Heavy news day. xAI's books got cracked open in SpaceX's IPO filing, and the numbers are wild — 6.4 billion in operating losses on 3.2 billion in revenue. Figure AI's humanoid robots just ran an autonomous warehouse shift and nearly beat a human in a 10-hour package sorting contest. SpaceX disclosed it's sitting on 18,712 Bitcoin worth 1.29 billion dollars. Bitcoin itself is stuck under 80,000, with traders bracing for a 6.25 billion dollar options expiry next week. And the CLARITY Act is finally getting a coordinated White House push in the Senate. Let's get into it.
SpaceX's IPO filing dropped this week and it's the first real look inside xAI's finances since the merger. The headline number: 6.4 billion dollars in operating losses last year on just 3.2 billion in revenue. For context, the year before it was 1.56 billion in losses on 2.62 billion in revenue. So losses quadrupled while revenue barely doubled. That's not a company finding leverage. That's a company stomping the accelerator.
The revenue mix is interesting. About 465 million came from AI solutions and infrastructure — 365 million of that from Grok and X subscriptions, 88 million from data licensing. Advertising added another 116 million. So Grok subscriptions are actually doing real numbers, but nowhere near enough to cover what's being spent on compute.
And the compute spend is the story. Capex hit 12.7 billion last year. The annualized run rate for the first quarter of 2026 is around 30.8 billion. That's Meta-tier infrastructure spending from a company that didn't exist three years ago. They're justifying it with vertical integration — owning Colossus and Colossus II data centers, roughly 1 gigawatt of compute, and an ambition to scale Grok to, quote, multiple trillions of parameters. There's also a line in the filing about orbital compute satellites starting as early as 2028, which is either visionary or a flex for IPO investors. Probably both.
On the product side, xAI launched Grok Build this week — their first real coding agent, aimed squarely at Claude Code. It's in beta, locked behind the 300-dollar-a-month SuperGrok Heavy tier. They also rolled out Grok Skills, which lets users save reusable workflows — basically turning the chatbot into a programmable workspace. Grok 4 is out there benchmarking above Gemini 2.5 Pro and OpenAI's o3 on Humanity's Last Exam and Arc-AGI-2. The tech is real.
But here's the tension. Grok has 117 million monthly active users out of 550 million across the X ecosystem. That's 20% penetration of a captive audience. Meanwhile xAI reportedly lost 50-plus researchers and engineers post-merger. So you have world-class infrastructure, a price-competitive top-tier model, a built-in distribution channel — and a company hemorrhaging cash and talent simultaneously. The IPO is going to test whether investors believe owning the full stack actually wins, or whether this is just the most expensive way to come in third behind OpenAI and Anthropic.
Figure AI did something this week that nobody else in humanoid robotics has done. They livestreamed their Figure 03 robots running an actual warehouse shift — package sorting, barcode scanning, conveyor belt loading — for over 48 hours straight with no logged failure. No cuts, no edits. Just robots doing the job, on camera, for the internet to pick apart.
The most telling part was the head-to-head. On May 17, they put a human intern named Aimé Gérard against the robot fleet in a 10-hour contest. Final score: human 12,924 packages, robots 12,732. The human won — but barely, and mostly because he didn't need a break. Per-package speeds were essentially identical. The intern averaged 2.79 seconds per package, the robots 2.83. We are talking about a four-hundredths-of-a-second gap between a trained human and an autonomous machine that can run 24 hours a day and doesn't ask for a raise.
The system underneath is Helix-02 — a unified neural network handling vision, touch, and full-body movement, trained on human motion data and large-scale simulation. The robots can self-diagnose and request backup from each other. That fleet behavior is arguably more impressive than the sorting itself.
Meanwhile, Tesla released video of an Optimus pilot production line, which is the first serious signal that they're moving from prototype to actual manufacturing. And on the deployment side, Memeburn pulled together the landscape: Humanoid and Schaeffler are fielding 1,000 to 2,000 wheeled humanoids globally by 2032. BMW is expanding its Figure deployment from Spartanburg into Europe. JAL is running Japan's first humanoid trial at Haneda airport for ground handling.
The cost curve is what matters. Western humanoids are targeting under 17,000 dollars per unit by 2030, with operating costs around 2 dollars an hour. At that price, in a warehouse running multiple shifts, the math gets brutal for human labor very quickly. UBS is projecting 2 million humanoids in workplaces by 2035. That number might be conservative. The bottleneck right now isn't capability — Figure just proved that. It's safety certification for fenceless operation. Once that clears, this scales fast.
