Bitcoin dipped below 75,000 dollars this weekend before bouncing on a Trump peace announcement around Iran. Nvidia delivered another monster quarter and confirmed Vera Rubin ships in the second half of the year. Strategy bought another 2 billion dollars of Bitcoin, pushing its stack past 4% of total supply. And Cursor just landed in the Leaders quadrant at Gartner, sharing the top spot with GitHub Copilot. Let's get into it.
Nvidia's Q1 fiscal 2027 numbers came in Tuesday, and the scale is getting hard to comprehend. Revenue of 81.6 billion dollars, up 85% year over year. Data center alone did 75.2 billion, up 92%. Adjusted earnings of 1.87 per share against expectations of 1.76. And the guide for next quarter is 91 billion at the midpoint, well above the 87 billion the street was modeling.
The headline beyond the numbers is Vera Rubin. Jensen confirmed first shipments in Q3 of this calendar year, volume ramp in Q4, with the real growth pouring into the first half of 2027. Rubin isn't really a chip anymore. It's a rack. The NVL72 configuration has 72 Rubin GPUs paired with 36 Vera CPUs, 1.3 million components per rack, and Nvidia claims roughly 10 times the performance per watt versus the prior generation. Each one of those racks needs over 1,000 terabytes of NAND, which is going to ripple through the memory market.
The new piece here is Vera, the CPU. Nvidia is now telling investors CPUs will generate around 20 billion dollars in annual sales and that Vera opens up a 200 billion dollar addressable market they weren't really competing in before. Early Rubin customers include OpenAI, Oracle, Anthropic, and SpaceX. Nebius is deploying NVL72 systems across the US and Europe starting in the second half.
Three things worth flagging. First, Jensen openly admitted supply constraints on Vera Rubin could persist through the product's life cycle. That's not a humble brag, that's a real bottleneck. Second, hyperscalers were over half of data center sales, more than 38 billion in one quarter. The concentration risk is real even if the demand is real. Third, the 80 billion dollar buyback announcement, on top of 30 billion in disclosed cloud computing commitments, tells you Nvidia thinks the cash flow is sustainable enough to return capital aggressively while still funding the next two generations.
The competitive picture isn't getting easier. AMD, Intel, custom silicon from every hyperscaler, Groq, and a wave of inference-focused startups are all chipping at the edges. But the move from selling accelerators to selling pre-integrated AI factories is Nvidia's real moat right now. You don't replace that with a chip. You replace it with a different operating model, and nobody else has one ready.
Bitcoin spent the weekend testing whether 75,000 dollars is a floor or a trapdoor. Price slipped to roughly 74,300 on Saturday, triggering around 917 million dollars in liquidations in 24 hours, with about 371 million of that on BTC longs. It then rebounded to nearly 77,000 after Trump posted that a peace framework with Iran had been largely negotiated.
Zoom out and the picture is uglier than the bounce suggests. Bitcoin is now down roughly 40% from its October 2025 all-time high near 126,000. It's trading below both the 50-day and the 100-day moving averages. Analysts at multiple desks are openly modeling a revisit of 60,000.
The interesting question is why. The standard narrative would say a pro-crypto Fed chair should be bullish. Kevin Warsh is in the chair. And yet Bitcoin is falling. The answer is the bond market. Two-year and ten-year yields are pushing toward 5%. Interest rate swaps are now fully pricing in at least one Fed hike by year-end, not a cut. Fed Governor Waller said this week the Fed should drop its easing bias entirely. Higher-for-longer real yields are the actual macro story, and risk assets are getting squeezed accordingly.
The options market agrees. Glassnode data shows 25-delta skew firmly in put territory after the 82,000 rejection. There's roughly 3.2 billion in negative gamma exposure clustered around 75,000, which means dealer hedging can amplify moves down through that level. One-week implied volatility dropped to around 31%, but the protection bid is persistent.
The constructive counter-argument comes from Mark Connors, former Credit Suisse, who argues Bitcoin just exited its longest stretch of relative underperformance in history, 142 days against stocks, bonds, and gold. His view: if inflation stays sticky and Treasuries keep selling off, hard money assets eventually catch a bid. Spot net flows on exchanges are negative 1.19 billion over 30 days, meaning coins are leaving exchanges into private wallets. That's classic hodling behavior even as price falls.
Near-term, watch 73,600 on the downside and 78,600 on the upside. A daily close above 78,600 on volume would invalidate the bear setup. A break of 73,600 likely opens the door to the low 70s and the 60K scenario everyone is whispering about.
