Sunday, April 12th. Bitcoin slips below 71,000 dollars after President Trump orders the U.S. Navy to blockade the Strait of Hormuz, sending oil futures up 7 percent on Hyperliquid and rattling every risk asset on the board. Meanwhile, the SEC's own enforcement review admits that dozens of Gensler-era crypto cases produced no investor benefit, and Chair Atkins says a sweeping Reg Crypto proposal is one signature away from publication. Google quietly drops an offline AI dictation app on iOS that runs entirely on-device with no cloud connection, proving edge inference is production-ready at consumer scale. And Meta's open-source AI strategy keeps getting murkier as reports surface that upcoming models will only be partly open. Let's get into it.
Bitcoin is sitting on a knife edge at around 71,000 dollars this weekend after a sequence of geopolitical jolts that would have been hard to imagine even a month ago. On Saturday, President Trump posted that the U.S. Navy would begin blockading any and all ships trying to enter or leave the Strait of Hormuz. Oil futures immediately spiked 7 percent on Hyperliquid, and Bitcoin dropped below 71,000 dollars in a swift risk-off move.
This follows the collapse of U.S.-Iran negotiations on Friday. Vice President Vance confirmed that talks in Pakistan had ended without a resolution, sending crypto prices down between 1.5 and 2 percent across the board. The two-week ceasefire that triggered a relief rally earlier in the week now looks fragile at best.
But here's the interesting part. Despite all this, Bitcoin has not fallen off a cliff. Open interest hit 5-week highs near 25 billion dollars, and funding rates are mimicking the patterns we saw before the BTC collapse below 60,000 dollars, which some analysts read as setup for a short squeeze rather than further downside.
On-chain data supports a cautiously constructive view. Realized losses are declining, which typically signals seller exhaustion. Spot markets are shifting toward net buying. Fidelity's Jurrien Timmer has called this environment a real buying opportunity, arguing that strong corporate earnings are absorbing geopolitical shocks better than most people expect.
And then there's Ray Dalio's war thesis, published in TIME on April 9th. Dalio explicitly argues that his indicators point to a simultaneous breakdown of the monetary order, domestic political orders, and the geopolitical world order. The Iran conflict is the immediate trigger, but the structural claim underneath is dollar debasement, and that argument lands squarely in Bitcoin's favor.
Some analysts are flagging triggers for a surge to 88,000 dollars, citing ETF flows, macro factors, and on-chain supply dynamics. Others note that Bitcoin and Ether are each less than 10 percent away from levels that could signal a trend reversal. The SpaceX treasury disclosure is also notable: Arkham data shows 8,285 BTC, about 603 million dollars, sitting in Coinbase Prime custody, even as the company swings to a nearly 5 billion dollar loss ahead of its IPO push. And Michael Saylor, predictably, is signaling another Bitcoin purchase, marking what would be Strategy's 106th transaction since 2020.
One structural risk worth watching: over 80 percent of Bitcoin ETF assets now flow through Coinbase custody, roughly 74 billion dollars at risk of a single-point-of-failure choke point. That's a concentration problem the industry needs to address before it becomes a crisis.
Bottom line: the geopolitical backdrop is genuinely scary, but Bitcoin's on-chain fundamentals and institutional plumbing suggest the market is absorbing shocks rather than cracking under them.
The SEC just did something remarkable. In its fiscal year 2025 enforcement review, the agency essentially admitted that its previous crypto crackdown was a mistake. Under former Chair Gary Gensler starting in 2022, the SEC pursued 95 enforcement actions against crypto firms specifically for book-and-record violations, racking up about 2.3 billion dollars in fines from those cases alone. The new report calls that approach a misallocation of resources and criticizes the pursuit of media headlines over actual investor protection.
The review singled out 7 crypto registration cases and 6 definition-of-a-dealer actions that it says produced no direct investor harm or protection. In other words, the agency chased firms that weren't actually hurting anyone. For context, the SEC filed 456 total enforcement actions in fiscal year 2025, including 303 standalone cases and 69 administrative proceedings. The overall monetary relief rose to about 17.9 billion dollars. But the message from Chair Atkins is clear: we're redirecting toward fraud, market manipulation, and genuine abuses of trust.
Since February 2025, the SEC has dismissed enforcement actions against Coinbase, Binance, Kraken, Consensys, Cumberland DRW, Dragonchain, and Balina. That's a wholesale unwinding of the Gensler-era enforcement posture.
But the bigger news might be what's coming next. Chair Atkins says a sweeping rulemaking package called Reg Crypto is one step from publication. It's sitting with the White House Office of Information and Regulatory Affairs for final sign-off. Once cleared, it goes out for public comment.
Reg Crypto would address how the Securities Act of 1933 applies to crypto fundraising. It would create startup exemptions, clarify when crypto transactions are securities versus non-securities, and potentially include an innovation exemption for DeFi platforms. The goal is to give projects a legitimate path to raise funds without the ambiguity that has defined this space for years.
Atkins has been blunt that the SEC will proceed with or without congressional action, though he also urged the crypto community to stay politically engaged for the midterms. Senator Cynthia Lummis is separately pushing the CLARITY Act, warning this is the last chance to pass it before 2030.
