Bitcoin is under pressure, down near 74,300 after two weeks of heavy ETF outflows totaling 2.26 billion dollars. Michael Saylor floated the idea that Strategy might actually sell some Bitcoin in 2026, which is not a sentence anyone expected to read. The SEC is hitting pause on its tokenized stocks exemption after pushback from traditional exchanges, and Hester Peirce is making clear synthetic tokens are off the table. Meta's Llama 4 saga keeps getting messier, with 600 layoffs in Meta AI and a full restructure under Alexander Wang. And Foundation, the Bitcoin hardware wallet maker, just closed a 6.4 million dollar round to build hardware that keeps AI agents in check. Let's get into it.
Foundation, the Boston-based Bitcoin hardware company, raised 6.4 million dollars in a round led by Fulgur Ventures, with Arche Capital participating. That brings their total funding to 16.5 million. And full disclosure, Fulgur Ventures publishes this podcast, so take that into account, but the product itself is genuinely interesting and worth understanding on its own terms.
The device is called Passport Prime, now generally available at 349 dollars. On paper it's a Bitcoin hardware wallet combined with FIDO keys, 2FA storage, a secrets vault, and 50 gigabytes of encrypted storage. But the framing is what matters. Foundation is calling this Human Authority Hardware, and the pitch is aimed squarely at the coming wave of autonomous AI agents.
Here's the problem they're solving. If you have AI agents executing actions on your behalf, signing transactions, moving money, accessing accounts, you need a trusted approval layer that isn't itself an AI. You need a piece of hardware where a human, in real time, says yes or no to high-stakes actions. Existing hardware wallets weren't designed for that. Enterprise HSMs weren't designed for that. FIDO keys weren't designed for that.
The device runs KeyOS, a Rust-based open-source microkernel with sandboxed apps and hardware-protected master keys. Bluetooth communication uses post-quantum cryptography, which they're calling QuantumLink. And they're opening the platform to outside developers with an SDK, simulator, and a USB MCP server so AI coding agents can be tested against real hardware. A KeyOS app store is planned by end of Q2. Cake Wallet, with over a million users, is the first external partner shipping on it.
The broader bet here is that single-purpose hardware wallets are becoming security stacks. Identity, MFA, agent authorization, Bitcoin custody, all collapsing into one device that you physically tap to approve. If the AI agent thesis plays out, this category is going to matter a lot.
Meta's Llama 4 launch has turned into one of the more dramatic stumbles in open-source AI. The release came back in April 2025 with three variants: Scout, Maverick, and the still-training Behemoth. All mixture-of-experts architecture. On paper, ambitious. In practice, the community testing was brutal. Maverick scored 16% on the Polyglot coding benchmark. There were accusations that benchmarks had been gamed and that test sets may have leaked into training data. Meta denied it and released an experimental open-source version to try to calm the storm, but the damage to credibility was real.
Fast forward to now. Meta AI is going through a massive restructuring under new head Alexander Wang. About 600 roles cut. FAIR, Meta's foundational research arm, folded into Wang's new Super Intelligence Lab. The original Llama team has been sidelined. A new Chief Scientist, Zhao Shengjia, was recruited from OpenAI on what's reported to be a very large package, creating internal tension over salary and status gaps. Prominent FAIR researchers are among those impacted.
The driver behind all this, according to reporting, is panic. Specifically, panic about DeepSeek. The Chinese open-source models have been moving fast, and DeepSeek v4 just dropped two MIT-licensed models, including a 1.6 trillion parameter version that fits on a single NVIDIA HGX B200 node. The headline isn't raw benchmarks, it's serving economics. DeepSeek claims over 10x reduction in cost-to-serve versus v3.2 through compressed sparse attention techniques. If that holds in production, it changes the math on deploying long-context models.
Meanwhile, the developer conversation is shifting. Gemma 4 is getting attention for Apache 2.0 licensing, native multimodality, and edge-to-server deployment with a consistent model family. Llama 4 Scout still has the best context window at 10 million tokens, but lagging coding performance and complicated licensing are hurting adoption.
