Bitcoin is having its worst stretch of the year. The price punched below 62,000 dollars overnight, wiping out roughly 1.5 billion in long positions and tagging the 200-week trend line that defined the 2022 bear market. Spot Bitcoin ETFs have now bled 4.4 billion dollars across 13 straight sessions. Meanwhile OpenAI rolled out a major GPT-5.5 update and a specialized life-sciences model called GPT-Rosalind, Interactive Brokers wired Claude directly into live trading accounts, and Japan's ruling party is pushing hard for spot crypto ETFs and yen stablecoins. Plenty to get through.
Let's start with the price action because it's brutal. Bitcoin dropped below 62,000 dollars, briefly touching 61,300 before bouncing to around 62,500. That's the first time since February we've seen these levels. Over two sessions, about 3 billion in liquidations got flushed out, with more than 600 million in long positions wiped in a single move. Derivatives traders are loading up on 60,000 dollar puts, betting this isn't over.
What's interesting is the chart itself. Bitcoin just tagged the 200-week trend line, the exact level that acted as resistance throughout the 2022 bear market. The weekly RSI is approaching its lowest reading in six years. More than half the bitcoin in circulation is now sitting on unrealized losses, which historically has marked every cycle bottom. So bulls have a setup, but it's the kind of setup that only looks good in hindsight.
The narrative war is more interesting than the price. Michael Saylor argues this is capital rotation into AI, not a bitcoin problem. Charles Schwab's Jim Ferraioli agrees, saying bitcoin is simply losing the momentum trade to AI stocks, IPOs, and gold. Presto Research points out every bitcoin drawdown this year has coincided with rallies in those exact assets as Fed rate cut expectations get scaled back.
Then there's the supply side noise. Mt. Gox-linked wallets moved 10,422 BTC on June 2nd, about 739 million dollars, right as the market was already weak. Strategy disclosed a bitcoin sale that became a whole Polymarket dispute, with UMA voters eventually ruling the June 1st disclosure counted for the June contract even though the actual selling happened in late May.
The darker reading from bears: this isn't rotation, it's exhaustion. The treasury company model is starting to crack. Bitmine's Ethereum bet is approaching a 9 billion dollar paper loss. FG Nexus just dumped another 17.8 million in ether. When leveraged treasury vehicles start unwinding, the reflexivity that powered the bull market runs in reverse. Pick your narrative, but the tape doesn't care which one you believe.
While bitcoin gets repriced, the AI agent story keeps building infrastructure underneath traditional finance. Two announcements this week make it concrete.
Interactive Brokers integrated Claude directly into live trading accounts. Clients can now connect Claude through a certified connector and have an AI agent analyze their portfolio, generate trading instructions, and prepare orders across 170 global markets. No API keys shared, no credentials on the client's machine. There's a human-in-the-loop approval layer, so the agent proposes, you confirm. Equities and ETFs at launch, other asset classes within a week. ChatGPT, Gemini and Grok integrations are in certification.
Morgan Stanley is taking a different angle. They're opening their ShareWorks and Equity Edge stock plan platforms to external AI agents that corporate clients run themselves. The agents bypass the human-facing software entirely and pull data directly. They're using the Model Context Protocol, the open standard that's quietly becoming the plumbing for all of this. Early access for a few clients now, expanding to about 3,400 administration clients next year. The strategic logic is sharp: handle more complex stock plans without hiring, and convert employees into wealth management clients as their stock vests.
And to confirm this isn't theory, Digital Quant 2026 just wrapped in Hong Kong after 60 days of live trading with autonomous AI agents running real money. 30 plus teams, 209 million USDT in trading volume, an interim high return of 48 percent. Agents like OpenClaw handled analysis, strategy adjustments, execution and risk management with minimal human input, across equities, metals and commodities, not just crypto.
The pattern is obvious. The interface to a brokerage used to be a screen and a human. Now it's increasingly an agent with credentials and a policy file. If you're building anything in fintech that assumes a human clicking buttons, that assumption has a shelf life.
OpenAI had a busy 48 hours. Three things worth tracking.