Sovereign Bitcoin is not one story anymore. It's at least three.
First, El Salvador. Their strategic reserve crossed 600 million dollars this week — 7,652 BTC as of mid-May. Bukele is still buying, still on the same dollar-cost-average cadence, and still doing it in apparent defiance of IMF loan conditions that supposedly capped state Bitcoin holdings. The IMF claims earlier increases were just internal Chivo wallet transfers. El Salvador's officials say no, we're actually buying. The country now ranks fifth globally in sovereign holdings, and Bitcoin makes up about 12% of its net international reserves. The on-the-ground adoption story remains weak — 92% of Salvadorans didn't use Bitcoin last year — but the treasury thesis keeps compounding.
Then there's Bhutan, going the other direction hard. Once holding around 13,000 BTC from hydro-powered mining, they're now down to somewhere between 3,100 and 3,400 coins. That's a 75% drawdown. On-chain analysts have tracked roughly a billion dollars in BTC leaving Bhutan-linked wallets since mid-2025. Bhutanese officials say they don't recall any sales, which is a phrase that should worry anybody. If trends continue, Bhutan could fully exit by October.
And then there's the United States. Patrick Witt at the White House said this week that the legal and custody framework for the Strategic Bitcoin Reserve is essentially done. The government is sitting on roughly 328,000 BTC worth about 25 billion dollars, and an official announcement on the framework is expected in coming weeks. The BITCOIN Act is moving through Congress to codify it past the next administration. The interesting tell: SpaceX's IPO filing revealed it holds 18,712 Bitcoin at 1.29 billion dollars in fair value. That puts SpaceX seventh among public companies. Nation states, public companies, and the U.S. government all stacking — while Bhutan quietly liquidates.
Market-wise, Bitcoin's having a rough patch through all this. It's stuck below 80,000, ETF flows have weakened, and CME FedWatch now shows a 54% chance of a rate hike in December — up from expectations of cuts just weeks ago. Long-term holder supply hit 16.3 million coins, near record highs, breaking a multi-year downtrend. The conviction is there. The price action just hasn't caught up.
The regulatory machinery is finally moving, and it's moving with intent. The CLARITY Act, which passed the House back in July 2025, is now getting a coordinated White House blitz in the Senate. Treasury, the SEC, and the CFTC are simultaneously publishing reports, op-eds, and proposed rules aimed at the Senate Banking Committee to break a stalled markup. The motivation is the 2026 midterms. If this doesn't pass before election season eats all the legislative oxygen, it sits in limbo.
What the bill does is create a pathway for digital assets to migrate from SEC jurisdiction as securities to CFTC jurisdiction as digital commodities, once they're decentralized enough. That's the structural reform the industry has been chasing for years. The catch — and CryptoSlate flagged this well — is that the CFTC's payroll workforce has dropped more than 30% in recent years. Handing them spot market oversight without resourcing them is going to be its own crisis.
On stablecoins, Bernstein put out a note arguing Circle has the cards if CLARITY passes, because the compromise language blocks a yield arms race between issuers. Total dollar-backed stablecoin supply just crossed 300 billion dollars. USDT and USDC together control 97% of the market. Adjusted monthly transaction volume is running around 15 trillion dollars. Annualized, that's approaching 100 trillion, up from 55 trillion last year. USDC's share of adjusted volume jumped from 41% to 60% year-over-year. These are payment-rail numbers, not speculation numbers.
Meanwhile the NCUA is out with proposed rules under the GENIUS Act for credit unions issuing payment stablecoins. The Fed proposed limited master accounts — the so-called skinny accounts — for fintech and crypto firms, after Trump's executive order directing them to study direct access to the payment system. Coinbase just expanded its white-label stablecoin infrastructure with Flipcash USDF on Solana. And in a notable corporate move, Tether bought SoftBank's 26% stake in Twenty One Capital, the Bitcoin treasury company, giving Tether significant control of a public BTC holder.
The direction is unmistakable. Stablecoins are stopping being a crypto product and becoming dollar settlement infrastructure. Whoever owns the rails when the regulatory framework hardens captures an enormous amount of global payment flow. That's the actual game being played in the Senate right now.
One observation to close on. A human warehouse worker just barely beat a robot at sorting packages. An IPO filing just revealed a 30 billion dollar annual compute run rate at a company that lost 6.4 billion last year. And the U.S. government is finalizing custody for 25 billion dollars in seized Bitcoin while Bhutan quietly sells its stash. The cost of compute, the cost of labor, and the cost of holding hard money are all being repriced at the same time. Whichever of those moves first dictates everything else.