Meta's bet is that aggressive reorganization and a product-first focus will close the gap. The risk is that gutting your research culture to chase quarterly competitiveness rarely produces the next breakthrough. It produces incremental catch-up.
The SEC just hit pause on what was shaping up to be one of the most consequential rulemakings of the year, the so-called innovation exemption for tokenized stocks. The plan, part of Chair Paul Atkins' Project Crypto agenda, would have created a regulatory path for digital tokens representing publicly traded shares to trade on decentralized platforms 24/7. In some versions of the framework, third-party issuers could wrap stocks like Apple or Nvidia or Tesla without the underlying company's consent.
That last part is where the pushback came from, and it came hard. The World Federation of Exchanges and traditional Wall Street infrastructure firms warned the SEC that allowing third-party wrappers could fragment liquidity across dozens of token issuers, complicate shareholder rights like voting and dividends, and create AML and KYC headaches on pseudonymous chains. Nasdaq, notably, is pursuing its own tokenized securities path that keeps trades on exchange with full shareholder rights, built on DTCC's blockchain. So the incumbents aren't anti-tokenization, they just want it to run through them.
Hester Peirce, the commissioner known as Crypto Mom, stepped in this week to clarify the scope. She pushed back on Bloomberg reporting that had suggested the rule might enable synthetic tokens. Her message: the exemption, when it eventually arrives, will be narrow. It will cover digital representations of existing equities, not synthetic derivative-like instruments tracking stock prices. Volume caps, white-listed participants, and temporary relief are all on the table.
The broader picture is that the SEC under Atkins is still pro-innovation. Safe harbors for developers, fundraising exemptions, investment contract safe harbors, all of that is in play. But the tokenized stock fight shows the limits. When you propose something that genuinely threatens existing exchange revenue, the lobbying machine engages, and the rulemaking slows down.
For the real-world asset tokenization sector, this is a setback but not a death blow. DTCC is still planning limited production trading of tokenized assets by July. ICE is working with OKX on tokenized products. The direction of travel is intact. The timeline just got longer.
Bitcoin is trading around 74,300 dollars, down sharply from the record high of 126,000 set earlier this cycle. That's roughly 38% off the top. Spot Bitcoin ETFs have bled 2.26 billion dollars over the past two weeks. The Dow hit new all-time highs on the same day Bitcoin fell under 77,000, which is the kind of divergence that makes the digital gold narrative harder to defend in real time.
Mark Cuban sold most of his Bitcoin and said it failed as a hedge when fiat confidence weakened and geopolitical risk rose. His words: not the hedge I expected it to be. You can disagree with him, but the price action this cycle has not made the macro hedge case easy.
Then there's Michael Saylor. In an interview this week, he said it's, quote, not unlikely that Strategy will sell Bitcoin in 2026. He framed it around maximizing Bitcoin per share by 2033, which is the corporate finance way of saying they might sell some to buy back stock or restructure if it accretes BTC per share. But the optics are still jarring. Strategy has been the loudest never-sell voice in the market. If they're now publicly modeling scenarios where selling makes sense, that recalibrates how the market reads every other corporate treasury.
On the other side of the ledger, Santiment is calling the ETF outflows a contrarian buy signal, noting these conditions have historically correlated with accumulation rather than panic. The SEC just approved Nasdaq to list Bitcoin index options under the ticker QBTC, pending CFTC sign-off. Tether keeps expanding its Bitcoin holdings. The institutional plumbing is still being built.
So two stories at once. Short-term, ETF flows are negative, sentiment is shaky, and even Saylor is hedging his language. Long-term, the infrastructure deepens every week. Which story you focus on probably says more about your time horizon than about Bitcoin.
One thing to watch as the AI agent wave actually arrives: the question isn't going to be how smart your model is. It's going to be who, or what, has the authority to press the final button on a 50,000 dollar transaction. That answer should probably not be another piece of software.