Latest / Tech IPO Conversations with Fexingo: Public Software Companies and Capital Markets / Why Tech IPOs Are Hiding Their World Models
Transcript
- Lucas: The most interesting shift in the tech IPO landscape right now isn't about how much money they raise, but what they refuse to tell you once they're listed. We are seeing a distinct move away from the hyper-transparency that defined the early cloud era, with companies like Roblox actively choosing to obscure their underlying AI architectures. Luna: Right, and it feels less like a compliance issue and more like a strategic defense mechanism. If the algorithm is the product, why give competitors a blueprint? Lucas: Exactly. Look at Roblox, which trades around forty-six dollars on the NASDAQ today. They have built a massive ecosystem where generative AI tools help creators build worlds, but they are notably vague about the specific models powering those features. In previous years, an IPO prospectus would list every vendor and technical dependency; now, it’s often just 'proprietary systems' and 'third-party partnerships.' Luna: So the moat isn't just the user base anymore, it's the opacity of the code itself. Does that make valuation harder for us retail-style investors, or does it just shift the risk? Lucas: It shifts the risk significantly toward due diligence. When you buy into a company like Roblox, you aren't just buying engagement metrics; you're betting on the efficiency of their inference costs. But without knowing whether they are running custom chips or relying on expensive third-party cloud instances, you can't accurately model their margins. It creates a black box that makes comparing them to pure-play AI infrastructure stocks nearly impossible. Luna: That’s a fair point. If I can’t see the engine, I can only judge the car’s speed. And right now, the car seems fast but the fuel mix is unknown. Lucas: Precisely. And this is happening across the board. Even established players are tightening up. Take Meta, which has been open about its Llama models for years, yet even they are becoming more circumspect about the energy consumption and compute requirements of their newest iterations. The narrative has moved from 'we are democratizing AI' to 'our efficiency is our secret sauce.' Luna: I’ve noticed that too. It used to be a badge of honor to say you were using the latest open-source weights. Now, silence is the new prestige. Lucas: It’s not just prestige; it’s protection against regulatory scrutiny and competitor poaching. If you disclose your exact prompt engineering strategies or your data filtering pipelines, a well-funded rival can replicate it in months. By keeping it vague, you force the market to price you based on outcomes—revenue, active users—rather than inputs. And right now, the market loves outcomes. Luna: But doesn’t that create a trust deficit? Investors feel like they’re being sold a magic wand rather than a machine. Lucas: They do, but the magic works until it doesn’t. The danger comes when the economics break down. If a company claims high margins but is actually burning through cash on invisible GPU clusters, the reveal can be brutal. We saw hints of this with some of the smaller ai native IPOs last year that couldn’t justify their valuations because their cost per inference was ten times higher than analysts assumed. Luna: So the opacity protects the stock price in the short term, but it amplifies the crash if the unit economics don’t hold up. Is there any company doing this better than others? Lucas: NVIDIA handles it differently because they sell the shovel, not the map. But for application-layer companies, I think we’ll see a bifurcation. One group will stay opaque and hope growth masks the details, while another will try to build trust by releasing standardized benchmarks, even if they hide the raw weights. Luna: Standardized benchmarks sound like a good middle ground. Like saying 'my model scores 95 percent on this test' without showing the training data. Lucas: Exactly. It’s an industry standard that hasn’t fully taken hold yet. Until then, we’re left guessing. And that brings me to something else I wanted to mention quickly before we dive deeper into the implications. Luna: Sure, what’s on your mind? Lucas: We deliberately don't run ads on these shows. If you want to support that choice, the link is buy me a coffee dot com slash fexingo. Luna: Yeah, it really helps keep the research independent. Thanks for listening. Lucas: Back to the opacity issue. Let’s talk about the regulatory angle. The SEC has been clear that material information must be disclosed, but 'how' AI works is increasingly seen as trade secrets rather than material financial risks. Where do you draw the line between a secret recipe and a hidden liability? Luna: That’s the million-dollar question. If the AI starts generating illegal content or has a bias bug, that’s a liability. But if the company hides the code, how do we know the fix is coming? Lucas: It’s a genuine governance gap. In traditional manufacturing, you can audit the factory floor. In software, the factory is a server farm in a different country, running code that changes every hour. Without transparency, shareholders are essentially trusting management’s word on safety and ethics. And historically, that trust has been tested. Luna: So maybe the next wave of IPO disclosures will include third-party audits of the AI systems themselves? Like a health inspection for the algorithm. Lucas: I think that’s likely within the next three to five years. We’re already seeing insurance companies start to ask for these audits. If insurers won’t cover your AI product unless you prove it’s safe, then you have no choice but to show your work. It’s market forces, not regulators, driving this change. Luna: That makes sense. Insurance is a great proxy for risk. If they won’t touch it, neither should the public markets. Lucas: Right. And look at the broader market context today. With the S&P 500 hovering near recent highs and tech volatility creeping back in, investors are getting pickier. They don’t want magic wands; they want predictable margins. A company that can say 'here is exactly how much it costs to serve one user' wins the premium multiple over the one that says 'trust us, it’s efficient.' Luna: So predictability is the new sexy. Not just growth, but explainable growth. Lucas: Explainable growth is the only kind that survives a bear market. In a bull market, you can get away with mystery. In a bear market, mystery looks like fraud. And with rates still sticky in late 2026, we aren’t in a bull market mood for speculation. Luna: I remember that feeling from a few years ago. The ones who survived were the ones who could show their math. Lucas: Exactly. So when you’re looking at a new tech IPO, don’t just ask what their AI does. Ask how they account for it. If they can’t give you a straight answer on compute costs, that’s your first red flag. Luna: And if they do give you a straight answer, check if it matches their revenue growth. Because sometimes the math gets stretched thin. Lucas: That’s a great way to put it. Stretching the math usually ends in tears. But for now, the trend of hiding the world models is going to continue. Companies will protect their IP aggressively. Luna: Until the regulators or the insurers force them to open the box. It’s only a matter of time. Lucas: Probably sooner than later. But in the meantime, we’re left playing detective. Which is fine by me, as long as the clues are real. Luna: Real clues, not magic spells. I’ll take that any day. Lucas: Agreed. Let’s see if any of the upcoming earnings calls this week finally crack that nut. Or if everyone stays beautifully, frustratingly silent. Luna: Silent it is, probably. But we’ll be watching.