Startup Thread

All right, everyone. One of the things I’ve been thinking a lot about is this. Why is Bitcoin not going up when all of the precious metals are going up? It feels like those two assets were very tied together. There was a lot of correlation between them. And the fact that Bitcoin is not going up has had me thinking a lot about what’s going on with Bitcoin. What are the market forces. What are all the macro factors that are playing into this. So I spent the weekend taking a deep dive trying to better understand it, and I want to explain what I learned and what my current thoughts are. First, let’s look at the metals themselves. Gold is up about 80% in the last year. Silver is up about 250%. Copper is up about 40%. Platinum is up nearly 200% over the last 12 months. At the same exact time, Bitcoin is down about 16% over the last year. So not only are the metals up, they are up significantly, every single one of them, but Bitcoin is down. It’s down double digits. And that obviously ...

Over the past few weeks, a cluster of unfamiliar terms has been circulating across AI Twitter and developer circles. Names like Ralph , Claudebot , Claude Co-work , and agentic coding have been popping up everywhere, often without much explanation. Taken together, though, they tell a coherent story about how AI-driven coding is evolving early this year. Entrepreneur and creator Riley Brown recently summed up the moment with a single post: “Cool cloud stuff. Remotion skill, Claudebot, CL AWD, Agent SDK, Ralph, and Co-work.” For many people, those words might as well have been in Greek. But the underlying shift is simpler than it sounds. What changed wasn’t the sudden release of a new model. Instead, it was a change in perception. Over the holidays, many developers had time to experiment with tools like Opus 4.5, Claude Code, and CodeX 5.2. As they worked on personal and professional projects, they realized that agentic coding had already progressed much further than expected. That r...

AI is often framed as a software story. Models. Algorithms. Training techniques. That framing breaks down once systems reach scale. At that point, AI becomes an infrastructure problem. And infrastructure runs on energy. Compute is constrained by power, not chips As AI workloads grow, the limiting factor is no longer GPUs alone. It is power availability. Large AI data centers behave like single industrial machines. They draw massive amounts of energy and create sharp spikes in demand. The question is not just where to place servers. It is where sufficient, reliable energy exists to support them. Data centers move toward energy sources One response is to colocate AI infrastructure directly next to energy production. Instead of pulling power through congested grids, companies build near natural gas, turbines, or other generation sources. This reduces transmission constraints and improves reliability. Energy availability begins to determine geography. Batteries smooth volatility...

For years, electric air taxis have existed mostly as demos and renderings. Short test flights. Controlled pilots. Carefully framed timelines. That starts to change in 2026. The shift is not that the technology suddenly appears. It is that certification and regulatory pathways finally line up enough to allow real operations to begin. The first deployments are expected to be limited. A small number of cities. Specific routes. Human pilots in the cockpit. But they are meant to be real, recurring flights, not demonstrations. Certification has been the bottleneck The aircraft themselves are not the main constraint anymore. The challenge has been certification. Air taxis sit in an awkward space between helicopters and airplanes. Regulators had to define new categories, safety requirements, and operational rules. That process has taken time. It is now far enough along that companies expect to begin service before full-scale certification is complete, under restricted operating framewor...

One of the big changes in cybersecurity right now is that attacks are not showing up the way people expect them to. For a long time, security was about malware. You looked for bad code. You looked for signatures. You tried to catch something that clearly did not belong. That model is breaking down. A lot of what is happening now does not involve obvious malware at all. The attacker is already inside. AI changes who can attack AI has lowered the barrier to entry in a way that is hard to overstate. Things that used to require deep technical skill can now be done by people who are not experts. You do not need to write sophisticated code. You can prompt a system to do it for you. That does not mean attacks are new. It means more people can carry them out. You end up in a situation where very capable attacks are no longer limited to a small group of actors. The techniques spread quickly once they work. Attacks don’t need to “phone home” anymore Another change is that attacks no lo...