Meanwhile, the CFTC is staking its own claim. Chair Mike Selig is arguing for exclusive regulatory authority over prediction markets, echoing the agency's ongoing court cases to cement jurisdiction.
The shift from regulation by enforcement to actual rulemaking is potentially the most important structural change for crypto in the United States since the ETF approvals. If Reg Crypto delivers what Atkins is promising, it could become a global benchmark for how digital assets are regulated.
Something quietly significant happened this week. Google released a free dictation app on iOS called AI Edge Eloquent that runs entirely on-device. No cloud connection. No subscription. No usage limits. It uses Google's Gemma models to do live transcription, automatically polishes your speech, and filters out filler words like um and uh. All of it happens locally on your phone.
This is not a research demo. This is a production app in the App Store, and it proves that edge inference at consumer scale is real, not theoretical. Zero latency, zero data leaving the device, zero server costs. Google plans to expand to Android and macOS, but the iOS-first launch is the statement: on-device AI can replace cloud for mainstream voice tasks right now.
This matters in the broader context of what Apple and Google are doing with personal AI. Apple has been rebuilding Siri from the ground up for iOS 27, expected at WWDC in June. The overhaul promises a redesigned conversational interface, deeper cross-app integration where Siri can pull data from email, add it to Maps, and text a friend in one flow, on-screen awareness so Siri can act on what you're looking at, and conversational memory across a session.
But here's the tension. Apple has four finished smart home products, the Apple TV 4K, HomePod 3, HomePod mini 2, and a new HomePad display, all sitting in a warehouse waiting for Siri 2.0. The hardware is done. The A17 Pro chips are in there. But Apple won't ship until the software experience is ready because the current Siri only hits about 78 percent accuracy compared to Google Assistant at 92 percent.
Apple is making a deliberate bet: delay hardware revenue to avoid another round of Siri is dumb headlines. That's a real strategic gamble when Amazon and Google are shipping competitive products with richer AI capabilities today.
The bigger picture is a fundamental shift in where AI computation happens. The old model was everything goes to the cloud. The new model, validated by Google's own app, is that meaningful AI work can run on the silicon already in your pocket. Apple's vertical integration from Neural Engine chips to iOS gives it a structural advantage for this approach. Google's strength is breadth, deploying models across platforms it doesn't control.
For developers, this opens up entirely new categories of privacy-first apps without server infrastructure. For users, it means AI that works offline, in airplane mode, in places with no signal. The shift from cloud-centric to edge-first AI is not coming. It's here.
Meta's relationship with open-source AI is getting complicated, and not in a good way.
Let's rewind. Llama 1, Llama 2, Llama 3, each release came with big proclamations about openness and democratizing AI. Llama 4 arrived with Scout, a 17 billion active parameter model with 16 experts and a 10 million token context window, the longest for any open-weight model, and Maverick, with 128 experts and 400 billion total parameters targeting GPT-4o-class performance. Both are multimodal, handling text, images, and video. Both trained on over 30 trillion tokens across 200 languages. Scout can run on a single H100 with INT4 quantization. These are genuinely impressive models released under open weights on Hugging Face.
But now reports from multiple outlets say Meta is developing proprietary frontier models, codenamed Avocado and Mango, and that the open-source versions of these will be stripped-down derivatives. Some training steps may be skipped. Some sub-models within the mixture-of-experts architecture may be held back. The open editions are expected to omit capabilities found in the closed versions.
Meta's new Chief AI Officer Alexandr Wang, who came over after Meta acquired 49 percent of Scale AI, is leading a hybrid strategy. The most capable models stay closed. Lighter versions get released openly. Safety is cited as the reason, but critics argue this is about competitive positioning dressed up as responsibility.
The term open-weight already carried an asterisk. Weights are public, but training data, code, and full reproducibility details are not. Now we're looking at a further narrowing where even the weights you get represent a diminished version of what Meta actually built.
Google, by contrast, just released Gemma 4 under Apache 2.0, a genuinely permissive commercial license. Four model sizes from 2 billion to 31 billion parameters, all handling text, image, and video. The edge models run under 1.5 gigabytes of memory on phones and Raspberry Pi devices. The shift from Google's custom Gemma license to Apache 2.0 is a real concession to the developer community.
So the landscape is inverting. Google, historically the more closed company, is becoming more permissive with its smaller models. Meta, which built its brand on openness, is pulling back on its most capable work. For developers and researchers who built workflows around the assumption that Llama would keep getting more open, this is worth watching carefully. The gap between Meta's open-source rhetoric and its actual release strategy is widening, and that matters for anyone building on top of these models.
Here's the thought to sit with. The two biggest stories this week, the Hormuz blockade rattling markets and the SEC admitting its own enforcement was misguided, are actually the same story told from different angles. Institutions, whether governments or regulators, are capable of dramatic reversals, sometimes within months. The world you're building for today may not be the world that exists next quarter. In Bitcoin, in AI, in regulation, the only durable advantage is building on foundations you actually control. Your keys, your models, your edge devices. That's it for today. I'm out.