Monday, April 6th. Here's what matters right now. Nvidia is ditching copper for light to wire up its next generation of GPU supercomputers. Strategy just bought another 4,871 Bitcoin while Japan's Metaplanet quietly became the third-largest public company holder. Bitcoin reclaimed 69,000 dollars on ceasefire talk between the US and Iran, even as oil prices and geopolitical chaos keep traditional markets on edge. And in the world of AI coding tools, Cursor just raised its bet to 2 billion dollars, demoting the IDE itself to a fallback. Let's get into it.
Nvidia is making a fundamental infrastructure bet. The company is shifting from copper interconnects to optical ones, and this isn't a science project. It's a response to a hard physics wall.
Copper has been the workhorse for wiring up GPU clusters. It's cheap, it's reliable, and for short distances, it works great. But as AI training demands scale, you need to connect hundreds and eventually thousands of GPUs into coherent systems. And copper just can't keep up. The bandwidth degrades over distance, the cables get thick and unwieldy, and power losses mount.
Nvidia's plan is to deploy photonic interconnects that can assemble systems with over 1,000 GPUs by 2028. Recent breakthroughs in co-packaged optics, where the optical components sit right next to the chip, are solving earlier problems with power consumption and integration. This is what lets you build AI supercomputers at a scale that copper physically cannot support.
Meanwhile, there's an equally dramatic story on the other side of the Pacific. DeepSeek, China's leading AI lab, has confirmed that its V4 model will run entirely on Huawei chips. No Nvidia, no AMD, no Western silicon at all. This is the AI decoupling we've been talking about for years, and it's now real. China is building a parallel AI stack from silicon to software, driven by necessity as US export controls tighten.
And those controls are biting. Nvidia's stock is down nearly 7% in 2026, hurt by enforcement actions against illegal chip shipments to China and the broader trend of hyperscalers building custom silicon. Nvidia's responding by investing 2 billion dollars in Marvell Technology, opening up its proprietary interconnect fabric to third-party chips. It's a smart hedge.
One more data point worth noting. Memory is projected to consume 30% of hyperscaler AI data center spending this year, up from about 8% in 2023. That's a 4X increase in just a few years. And Nvidia reportedly gets preferential supply terms well below standard market rates, giving it yet another structural advantage. The AI hardware race isn't slowing down. It's just getting more complex.
There's a fascinating split happening among the biggest corporate Bitcoin holders, and it tells you a lot about where we are in this cycle.
On one side, you have the relentless accumulators. Strategy, formerly MicroStrategy, just added another 4,871 Bitcoin for 330 million dollars. Their total holdings now approach 767,000 BTC. The purchases were mostly funded through sales of the company's STRC preferred stock. Michael Saylor posted "back to work" on X on Sunday after a brief pause, signaling the buying machine is running again. And their paper losses are enormous. Strategy reported 14.5 billion dollars in unrealized losses in Q1. But they clearly don't care. The strategy hasn't changed.
Then there's Metaplanet, Japan's largest corporate Bitcoin holder, which bought 5,075 BTC in Q1, bringing its total to 40,177 Bitcoin at a cost of about 3.92 billion dollars. They've now passed MARA to become the third-largest publicly traded Bitcoin holder. Their target is 100,000 BTC by year end and 210,000 by 2027. Aggressive doesn't begin to describe it.
But here's where it gets interesting. MARA, which used to be the second-largest holder, is going the other direction. They sold 15,133 BTC for 1.1 billion dollars and laid off 15% of their staff to pivot toward AI infrastructure. And they're not alone. Riot, Genius Group, and other listed miners disclosed selling more than 19,000 BTC from their treasuries just last week.
So you have this divergence. Some companies are treating Bitcoin as the ultimate reserve asset, doubling down regardless of price. Others are liquidating to fund pivots, mostly into AI. The miners in particular are under pressure. They're approaching what looks like a washout level of selling, which historically has preceded relief in market pressure.
One heartwarming footnote: a solo Bitcoin miner beat 1-in-28,000 odds to mine a block worth 210,000 dollars. The little guy can still win.
