Bitcoin is back near $67,000 after Trump announced a framework deal with Iran to reopen the Strait of Hormuz, sending oil down roughly 5% and lifting risk assets. But the rally is shaky — the Bank of Japan just hiked rates 25 basis points to 1%, the highest since 1995, and traders are bracing for global liquidity to tighten. Strategy added another 1,587 BTC last week for $100 million, pushing its stack to 846,842 coins. BlackRock launched a new bitcoin income ETF using covered calls. Six federal agencies are sprinting to finalize GENIUS Act stablecoin rules by July 18. And the BIP-110 activation window is closing in fast, with less than 10,000 blocks to go on what could be the most contentious Bitcoin fork debate in years.
Let's start with the tape. Bitcoin hit an intraday high near $67,300 yesterday on the back of the US-Iran framework, then gave most of it back. The market wants the deal actually signed before pricing it in, and you can see that hesitation in the ETF flows — bitcoin funds saw outflows Monday while ether, XRP, Solana, and Hyperliquid funds all took in cash. Most of the bitcoin outflow was just Grayscale's GBTC bleeding, as usual.
The more interesting story is Japan. The BOJ raising to 1% is the kind of move that quietly drains global liquidity. The yen carry trade has been a massive hidden bid for risk assets for years. Some analysts are now warning of a potential 26 to 38% drawdown in BTC if liquidity conditions worsen. Others are looking at the same chart and seeing a double-bottom setup with weekly RSI divergence that could push BTC to $100,000 before October. Pick your poison.
What I find more compelling than the price predictions is the accumulation data. Glassnode shows buyers added over 250,000 BTC in the $59,000 to $67,000 range, with the Accumulation Trend Score at its strongest level of the entire drawdown. That's broad-based — retail and whales both. Mining difficulty also just dropped 10%, the 11th largest downward adjustment ever, which gives surviving miners more breathing room. And Michael Saylor is still buying, still arguing Bitcoin doesn't need staking or yield, that returns should come from a five-layer credit and equity stack built around BTC rather than from the protocol itself. Whether you buy his framing or not, Strategy keeps acting on it.
July 18. That's the date. Six federal agencies — the OCC, FDIC, NCUA, Treasury, FinCEN, and OFAC — have until July 18 to publish final rules implementing the GENIUS Act. Comment periods closed June 9, which means regulators have a 35-day sprint to finalize a framework that will define who gets to issue dollar stablecoins in the United States.
The OCC proposal sets a $5 million minimum capital floor for new federal stablecoin issuers. It creates a three-tier liquidity structure: at least 10% redeemable same-day in Fed deposits or cash, 30% redeemable within five business days in high-quality liquid assets, 60% in standard reserves. The FDIC has made clear stablecoin holders get no deposit insurance, period, regardless of the issuer's charter. And critically, the act bans US-compliant stablecoins from paying any yield or interest to holders. That's a statutory ban, not a regulatory choice.
Who wins here? Large bank holding companies with capital to spare — JPMorgan, US Bancorp, the usual suspects. Who loses? Crypto-native and smaller fintech issuers, who either need a state charter capped at $10 billion in issuance, or they need to find another business. Once final rules drop, issuers have about 120 days to comply. By September, the US stablecoin market could look very different.
There's already a workaround emerging. Ethena's USDe is a delta-neutral synthetic dollar, not a reserves-backed stablecoin. Because it hedges crypto collateral with perpetual futures rather than holding cash, it doesn't meet the GENIUS Act's definition of a payment stablecoin — and therefore the yield ban doesn't apply. USDe was paying around 4% APY earlier this year. The OCC is trying to extend the yield ban to affiliates through a rebuttable presumption, but it's not clear that captures synthetic dollar mechanics. Expect more of this. If you ban yield in one bucket, the yield migrates to a bucket the rule doesn't name. Meanwhile State Street just launched a money market fund specifically aimed at managing stablecoin reserves, joining BlackRock and Franklin Templeton. The plumbing is being built.
Two protocol-level debates are heating up simultaneously, and both are governance stress tests.
First, quantum. Coinbase's cryptography advisory council — actual top-tier cryptographers — just released their assessment. Quantum computers are not an immediate threat. But sufficient quantum capability is probably about a decade out, and migration planning needs to start now. The exposure is real: roughly 1.7 million BTC sit in early pay-to-public-key addresses, which already exposed their public keys on chain. That includes coins likely belonging to Satoshi. Another 5 million or so are exposed through address reuse. Call it 6.7 to 6.9 million BTC potentially vulnerable long-term.
