Happy Sunday. A lot happened this weekend. Strategy just made its third-largest Bitcoin purchase ever, dropping $2.54 billion on 34,164 coins and pushing its total holdings past 815,000 BTC. Meanwhile, the DeFi world is reeling. The KelpDAO exploit drained $292 million and triggered a cascade that wiped more than $13 billion in total value locked across the sector in just 2 days. On the AI front, Meta is walking a tightrope between open and closed source, releasing Llama 4 open-weight while simultaneously shipping its first proprietary model. And Y Combinator just wired its first seed deal entirely in USDC. Let's get into it.
Let's start with Meta, because the AI strategy coming out of Menlo Park right now is genuinely hard to follow. On one hand, Meta released Llama 4, the latest in its open-weight model family. It's a 400-billion-parameter mixture-of-experts model, available on Hugging Face and Meta's developer portal. Meta claims state-of-the-art results on the MMLU Pro reasoning suite and says it beats GPT-4o on both MATH and HumanEval coding benchmarks in their internal tests. Yann LeCun was out there championing this as another big step for open AI — wider access, better safety research, less industry concentration. AWS and Azure will both offer it through managed APIs.
But here's where it gets weird. Just days before, Meta shipped Muse Spark — the company's first closed-source model. No public weights, no open license, API-only access through Meta's own infrastructure. It powers Facebook, Instagram, WhatsApp, Messenger, and the Ray-Ban Meta AI glasses. This thing came out of Meta Superintelligence Labs, which is led by Alexandr Wang, formerly of Scale AI. Meta reportedly invested $14.3 billion for a 49% stake in Scale AI to make that happen.
So you've got the same company releasing an open-weight model and a proprietary one within days of each other. The open-source crowd is understandably confused. Meta's 2024 manifesto was all about democratizing AI through open weights. Muse Spark is a pretty clean break from that promise. It's small, it's fast, it uses way less compute than Llama 4 Maverick, and Meta says it's competitive with OpenAI and Anthropic on many tasks, though not dominant across the board.
Meanwhile, Mistral quietly dropped Voxtral 4B, an open-weight text-to-speech model that might be more practically interesting than either Meta release. It does zero-shot voice cloning from just 3 seconds of reference audio, supports 9 languages out of the box, and reportedly beat ElevenLabs Flash v2.5 in human preference tests at a rate of 68.4%. It's open-weight, on Hugging Face, and you can run it locally on edge devices. That's the kind of open release that actually moves the needle for developers.
The broader trend is clear. The best frontier models are increasingly moving behind proprietary walls, while the open-source ecosystem is getting strong enough that you can build production-quality applications without them. It's a two-track world now.
Strategy, formerly MicroStrategy, just disclosed its third-largest Bitcoin purchase on record. 34,164 BTC for $2.54 billion, at an average price of roughly $74,336 per coin. That brings total holdings to 815,061 Bitcoin, acquired for a cumulative $59.02 billion at an average cost of approximately $75,577 per coin.
To put that in perspective, Strategy now holds over 3.7% of all Bitcoin that will ever exist. And they funded this latest buy entirely through sales of STRC, their perpetual preferred stock. No common shares diluted. No convertible notes issued. Just preferred stock sales. In fact, STRC had a single-day record capital raise of $1.16 billion on April 13th alone.
The capital structure is getting increasingly complex. There's about $8.25 billion in convertible notes at the top, then roughly $10.3 billion across multiple preferred stock series — STRC, STRK, STRD, STRF — and common MSTR shares at the bottom. The amplification metric, debt plus preferred capital divided by Bitcoin reserves, has risen to about 33%. That's meaningful leverage.
Saylor himself posted "Think Even Bigger" on X over the weekend, which historically means another purchase is coming soon. Analysts think the next buy could exceed 40,000 BTC. Strategy still has approximately $56 billion in remaining capacity across its at-the-market programs. If fully deployed at current prices, that could theoretically fund another 763,000 BTC, pushing them past 1.5 million coins — though obviously a buyer that size would move the market substantially.
The preferred stock angle is particularly interesting. STRC pays an 11.5% annual dividend and trades near its $100 par value with very low volatility. It's essentially a bridge product for pension funds, insurance companies, and bond funds to get Bitcoin exposure through a steady-income instrument they're comfortable with. Strategy is also proposing to switch from monthly to semi-monthly dividend payments — 24 per year — to reduce the stock price cyclicality around ex-dividend dates.
Meanwhile, Morgan Stanley launched its own spot Bitcoin ETF on April 8th. MSBT, listed on NYSE Arca, with a sponsor fee of just 0.14%, the lowest of any Bitcoin ETP. It pulled in $116 million in its first 7 trading sessions. Tiny compared to Morgan Stanley's $1.9 trillion in assets, but that's exactly why it matters. It signals where the puck is going. Bitcoin fund inflows overall hit $1.4 billion last week, the second-strongest since January, with total AUM across crypto investment products reaching $154.8 billion.