Stablecoins are often discussed as a crypto-native idea. A new asset class, a new ideology, or a challenge to banks. That framing misses the more practical reason they matter. The simplest way to think about stablecoins is as a payments product. Across fintech, the products that win tend to follow the same pattern. They make an existing financial activity faster, cheaper, or more durable. When that happens, adoption follows. That pattern shows up repeatedly. Square reduced the cost and friction of accepting card payments. Brex made it easier for startups to access credit without pledging personal assets. Robinhood removed trading commissions entirely. Each of these products worked because the economic improvement was obvious. Stablecoins fit into that same category. They allow money to move faster or cheaper than traditional rails. Over time, that tends to win. Payments innovation rarely looks clean From the outside, payments look trivial. A balance moves from one account to anot...

Stablecoins get treated like a crypto market structure story. The more practical way to look at them is as a payments product. If a new rail makes money movement faster or cheaper, it tends to win. If it does not, it stays a niche tool. That is how Zach Abrams, CEO of Bridge, frames the space. His baseline comparison is fintech, not ideology. Square lowered the friction to accept card payments. Brex reduced friction for startups to get a corporate card. Robinhood made trading feel free. These products worked because the value proposition was obvious in dollars and cents. He sees stablecoins the same way. They can move value in a way that is faster and cheaper, and that makes them hard to ignore over time. Payments are “simple” until you look at the rails A common misconception is that payments are trivial. Money moves from one account to another. In practice, payments products are built by stitching together messy systems. Abrams gives a clean example from outside crypto. Early ...

A recurring idea in the discussion is that AI infrastructure is being treated as a winner-take-all market. Hyperscalers themselves describe it that way. Scale matters. Not everyone can win. That framing has consequences once you move away from equity and start thinking about debt. What winner-take-all means for lenders If AI really is winner-take-all, then by definition only a small number of companies will end up on top. That is not a controversial point in the discussion. It is taken as a given. The issue is what happens when investors provide debt financing across multiple players in that environment. From a bond investor’s perspective, lending broadly means you are likely financing companies that will not ultimately be the winners. You may be lending to five or six companies when only one ends up dominating. That is a very different setup than equity investing. Debt does not benefit from upside Equity investors can live with that structure because their payoff is asymmetric...

The debate around AI infrastructure is often framed as a question of scale. How many data centers get built, how fast capacity comes online, and which companies dominate. Less attention is paid to how that infrastructure is being financed, and whether the structure of that financing matches the assets underneath it. Recent discussion on WSJ’s Take on the Week highlights a growing gap between enthusiasm for AI buildout and caution around long-term debt used to fund it. Long duration meets uncertain asset life A key concern raised in the discussion is the growing use of long-term unsecured debt by hyperscalers to finance AI infrastructure. From a fixed-income perspective, this creates tension. The assets being funded are not static. Chips, racks, and supporting hardware evolve quickly. Their useful life remains uncertain, and in many cases relatively short. Issuing debt with maturities stretching decades into the future assumes stability that does not yet exist. Forty years is a l...

The current phase of the AI buildout is being driven less by software breakthroughs and more by capital intensity. Behind the expansion of data centers sits a financing requirement large enough to shape bond markets over the next two years. Estimates discussed on WSJ’s Take on the Week put the amount of debt that needs to be financed at roughly $5.3 trillion over the next 24 months . That figure is tied primarily to data centers and the infrastructure supporting large-scale AI compute. What matters is not just the size of the number, but where the money is expected to come from. Most of the financing still has to reach the bond market Of the estimated $5.3 trillion, about $1.5 trillion is expected to be covered through free cash flow generated by hyperscalers. The remaining roughly $4 trillion will need to be financed externally. That means a large volume of new issuance is expected to reach the bond market in a relatively compressed timeframe. This is not a marginal increase ...

 For a long time, blockchains have been described as “decentralized computers.” It’s an appealing phrase, and on the surface it sounds reasonable. Blockchains run code. They store data. Applications exist on top of them. So why not call them clouds? Because once you look at how these systems actually work, the comparison starts to fall apart. Not gradually. Structurally. Most blockchains were never designed to be general-purpose cloud infrastructure. They were designed to do something much narrower and much more specific: agree on shared state and enforce rules around it. Everything else people try to build on top flows from that choice. That difference matters more than it sounds. Ledgers First, Applications Second At their core, most blockchains behave like distributed ledgers that execute small pieces of logic. Smart contracts are not programs in the traditional sense. They are closer to constrained scripts that update state when certain conditions are met. One way the di...