flyintiger_36 🥷

A follow-up to last Friday's post. Last weekend, a few friends came back to me with similar questions, mostly boiling down to this: Why would Washington keep leaning on the banks to fund this buildout, with long yields at their highest since the early 2000s and AI bubble talk everywhere? I suspect some of you might be wondering the same thing. The short answer is that there's a much larger game at play here. Let me explain… The game is called the Red Queen. In Lewis Carroll's Through the Looking-Glass, the Red Queen tells Alice that in her world you have to run as fast as you can just to stay in the same place. Evolutionary biologists borrowed the idea to describe species locked in an arms race, where nobody gets ahead for long, but whoever stops adapting doesn't survive. What sent me down this rabbit hole was a paper published in January by two researchers at the Chinese University of Hong Kong and the University of Macau, called The Economics of Digital Intelligence Capital. Their insight was that AI capital doesn't lose value the same way a factory or a truck does. It loses value every time a competitor pushes the frontier forward. Stop spending for a quarter and everything you've already built is suddenly worth less, just because the other guy didn't stop. Think of it as two Formula 1 teams. Every race weekend, each team brings new upgrades that make the car a little faster. As long as both keep developing, the race stays close. But if one decides the upgrades are too expensive and skips a few, the gap opens... and skip enough and it's a gap that's almost impossible to close, because the other team is compounding the advantage every single race. That's Microsoft and Google, OpenAI and Anthropic. Now zoom out and swap the two teams for the US and China. The US was happy to let China become the world's largest factory, because cheap goods made the US consumer economy work. Letting China become the world's source of intelligence is a completely different proposition, because whoever leads there leads everything downstream of it: productivity, military capability and eventually financial power. Every great power rivalry before this one was fought over land, oil or arsenals. This time the prize is intelligence itself. That's why the leverage rules were rewritten and why the banks are being positioned to finance the buildout alongside the hyperscalers' own cash flow. And it's why nobody in Washington is going to tighten the taps significantly because asset prices look stretched, not while Beijing is backing its own race just as hard. When both sides are funding a race neither can quit, liquidity stops being a policy choice and becomes a strategic necessity. Think of it as the defense budget of the 21st century. Win the intelligence race and national security comes with it. The pushback I hear most often is that every great infrastructure boom ends in a bust, usually with the railroads and the dot-com fiber buildout as exhibit A. I wrote a much longer piece on exactly this back in May, but here's the short version... Look at why those busts happened. Both built capacity for demand that didn't exist yet. The track was laid long before there was enough traffic to pay for it, and the fiber went in when only a few hundred million people were online, mostly on dial-up. When demand arrived more slowly than the debt came due, the bet broke. AI is the inverse. The distribution already exists, with more than 5 billion people carrying a connected device that can reach these models the day they go live. The demand is already here too, with more than $2 trillion of contracted backlog across the big cloud providers and capacity expected to stay tight through 2027. And that's before the largest wave of immigration in all recorded history, except none of the new arrivals are human. As agents come online, the biggest users of compute won't even be people. They'll be software working on our behalf, consuming compute long after we've put our phones down at night. This isn't capacity in search of customers. It's customers waiting on capacity. There will be a speculative bubble eventually, there always is... but it comes at the end of the cycle, not here. When people ask what slows this race down, my answer is nothing. That doesn't mean there are no speed bumps. Higher rates make it more expensive. Power shortages delay projects. Financing problems will take out individual players... but even then, the compute doesn't disappear. If one of the big names went under tomorrow, every rival would be fighting over its data centers, because that's capacity you simply can't buy anywhere else. None of it removes the incentive for everyone else to keep going. Eventually, I think the two sides reach some kind of truce over compute, because their supply chains are still too intertwined: the US designs most of the leading chips, Taiwan makes them and China processes most of the critical minerals that the whole buildout depends on. Each side holds something the other can't do without. But not yet. Right now, neither side is anywhere near ready to back down. Just look at Tuesday's White House meeting. Dario Amodei, whose company has been one of the loudest voices calling for the industry to slow down, stood next to Trump and told reporters: "As the president said, whoever wins AI wins." His only caveat was that the US should win safely. Even the people who want to slow AI down aren't prepared to lose the race. So back to Gundlach's dilemma. Hike or cut is the wrong question, because it assumes someone gets to ease off. Nobody does. Washington, Beijing, the hyperscalers, the banks... every major player in this race has the same job right now. Keep upgrading the car.