Latest / Tech IPO Conversations with Fexingo: Public Software Companies and Capital Markets / How Post-IPO Companies Use AI Compute as a Competitive Moat
Transcript
- Lucas: Let's talk about something that's quietly reshaping how post ipo software companies compete: their ability to throw compute at problems. Not just any compute — ai grade GPU clusters that cost tens of millions to stand up. Luna: You're talking about the moat that public-market scale buys you in the AI arms race. Lucas: Exactly. We've been watching Palantir for a while now, and its five-day gain of almost 15 percent — closing today at 152 dollars and change — isn't just about a good quarter. It's about the market re-rating their compute infrastructure as a durable asset. Luna: So the thesis is: owning your own compute cluster is the new factory floor. Lucas: Precisely. And Palantir is a perfect case study because they've been investing heavily in what they call 'AI platforms' — essentially massive GPU deployments that let them run classified workloads for government clients. That's something a private startup can't easily replicate. Luna: Right, because a startup would have to either rent capacity from a cloud provider at variable prices or raise billions just to buy the hardware. Lucas: And even then, the lead time on NVIDIA's H100 and B200 chips is still measured in months. Public companies that started ordering two years ago are now sitting on a stockpile that's become a competitive moat. Shopify is another example — up 9 percent in the last five days. Luna: Shopify's move has been interesting because they're not just using AI for internal tools; they're offering it to their merchants as a bundled service. Lucas: That's the second layer. Compute isn't just an operational advantage — it's becoming a product feature. Shopify's AI assistant for merchants, which helps with inventory forecasting and customer service, is powered by models running on their own infrastructure. They control the latency, the cost, the fine-tuning. Luna: And that control translates into margin. If you're a merchant paying Shopify for AI features, you're essentially renting access to their compute moat. Lucas: Now contrast that with the private AI companies — and there are dozens of them — that are burning cash on cloud compute leases. Their unit economics are structurally worse because they don't have the balance sheet to pre-purchase capacity or negotiate volume discounts. Luna: That's why we're seeing more late-stage private companies stay private longer. They don't want to expose those fragile margins to public scrutiny. Lucas: And that's a reversal of the old playbook. For years, the advantage of going public was cheap capital for growth. Now, the advantage might be the ability to place massive, long-term bets on physical infrastructure. Luna: But there's a risk here, right? If you build too much compute and the AI demand curve flattens, you're left with stranded assets. Lucas: It's a real concern. Microsoft just announced a new tool that lets developers spin up AI behavior tests using text descriptions — that's a software play, not a hardware one. It suggests they're trying to make AI development more efficient, which could reduce the raw compute needed per application over time. Luna: So the moat might have a shelf life. If the cost of training and inference keeps dropping, owning your own cluster becomes less critical. Lucas: That's the bear case. But for now, the market is rewarding companies that show they can translate compute into revenue. Let me give you a concrete number: Palantir's government segment grew at a compound annual growth rate of about 30 percent over the last three years — and their ability to run classified workloads on their own hardware is a big reason why. Luna: So the question for investors becomes: which post ipo companies are building real compute capacity versus just renting it and calling it AI strategy? Lucas: That's exactly the kind of scrutiny we're seeing now. And it's one reason why some of the big names like Alphabet have seen weakness — down 6.3 percent in the last five days — because the market is questioning whether their massive compute spending is translating into defensible revenue. Luna: Meanwhile, a company like Rivian, up almost 18 percent this week, doesn't even compete on AI compute — they're in a completely different sector, so it's not a factor for them. Lucas: Right. This is very sector-specific. But for software companies that pitch themselves as ai native, the compute question is becoming the first due diligence item. Luna: If conversations like this are useful for what you're building or running, it's worth knowing that these episodes stay ad-free thanks to listener support. People who find value here sometimes toss a few dollars at buy me a coffee dot com slash fexingo. Lucas: It genuinely helps keep the show independent. And it means we can keep digging into specific angles like this one without chasing commercial sponsors. Luna: Yeah, we appreciate it. Now back to the moat — Lucas, what about the risk that AI compute becomes commoditized? We're already seeing cloud providers offer GPU instances at increasingly competitive prices. Lucas: That's a real threat to the moat thesis. If you can get equivalent compute from AWS or Azure at a price that makes building your own cluster uneconomic, then the advantage disappears. But right now, the bottleneck is still availability. Luna: And that's why the big public companies are leaning in — they have the cash and the credit to secure supply. Lucas: Exactly. Think about it: a company like Shopify can issue debt at 4 percent and use that to pre-pay for three years of GPU capacity. A private startup would pay venture debt at 12 percent or give up equity. The cost of capital advantage is real. Luna: So we're seeing a bifurcation: public companies with AI compute moats are getting premium valuations, while those that rely on rented capacity are being discounted. Lucas: That's the new reality. And it's going to force more private AI companies to either go public sooner than they'd like or find creative ways to pool compute resources. Luna: Like compute cooperatives or shared infrastructure consortia? Lucas: Potentially. There are already startups trying to broker bulk GPU deals for mid-size firms. But it's early days. For now, the clear winners are the post ipo software companies that made big compute bets two to three years ago. Luna: And the losers are the ones that waited. Lucas: That's the punchline. And it's a reminder that in tech, the moat you build today is the advantage you collect tomorrow — or the anchor you drag if you get it wrong. Luna: Alright, that's a good place to leave it. Thanks for joining us on Tech IPO Conversations.