The 900 Billion Dollar Question Nobody Is Asking About AI

0:00Lucas: So Broadcom is down seventeen and a half percent in a week. Micron, nearly eleven percent. The SOXX semiconductor index shed five and a half percent over the same stretch. And here's the weird part — none of these companies reported bad news. Luna: Right — it's not earnings. It's not guidance. So what spooked the market? Lucas: I think the market is starting to ask a question that nobody wanted to ask during the AI gold rush. And that question is: Does any of this spending actually generate a return on invested capital? Luna: That's one of those questions that sounds simple but gets really uncomfortable really fast. Lucas: It does. And look — I want to be clear: I'm not saying AI is a bubble or that the spending is wasteful. But when you look at the numbers, there's a gap between what companies are spending on AI infrastructure and what they're earning from it. And the gap is getting harder to ignore. Luna: If today's conversation gave you something to think about — and honestly, if it was worth a coffee to you — there's a link at buy me a coffee dot com slash fexingo. No pressure, just a way to keep the show ad-free and independent. Lucas: Yeah, that's the only ask we make. And it genuinely helps. So back to that uncomfortable question — because it's the one every CFO is wrestling with right now. Lucas: Let me put some numbers on this. According to recent estimates, the top four cloud providers — Amazon, Microsoft, Google, and Oracle — are on track to spend over 200 billion dollars on ai related capex this year alone. That's up from maybe 120 billion last year. Luna: And what's the revenue from those AI services? I've seen numbers around 30 to 40 billion. Maybe. Lucas: Maybe is the right word. So you've got a ratio of roughly five to one — five dollars spent for every one dollar of revenue. Now, some of that spending builds long-term assets, sure. But the market is looking at that gap and starting to sweat. Luna: And that's why a company like Broadcom gets crushed even when they beat earnings. Investors are pricing in the risk that the spending wave slows down. Lucas: Exactly. Broadcom's networking chips are essentially the plumbing for AI data centers. If the plumbing contractor sees a slowdown in new construction, they feel it fast. And Broadcom's stock is down almost eighteen percent in five days — that's not a reaction to a single data point. That's a sentiment shift. Luna: Let me play devil's advocate for a second though. Historically, every major technology buildout has had this phase. The railroad bubble in the 1800s, the dot-com buildout — overinvestment early, then consolidation, then the real value emerges. Isn't that just how infrastructure works? Lucas: It's a fair argument. And I think there's truth to it. But the scale here is different. The amount of capital being deployed is an order of magnitude larger than any previous tech cycle. And the timeline to see returns is longer because AI is still figuring out its killer apps. Luna: So the market is saying 'show me the money' — and the AI companies are saying 'trust the process'. Lucas: Basically. And investors are starting to demand more concrete evidence that these investments will pay off. That's why you see stocks like NVIDIA and AMD also taking a hit — down six percent each over the past week. Even the picks and shovels suppliers are getting lumped into the selloff. Luna: But isn't the spending still happening? I mean, the hyperscalers aren't slowing down their buildouts just because the stock price wobbled. Lucas: No, they're not. And that's the disconnect. The actual capital deployment continues. The orders for chips and data center equipment are still flowing. But the stock market is forward-looking. It's pricing in the possibility that those orders start to slow six to twelve months from now, especially if the ROI picture doesn't improve. Luna: So who's most exposed if the slowdown actually happens? Lucas: The companies with the least diversification. Broadcom is heavily tied to AI networking. Micron is tied to high-bandwidth memory for AI accelerators. Both are down big this week. On the other hand, Intel is actually up two percent — they have a broader portfolio, including PCs and traditional servers, so they're less exposed to a pure AI pullback. Luna: What about the software layer? Like Microsoft or Meta — they're spending billions too, but they also have revenue from AI features. Lucas: Microsoft is down almost seven percent this week. That tells you the market is questioning their AI monetization too. Meta is down only two percent — relatively resilient — probably because their AI investments are tied to ad revenue, which is more measurable. But the overall story is the same: the market wants proof. Luna: I think there's another layer here that doesn't get enough attention. The return on invested capital isn't just about revenue. It's about whether the capital is being used efficiently. And there's a growing sense that a lot of this spending is duplicative. Lucas: That's a great point. You have three or four hyperscalers all building essentially the same infrastructure. And each one is trying to lock in supply agreements with NVIDIA, AMD, and others. That competition drives up prices and reduces returns for everyone. It's a classic coordination problem. Luna: So what would change the narrative? What would make the market feel better? Lucas: Two things. First, a clear signal that AI services are generating real, recurring revenue at scale — not just experimental usage. Second, some consolidation or specialization among the infrastructure players so that capital isn't being wasted on redundant buildouts. If we see one of the hyperscalers announce a partnership or a co-investment model, that could be a positive sign. Luna: Like a shared data center consortium or something? Lucas: Exactly. Or more enterprise customers publicly committing to long-term AI workloads. Right now, a lot of the demand is still from the tech giants themselves. If you see a major bank or a manufacturer saying they're spending a billion dollars on AI compute over five years, that changes the math. Luna: So the market isn't saying AI is dead. It's saying 'prove it'. Lucas: That's exactly where we are. And I think that's healthy. A little skepticism forces discipline. The question is whether the discipline arrives before the investment wave crests. Luna: And between the SOXX down five and a half percent and Broadcom down eighteen, it feels like the market is already voting. Lucas: It is. But remember — markets overreact in both directions. The same stocks that are getting hammered today could be the ones that look cheap in six months if the spending cycle extends. The key is watching the ROIC trajectory. If it starts to inflect positively, that's when you want to be buying. Luna: So the bottom line: AI infrastructure spending is real, but the returns are still unproven. And until that changes, expect more volatility. Lucas: Exactly. And that's where we are on June 9th, 2026. The 900 billion dollar question — because that's roughly the cumulative AI capex projected over the next three years — is whether the returns will eventually justify the outlay. History says yes, eventually. But eventually can be a long time when you're holding the stock.