Bitcoin reclaimed 69,000 dollars on Monday after reports surfaced that the US and Iran are discussing a 45-day ceasefire. Short liquidations outpaced longs nearly 3-to-1 over 12 hours. It was a textbook short squeeze driven by a geopolitical headline.
But zoom out and the picture is messy. Oil prices have surged roughly 74% since the Iran conflict escalated, with refining capacity in the Gulf region taking real damage. Trump continues to send contradictory signals, threatening that Iran could be "living in Hell" while simultaneously telling reporters a deal is getting close. Polymarket odds of the US invading Iran this year hit 63% after one of his posts.
Traditional markets are struggling. The S&P 500 and Dow declined in April, marking their third consecutive month of losses. Goldman Sachs estimates the energy disruption alone will add 0.2 percentage points to US inflation this year. The US economy contracted in Q1. None of this screams stability.
And yet Bitcoin is finding its footing. There's a growing narrative, backed by actual capital flows, that Bitcoin is functioning as a safe haven. The Financial Times reported that investors are shifting funds into Bitcoin ETFs while pulling out of gold ETFs. That's a meaningful signal. When people reach for Bitcoin over gold during a geopolitical crisis, something has structurally changed in how the asset is perceived.
Now, not everyone is bullish. Bloomberg's Mike McGlone reiterated his long-standing call that Bitcoin could fall to 10,000 dollars unless it reclaims 75,000. And buyers attempting to push above 70,000 are meeting resistance, with some analysts warning a drop below 60,000 is possible. Trading volumes remain thin, and there's no clear consensus on direction.
The dollar, for its part, is strengthening. The US economy's relative resilience compared to Europe, which depends more heavily on energy imports, is supporting the greenback. That's typically a headwind for Bitcoin, so the fact that BTC is holding up despite dollar strength is notable.
One interesting meta-development: prediction markets are becoming real-time macro tools. Sygnum Bank noted that crypto trading desks are now using Polymarket's Iran war odds as a live sentiment indicator, blending geopolitics and DeFi in ways nobody expected a few years ago.
90% of developers now use AI tools at work. That stat alone would have been unthinkable 3 years ago. But the more interesting question is what's happening at the frontier.
Cursor just shipped version 3, and the headline is provocative. The IDE, the integrated development environment that has been the center of a developer's workflow for decades, is now a fallback. Not the default. The default is a chatbot interface where you describe what you want in natural language, and the system writes the code.
Cursor 3 runs multiple AI agents in parallel. Some are cloud-based with access to heavy compute for complex tasks. Others run locally for quick edits and testing. There's a new Design Mode that lets you modify UI elements by just describing what you want changed. And a multi-LLM request feature that fires your prompt at several models simultaneously and lets you pick the best response. The company has raised over 3 billion dollars from Nvidia, Google, and others. They're not playing around.
On the open-source side, the gap is narrowing but it's still real. An XDA Developers comparison tested Claude Code against 3 open-source alternatives: OpenCode, Cline, and Aider. OpenCode came closest to matching Claude Code's ability to handle complex end-to-end tasks, and it supports local models, which matters for privacy and cost. Cline integrates tightly with VS Code, and Aider is the terminal purist's choice with strong Git integration. None of them quite match the polish of the proprietary tools, but they're getting there.
A new academic study looked at over 110,000 pull requests to understand how autonomous AI agents actually perform in the wild. The finding that stands out: AI-generated code experiences more churn over time than human-authored code. It gets merged, but then it gets modified more frequently afterward. Creating code is one thing. Maintaining it is another, and that's where AI agents still have a gap.
The implication is clear. We're moving toward a world where most initial code is AI-generated, but human judgment remains essential for architecture, maintenance, and knowing when the AI got it wrong. The tools are getting extraordinary. The need for good engineers isn't going away. It's just changing shape.
Here's the thing to sit with. In hardware, in treasury strategy, in geopolitics, and in developer tools, we're watching the same pattern. The old defaults are being replaced, not all at once, but decisively. Copper gives way to light. Gold gives way to Bitcoin. The IDE gives way to a chatbot. The question isn't whether these shifts are happening. It's whether you're positioned for where they land. That's your Monday. I'll see you tomorrow.