Bitcoin is fighting to hold 80,000 after a brief slip below, with options traders betting the dip is short-lived. Spot Bitcoin ETFs just logged their sixth straight week of net inflows, the longest streak in nine months. The CLARITY Act markup is set for May 14 in the Senate Banking Committee, and it could either become a milestone or fall apart over a fight about Trump family crypto ethics. CleanSpark mined 640 BTC in April while pivoting hard toward AI hosting. And a new study says developers using AI coding tools feel 20% faster but are actually 19% slower. Let's get into it.
There's a study making the rounds that engineering leaders should probably print out and tape to the wall. METR ran a controlled experiment with experienced developers, median around 10 years of experience, using tools like Cursor Pro and Claude 3.5 and 3.7 Sonnet. The developers reported feeling 20% faster with AI assistance. When researchers measured actual task completion times on matched tasks, they were 19% slower. That's a 39-point gap between perception and reality.
This is a big deal because almost every internal report justifying AI tooling spend is built on satisfaction surveys and vibes. Developers love these tools. They feel productive. The dopamine hit of watching code generate is real. But when you measure actual throughput on real work, the gains often vanish, and on complex tasks they reverse.
The nuance matters. AI helps a lot with boilerplate, tests, and documentation. It hurts on novel architecture, complex reviews, and security-sensitive work. Which tracks with another data point from this week: Sherlock Forensics audited 50 AI-built apps and found 92% had critical-severity vulnerabilities. ProjectDiscovery's 2026 report shows two-thirds of cybersecurity practitioners now spend more than half their time manually validating AI output rather than fixing real issues.
Meanwhile GitHub just shipped a usage metrics API breakdown for Copilot code review by comment type, security, bug risk, and so on, with counts of suggestions posted versus actually applied. That's the right direction. If your developers apply 5% of security suggestions, that tells you something. If they apply 80% of style suggestions, that tells you something different. Stop measuring AI by how fast people feel. Measure what gets shipped, and what gets reverted three days later.
Two announcements this week tell you everything about where AI infrastructure is headed. Terrestrial Energy and Riot Platforms signed an MOU to deploy multiple 390 megawatt molten salt reactors at Riot's data center sites in Texas and Kentucky, targeting up to 4 gigawatts of nuclear capacity. Separately, NANO Nuclear and Supermicro signed an MOU to put 15 megawatt KRONOS microreactors directly inside AI data centers, with Supermicro tailoring server racks and cooling to share a closed loop with the reactor.
The Riot deal is the more concrete one. IMSR is a Gen IV design, runs on standard low-enriched uranium under 5%, claims about 44% efficiency, and Riot is even open to natural gas as a bridge fuel during buildout. Riot wants energy resilience for its compute workloads. Terrestrial wants a flagship deployment.
The NANO deal is more speculative. Non-binding MOU, NRC microreactor reviews are still early stage, NANO has substantial historical losses and basically no revenue. Realistic deployment timelines are early 2030s at best.
But zoom out. The Bitcoin miners pivoting to AI hosting, like Riot and CleanSpark, both made it clear this quarter that the moat isn't hash rate anymore. It's gigawatts under contract. CleanSpark has 1.8 gigawatts under contract with 808 megawatts utilized, and is openly courting hyperscale AI tenants. Riot already has a 50 megawatt AMD contract producing 33.2 million dollars in data center revenue last quarter at a 90.8% gross margin on the lease portion.
The punchline: Bitcoin miners spent a decade learning how to acquire cheap power and build out behind-the-meter substations. That skill set is suddenly the most valuable thing in tech. AI didn't kill mining. AI made miners landlords.
Bitcoin is sitting in an interesting spot. It rallied 36% off the 60,000 lows, briefly poked above 82,000, then dipped below 80,000 this week. The RSI flashed an overbought signal not seen since early 2026, and CryptoQuant's Julio Moreno is pointing out that despite the recovery, we're still technically in a bear market structure. There's a heavy resistance cluster between 84,000 and 92,000 that traders are struggling to crack.
ETF flows are sending mixed signals. Six straight weeks of net inflows, the best streak since summer 2025 when 7.57 billion came in over seven weeks. Then this week, 268 million in outflows in a single day. Options traders, though, aren't panicking. They're treating the dip as a backtest, not a breakdown.
The macro setup matters more than the chart. The dollar index is weak, and the eventual appointment of a new Fed chair is the next big catalyst the market is pricing. Strategy's MSTR is forming an ascending triangle that some technicians read as an 80% rally setup toward 350.
A few things worth flagging. The Swiss Bitcoin reserve referendum campaign just lapsed, they couldn't gather enough signatures, so the Swiss National Bank is not going to be forced to hold Bitcoin. Strike CEO Jack Mallers got asked whether Wall Street is a threat to Bitcoin. His answer was blunt: if Wall Street can kill Bitcoin, Bitcoin was never going to make it anyway. Hard to argue with that. The protocol doesn't care about ETF flows. Holders care, traders care, but the chain keeps producing blocks at 80,000 or 800,000.
The Digital Asset Treasury landscape keeps getting weirder and bigger. Strive crossed 15,000 BTC after another 444 BTC purchase at 76,307 per coin. Their treasury is now worth around 1.2 billion. Strive completed its Semler Scientific acquisition in January and is openly running a Saylor-style playbook, including holding STRC preferred stock issued by Strategy itself.
Metaplanet in Tokyo is the more aggressive story. They've reached 40,177 BTC after Q1 buys, making them one of the top five public corporate Bitcoin holders globally. CEO Simon Gerovich is targeting 100,000 BTC by year end, financed entirely through equity and warrants, no debt. The catch: their FY2025 results showed a 620 million dollar net loss driven by unrealized Bitcoin valuation losses, with an average cost basis around 107,000 per coin. Stock is down 28.6% year to date. When your equity is a leveraged Bitcoin proxy, that cuts both ways.
Then there's DDC Enterprise, formerly DayDayCook, holding 2,383 BTC alongside an Asian food media business. That's the hybrid DAT model, and it's becoming common in Asia where Japan's flat 20% corporate Bitcoin tax rate makes it workable.
On the regulatory front, the CLARITY Act markup is scheduled for May 14, which Coinbase's policy chief called a big step forward. But the markup may fall apart over a fight about ethics restrictions on federal officials with crypto holdings, which is a not-so-subtle reference to the Trump family's various tokens and ventures. SEC Chair Paul Atkins gave a notable speech this week floating a 1990s-style innovation pathway for on-chain trading systems, comparing it to how the SEC handled electronic exchanges decades ago. And Kraken's parent Payward applied for an OCC charter, joining Coinbase, Ripple, BitGo, Circle, Fidelity Digital Assets, and Paxos in the federal crypto bank queue.
The pattern is clear. Crypto is being absorbed into the regulated financial system at every layer, treasury companies on the asset side, federally chartered crypto banks on the institutional side, tokenized funds from BlackRock on the product side. Whether that's good or bad depends on what you wanted Bitcoin to be in the first place.
One observation to close on. The two biggest stories this week, the AI productivity paradox and miners pivoting to nuclear-powered AI hosting, are the same story told from opposite ends. The hype is real, the spending is real, the buildout is real. But somewhere between the gigawatts and the generated code, somebody has to actually ship work that doesn't break. Measure that, or you're just buying expensive feelings.