Michael Saylor's company, now just called Strategy, did it again. The 8-K hit Monday morning confirming the purchase of 24,869 Bitcoin for about 2.01 billion dollars at an average price of 80,985 per coin. That brings the total stack to 843,738 BTC, acquired at an average cost of 75,700, for a total spend of 63.87 billion dollars.
That 843,738 number is the one that matters. It's just over 4% of all Bitcoin that will ever exist. One company. One balance sheet. One increasingly leveraged bet.
The funding mechanism is worth paying attention to because it's evolving. This round came almost entirely from STRC, the perpetual preferred stock paying an 11.5% annualized dividend. Strategy sold 19.5 million STRC shares and only 430,000 shares of MSTR common during the week of May 11 to 17. STRC volume hit an all-time high of 15.1 million shares in a single session. Roughly 80% of STRC is held by retail. The company is now seeking shareholder approval to shift the dividend from monthly to bimonthly, which would let them time issuance more tightly around ex-dividend dates.
There's still significant ATM capacity left, 17.5 billion in STRC and 26.3 billion in MSTR common. So the buying program isn't slowing down on paper. But notice the average price of this purchase, just under 81,000, versus a spot price that spent the weekend below 76,000. They're now buying above market, which is fine when STRC is trading at premium and demand is strong, but it's a tighter window than it was when Bitcoin was at 90,000 and they were buying at 70.
The other piece of news: Strategy plans to buy back 1.5 billion of its zero-coupon 2029 convertible notes for about 1.38 billion. They're effectively retiring older, cheaper debt with new preferred equity. It cleans up the capital stack and reduces dilution risk if the converts ever became a problem.
CEO Phong Lee is touting a 6.6 billion dollar BTC Gain year-to-date and a 12.6% BTC Yield for 2026. Those metrics only work if Bitcoin holds or appreciates. At 75,000, the average cost basis of 75,700 is right at the line. Strategy isn't in trouble, the preferred dividend is well covered and there are no near-term forced sellers, but the cushion is thinner than it has been in a while.
Gartner dropped its 2026 Magic Quadrant for Enterprise AI Coding Agents and the results say a lot about where this market actually is. GitHub Copilot is a Leader for the third year running. Cursor is now also a Leader, called out specifically for completeness of vision. That's the headline shift. Two years ago Cursor was a curiosity. Now over 70% of Fortune 500 companies use it.
What's interesting is how differently the two leaders are positioned. GitHub is leaning hard into governance, the full software lifecycle, and the fact that Copilot is now embedded across 140,000 organizations. The pitch is: we own the platform where the code already lives, we integrate with your existing controls, and our agentic workflows handle issues, reviews, pull requests, and actions natively. Copilot CLI usage is nearly doubling month over month. That's a real number.
But GitHub has had a rough stretch. Repeated outages since March have dented the reliability story right when enterprises are deciding which agent to standardize on. Microsoft's broader AI narrative is also a little muddled, balancing OpenAI bets against in-house ambitions, and Copilot's move to usage-based pricing is creating friction with developers who liked the flat-rate predictability.
Cursor is going a different direction. They just shipped Composer 2.5, an in-house mixture-of-experts model built on Moonshot's Kimi K2.5 base, trained on roughly 25 times more synthetic tasks than the previous version. Same pricing as Composer 2, 50 cents per million input tokens, 2.50 per million output. They're being unusually transparent, including publishing failure modes and a candid note about reward hacking they caught during training. They've also got a partnership with what's referred to as SpaceXAI for a larger-scale frontier model. Cursor's bet is that owning the model, the editor, and the agent layer end-to-end gives them better task focus and faster iteration than wrapping somebody else's API.
Meanwhile, OpenAI and Dell announced a Codex enterprise partnership this week. Codex can now run inside Dell's AI Factory infrastructure, much closer to internal codebases, documents, and approval chains. Dell claims 5,000 AI Factory customers already. OpenAI says Codex reaches over 4 million developers weekly. The framing is on-prem and policy-controlled, which is exactly what regulated industries have been asking for and exactly what pure cloud tools couldn't offer.
The pattern across all three stories is the same. Coding agents are no longer about autocomplete. They're about agentic workflows running across the full development lifecycle, with enterprise governance, security controls, and increasingly the option to deploy near the data rather than in someone else's cloud. The market is big enough for multiple winners, but the differentiation is shifting from model quality to integration depth and deployment flexibility.
Here's the trade that's actually defining this moment: Nvidia prints 81 billion in a quarter and announces an 80 billion buyback, while Bitcoin sells off because the bond market won't let the Fed cut. The AI capex cycle and the hard-money thesis are both real, and right now they're pulling in opposite directions. Whichever one bends first sets the tone for the rest of 2026.