Now to the story that dominated crypto this weekend. On Saturday, an attacker drained roughly 116,500 rsETH tokens — about $292 million worth — from KelpDAO's cross-chain bridge. LayerZero, whose infrastructure was involved, says the attackers compromised 2 RPC nodes that the verifier relied on and DDoS'd the rest. They attribute the attack to North Korea's Lazarus group. Critically, LayerZero says the exploit only worked because Kelp had ignored their multi-verifier recommendations. In other words, this was a single point of failure in Kelp's setup, not a protocol-level vulnerability.
But the damage didn't stay contained. The stolen rsETH was used as collateral on Aave to borrow wrapped ether, leaving Aave holding what amounts to bad debt backed by compromised tokens. Aave's total value locked dropped $8 billion in about 24 hours. The AAVE token fell nearly 20% to $89.50. A $300 million borrowing spike hit the protocol as stablecoin markets tightened.
And it spread further. Roughly $10 billion fled the broader DeFi sector over the weekend. Total DeFi TVL dropped more than $13 billion in 2 days as users pulled funds from lending and yield protocols across the board. Multiple protocols froze markets tied to rsETH. Ledger's CTO called 2026 DeFi's worst year for hacks. Coming on the heels of the Drift Protocol breach on April 1st and Venus's March post-mortem, confidence is genuinely shaken.
The Curve Finance founder made a sharp observation here. The contagion could have been contained if DeFi lending used isolated pools rather than shared liquidity. But that comes at a direct cost to capital efficiency — the very thing that makes DeFi attractive in the first place. It's the fundamental tension. Composability and shared liquidity are DeFi's superpowers, but they're also the attack surface.
Bitcoin itself held up relatively well. It's trading near $74,335 after a modest 1.6% pullback, which is remarkable given that geopolitical risk also spiked — Iran reimposed controls on the Strait of Hormuz, oil jumped 5.7%, and European equity futures dropped. Some technical analysts see $75,000 as a potential new floor, with a $7.9 billion options expiry this week where Bitcoin sits above max pain. On-chain activity hit an 8-year low, but price barely moved. Wall Street has effectively replaced retail as the marginal buyer.
Let's close with something that didn't make as many headlines but might matter more long-term. Y Combinator just completed its first seed investment paid entirely in USDC. The recipient is Totalis, a prediction markets infrastructure startup. The $500,000 went out on Solana in 3 on-chain transfers — a $1 test transaction, then $124,999, then $375,000 — directly to Ramp, the company's treasury platform.
YC announced earlier this year that it would start offering stablecoin funding options for its Spring 2026 batch. Standard deal stays the same — $500,000 for 7% equity — but payouts can now happen in USDC or USDT on Base, Solana, or Ethereum. The rationale is simple: it's faster than wires, cheaper for cross-border transfers, and removes a ton of friction for founders in Africa, Southeast Asia, and Latin America who might wait days or weeks for traditional banking rails to settle.
On the institutional side, Mastercard is testing stablecoin settlement with something called SoFiUSD, aiming to speed up card transaction clearing by bridging traditional finance and blockchain settlement. And the BIS weighed in too — their general manager warned that dollar stablecoins could pose risks to financial stability and urged stronger global coordination on regulation.
Moody's took the opposite view, saying stablecoins aren't a near-term threat to banks, partly because yield-bearing stablecoins face regulatory prohibitions in the US and existing payments infrastructure is robust enough to absorb the competition.
Meanwhile, Paxos Labs raised $12 million led by Blockchain Capital to launch Amplify, a suite that lets platforms convert digital asset holdings into financial products. It has 3 modules: Earn for yield, Borrow for digital asset-backed lending, and Mint for branded stablecoins. One SDK, revenue sharing built in, and Paxos handles the liquidity and counterparty vetting behind the scenes. An early partner called Hyperbeat hit $510,000 in AUM within days of going live.
And Tether backed a UAE tokenization firm called KAIO in an $8 million round to bring institutional Emirati funds on-chain. The pattern is unmistakable: stablecoins are quietly becoming the default settlement layer, not just for crypto, but for real economic activity. When YC is paying founders in USDC and Mastercard is testing stablecoin clearing, the debate about whether this technology matters is over. Now it's just about the plumbing.
Here's what I'd sit with this week. Bitcoin barely flinched through a $292 million DeFi exploit, a geopolitical flare-up in the Strait of Hormuz, and a $13 billion TVL wipeout across decentralized finance. On-chain activity is at an 8-year low, and the price is still near $74,000. The marginal buyer isn't on-chain anymore — it's sitting in a brokerage account buying ETFs and preferred stock. That's a fundamentally different market than even 2 years ago. Whether that's bullish or something to worry about probably depends on why you showed up in the first place.