First, GPT-5.5 Instant got a quality update. The headline change is style: less bullet-heavy, more human writing, better pacing for practical tasks. The retirement schedule is the real news. GPT-4.5 is gone on June 27th after a 30-day sunset. The original o3 reasoning model retires August 26th with a 90-day sunset. OpenAI is consolidating compute and forcing everyone forward.
Second, and this is the more interesting one: GPT-Rosalind. It's a life-sciences specialized model, fine-tuned for medicinal chemistry, genomics and drug development on top of GPT-5.5's reasoning. The numbers are striking. On MedChemBench it scores 27.5 versus GPT-5.5's 25.1, while using 7.2 percent fewer tokens. On LabWorkBench, 63.2 versus 55.8 with 5.3 percent fewer tokens. Genomics analyses use 31 percent fewer tokens. Better and cheaper at the same time, in a vertical. Novo Nordisk is among the testers. Access is gated through a research preview for organizations with legitimate scientific use cases.
Third, GPT-5.5, GPT-5.4 and Codex are now generally available on Amazon Bedrock. That means enterprises can run OpenAI models inside AWS with AWS security, governance and existing commitment credits. Codex pricing matches OpenAI first-party rates. This quietly dissolves one of the bigger objections enterprise buyers had about OpenAI lock-in.
Two takeaways. One: specialization is back. The bitter lesson said scale beats domain expertise, but Rosalind suggests that once you have a strong general model, a domain-tuned version can win on both quality and cost. Expect more of these vertical models. Two: distribution is consolidating. OpenAI on Bedrock, Claude on Interactive Brokers, agents inside Morgan Stanley. The labs aren't trying to own the customer relationship anymore. They're trying to be the engine inside everyone else's product.
Two threads on the bitcoin infrastructure side worth pulling together.
Japan first. The Liberal Democratic Party formally asked Finance Minister Satsuki Katayama to build a legal framework for crypto ETFs and to promote yen-denominated stablecoins for Asian settlement. This builds on April's cabinet decision to reclassify crypto as financial products rather than payment tools, which is what makes ETF wrappers legally coherent. JPX has signaled it's ready to list crypto products within one to two years once the law is amended. Separately, as of June 1st, foreign stablecoins like USDC are now legally usable for payments in Japan through licensed intermediaries. USDT's status remains uncertain given its offshore reserve structure.
The strategic frame: Japan wants to be the regulated Asia hub and outflank Hong Kong and Singapore. SoftBank is positioned well here. PayPay owns 40 percent of Binance Japan, and PayPay has over 70 million users. If yen stablecoins and spot ETFs both arrive, SoftBank ends up owning a meaningful chunk of the payment-to-crypto-to-settlement pipeline without issuing a stablecoin or running a bank.
On the bitcoin-native side, two things happened. Tether is bringing USDT back to bitcoin through RGB and the Lightning Network, with Utexo as the first integrated platform. Tether co-led a 7.5 million dollar seed round in Utexo with BigBrain VC and Portal Ventures. That's a real endorsement of RGB as the path for USDT on bitcoin rails. The pitch is fixed USDT-denominated fees, instant Lightning finality, self-custody throughout. Lightning hit a record 1.17 billion in transaction value last November across 5.2 million transactions, so the underlying activity supports it.
And David Marcus launched Grid Global Accounts at Bitcoin 2026 Las Vegas. It's a dollar-denominated payment layer on bitcoin, built on Spark, connected to 175 million Visa merchants across 33 countries, scaling toward 100 by year-end. Lightspark is now a principal Visa member. Grid is sold to platforms and developers, not consumers, so the platforms own the customer relationship while bitcoin rails carry the payment underneath.
The through-line: stablecoins, regulated wrappers, and bitcoin settlement are converging into the same stack. Different jurisdictions are picking which layer they want to own.
One observation to end on. The bitcoin price is being driven by capital rotating into AI trades, while AI agents are being plugged directly into the brokerage accounts doing the rotating. At some point soon, the agents will be the marginal buyer of bitcoin too. Worth thinking about who's actually making the trade by this time next year.