Polymarket position is in chaos because of the timing of the disclosure. Meanwhile, Apple is reportedly distilling a version of Google's Gemini to run on iPhones ahead of WWDC next Monday. And on the AI front, Penn Medicine is rolling out clinical AI agents across its electronic health record, while Boston Children's says OpenAI's tools helped surface more than 40 rare disease diagnoses. Let's get into it.
So here's the story that's dominating everything. Strategy, Michael Saylor's company, sold 32 Bitcoin to cover dividend payments on its preferred stock. 32 coins. Out of 843,738. A rounding error in terms of size. But it's the first time the company has sold any Bitcoin since 2022, and the symbolism is enormous.
The disclosure came in an 8-K filing on June 1, but the sale itself happened in late May. That timing detail is now the center of a roughly $150 million Polymarket battle over whether the contract resolves yes or no. The traders who correctly bet Strategy would sell are watching the platform potentially deny their payouts because the public disclosure landed after the deadline. Pure information asymmetry, surfaced.
Markets reacted hard. Bitcoin dropped under $70,000, briefly testing $69,000. Spot Bitcoin ETFs have now bled cash for 11 straight sessions, totaling around $3.4 billion. Derivatives open interest is at 773,000 BTC, near record highs, and funding rates are still elevated despite weak spot demand. That's a setup that tends to resolve violently in one direction or the other.
Now, the bull case. Tom Lee is calling this classic bottom behavior. Strive added 2,500 Bitcoin last week to reach 19,000 BTC total. Capital B, the European treasury company, is asking shareholders for authority to raise up to $122 billion to buy more Bitcoin. So the treasury thesis is not dead. It's narrowing.
And here's what actually matters about Strategy's move. Two weeks before this sale, they repurchased $1.5 billion of their 2029 convertible notes at roughly an 8% discount to par. They cut convertible debt from $8.2 billion to $6.7 billion. They issued $2 billion of variable-rate preferred stock. They bought 24,869 Bitcoin in the same window. The 32 coin sale was almost certainly a mechanical action to top up the USD reserve that pays preferred dividends, which now sits at $871 million.
This isn't capitulation. It's plumbing. But the market is treating it like a regime change because for three and a half years, the narrative was that Saylor only buys. That story is over. What replaces it is a more complicated one about a Bitcoin-backed financial machine that has obligations to service, and sometimes that means trimming a tiny slice. The question is whether the market can hold that nuance, or whether it just sees a seller and panics.
Pivot to Apple. WWDC kicks off Monday June 8, and the leaks are flying. The big one: Apple is reportedly working to distill Google's multi-trillion-parameter Gemini model down to something that can run on an iPhone, and use cloud Gemini for the heavier lifting.
Let that sink in. Apple, the company that has spent a decade telling you AI happens privately on your device, is now licensing the core intelligence of its flagship assistant from its biggest competitor. Siri, in iOS 27, will reportedly be a three-layer system. Simple stuff runs locally on Apple silicon. Medium-complexity tasks go to Apple's Private Cloud Compute. And the hardest queries get routed to Google Cloud running Gemini, with Nvidia's confidential computing platform handling encryption in transit and at rest.
There's also a standalone Siri app coming. Chat-style interface, past conversation history, document and photo uploads, system-wide integration with Mail, Messages, Photos, Calendar. It's basically ChatGPT, but with access to your personal data on the device. The Dynamic Island gets used as the surfacing layer for quick responses. Spotlight gets rebuilt around AI cards.
The strategic read here is interesting. Apple lost the on-device AI race. They've been working on their own foundation models for years, the results have been underwhelming, and the original Apple Intelligence rollout was a debacle. So they did what Apple does when they're behind. They wrote a check. The Google deal gives them best-in-class capability immediately, and they get to keep the privacy narrative by wrapping it in Private Cloud Compute and confidential computing.
But there's a tension. If your iPhone's smartest features require a round trip to Google's data centers, you've fundamentally changed what an iPhone is. It's no longer a device. It's a thin client. Battery and latency become external dependencies. And the privacy story gets harder to tell when the model weights belong to your search-engine competitor.