The technical solutions are converging. BIP-360 and BIP-361 work on post-quantum signature schemes. Hourglass would cap how much vulnerable BTC can be moved per block to prevent a sudden flood. PACTs let users commit to future quantum-safe addresses without exposing data now. The council deliberately refused to weigh in on the political question — whether vulnerable coins, including Satoshi's, should eventually be frozen, burned, or left untouched. That's a community decision. Their point: don't wait for that fight to be resolved before doing the engineering work. Institutional due diligence checklists are already asking about quantum risk, so this isn't theoretical anymore for capital allocators.
Then there's BIP-110. This is the proposal to restrict non-financial data in Bitcoin transactions — aimed squarely at the inscription and spam problem. Activation is set at block 961,632, less than 10,000 blocks away. The controversy isn't really about the data restriction itself. It's about the activation mechanism: a 55% miner signaling threshold with a mandatory enforcement backstop. That's a low bar, and it shifts activation away from broad consensus toward a low-threshold enforceable rule. Critics are calling it the most divisive fork battle since 2017's Bitcoin Cash split. Most miners and major pools aren't aligned, and the economic support looks insufficient for a sustained chain split. But the activation window will absolutely create volatility, and exchanges and custodians are quietly drawing up contingency plans — potentially pausing deposits and withdrawals to handle replay risk. Even if nothing dramatic happens on chain, this tells you something about where Bitcoin governance pressure is building.
The coding agent space is moving from "AI helps you write code" to "AI runs the entire software lifecycle." Three releases this week make that concrete.
Xiaomi open-sourced MiMo Code, a terminal-native coding harness under MIT license. It's a fork of OpenCode with their own memory architecture bolted on. The claim: on long-horizon tasks of 200 or more steps, MiMo Code's win rate against Claude Code exceeds 65%. It runs on MiMo V2.5, a 310B parameter multimodal model with 15B active and a 1-million-token context window. Self-reported benchmarks, obviously, and the free hosted access runs through Xiaomi servers — so data residency is a real consideration. But the architectural bet is interesting: multi-layer memory with persistent files, scratch notes, and per-task state, specifically engineered for tasks that don't fit in a single context window.
Cohere went the opposite direction — small and self-hostable. North Mini Code is a 30B mixture-of-experts model, 8 experts active per token, 3B active parameters at inference, 256k context, Apache 2.0, runs on a single H100. The pitch is agentic coding for organizations that need data residency and predictable costs. It's verbose — generates roughly 3x the output tokens of comparable models — which is a real cost factor, but the single-H100 deployment story is compelling for enterprises that can't ship code to a hosted API.
And then there's Alibaba's Qoder 1.0 and Factory's 2.0 release, both pushing the same vision: agentic engineering. Not a single assistant, but a coordinated team of agents handling planning, research, coding, review, testing, and delivery, with humans focused on architecture and risk. Augment Code is reporting hard numbers from their Cosmos platform — 81.3% of incidents handled by agents, on-call engineers merging 44% more PRs per week. And Y Combinator just revealed Locus Founder, which lets you text a business idea via iMessage and the AI builds and runs the business, settling payments in USDC.
Meanwhile, Meta's strategy on open-source AI is fracturing. Llama 4 launched as the open enterprise play with claims of $47 billion in incremental GPU demand through 2027. But Meta also released Muse Spark as a closed, hosted proprietary model, and Meta Superintelligence Labs put out MSI-1 with no open weights at all. The open frontier model story is getting complicated, even at Meta. Grayscale, predictably, is pointing at the recent US government order forcing Anthropic to cut access to certain models as evidence that decentralized AI tokens deserve a premium. That's a stretch, but the underlying point — that centralized AI creates single points of regulatory failure — is real.
July 18 is the date that matters most in the next month. If the GENIUS Act final rules land on schedule, the US stablecoin market reorganizes around well-capitalized banks within 120 days, yield migrates to synthetic dollars and offshore issuers, and the regulatory perimeter starts looking very different from the crypto-native one most of this industry was built on. Watch where the yield goes. That tells you who actually got regulated and who just got rebranded.