For developers and for the broader AI competitive landscape, the WWDC announcement next Monday is going to be a tell. If Apple emphasizes on-device, they're trying to control the narrative. If they lean into the Gemini partnership openly, they're admitting the game has changed. My guess is they thread the needle and most users will never know which layer is answering their question. Which is probably the point.
AI in healthcare is having a quiet but real moment, and it's worth paying attention because this is what actual deployment looks like, not benchmarks.
Penn Medicine announced a partnership with K Health to roll out a suite of AI clinical agents across its electronic health record. Starting with virtual urgent care, expanding to in-person primary care, cardiology, dermatology. Both patient-facing and clinician-facing agents, integrated into the EHR rather than bolted on as a separate app. Penn is framing this as infrastructure, not a feature. That language matters.
OpenAI separately publicized that Boston Children's Hospital has used its tools to help surface more than 40 rare-disease diagnoses. Important caveat: this is not a diagnostic product. It's clinicians using the model to organize case information and find relevant medical literature faster. No published validation data, no accuracy metrics, no disclosed model details. So treat the 40 number as marketing, not science. But the underlying use case, helping doctors do the cognitive grunt work in cases with sparse data, is exactly where these tools should be earning their keep.
The more rigorous story is Bayesian Health, which just became the first company to receive FDA clearance for an AI-powered continuous monitoring system for sepsis. This is a different category entirely. Years of real-world deployment, published Nature Medicine studies with around 750,000 patients, and at MemorialCare they're reporting more than double sepsis sensitivity while reducing alert volume. Early engagement with the tool, within the first hour, is associated with a 3.6% absolute mortality reduction.
And internationally, Alibaba's DAMO Academy signed a strategic partnership with Zhejiang University's Second Affiliated Hospital on May 26, focused on cardiovascular screening, cancer diagnosis, pathology, and imaging.
The pattern across all of these: the wins are not autonomous AI doctors. The wins are workflow tools that earn clinician trust over years, integrate into existing systems, and have measurable outcome data. The losses, and there will be losses, will come from products that try to skip those steps. The FDA clearance for Bayesian is the more important headline than the OpenAI press release, even if it gets less attention.
Last segment. Bitcoin miner earnings season is wrapping up, and the data is telling a very clear story. The pure-play Bitcoin miner is becoming an endangered species.
Riot Platforms reported Q1 2026 revenue of $167 million. Bitcoin mining contributed $112 million of that, but the data center arm contributed $33 million and engineering services another $22 million. All-in mining costs hit $44,629 per Bitcoin. AMD just expanded its contracted capacity with Riot to 50 megawatts. The CEO called Q1 an inflection point and explicitly said the company is transitioning into a revenue-generating data center operator alongside its mining activities.
HIVE reported full-year revenue of $297.8 million, up 158%. Their BUZZ HPC division, which is the AI infrastructure piece, grew 94% to $19.5 million in revenue with $35 million in contracted annual recurring revenue. They're planning a 320 megawatt AI gigafactory in the greater Toronto area designed to host more than 100,000 GPUs, with capital expenditures of around 3.5 billion Canadian dollars.
Then there's Cango, which had a brutal quarter. Net loss of $261 million driven by impairment charges on mining machines and Bitcoin fair-value changes. They produced 1,266 Bitcoin at an average cash cost of $76,928. They're also pivoting toward AI compute with a platform called EcoHash.
Fidelity Digital Assets just put out a report noting that the 30-day average Bitcoin hashrate and mining difficulty have actually declined about 8 to 9% from highs, which they interpret as miners reallocating capacity toward AI workloads with better margins.
This is the structural story under the price action. Bitcoin's security budget is paid for by miners, and miners are increasingly running mixed businesses where Bitcoin is one workload among several. That's not inherently bad. It can mean more resilient operators with better balance sheets. But it also means the network's security is implicitly tied to whether AI compute remains more profitable than mining, which is a dependency that didn't exist five years ago. Worth watching closely as the next halving cycle approaches.
One observation to close on. Apple is licensing Gemini. Bitcoin miners are renting their power to Nvidia GPUs. Strategy is selling Bitcoin to pay preferred dividends. The pure plays are getting hybrid, whether they want to or not. The companies that win the next five years will be the ones who admit that earliest.