Latest / Tech IPO Conversations with Fexingo: Public Software Companies and Capital Markets / How IPO Proceeds Are Funding AI Compute Instead of Growth
Transcript
- Lucas: If these conversations are useful for what you're building or running, we want to keep them coming without any ads. That only works if people who find value here chip in a little. Luna: Absolutely. A couple of dollars a month is genuinely what keeps these going — buy me a coffee dot com slash fexingo, if you've gotten something out of them. It's that simple. Lucas: Exactly. And on that note — let's talk about something I've been tracking in the latest S-1 filings. There's a quiet shift in what companies say they're going to do with the money they raise. Luna: I've noticed that too. It used to be 'sales and marketing' or 'general corporate purposes.' Now it's 'infrastructure and compute.' Lucas: Right. I pulled the data for Q1 2026 — every tech company that filed an S-1 in that period. The average allocation to compute infrastructure hit 38 percent. Two years ago it was 6 percent. That's not marginal. That's a structural redefinition of what IPO capital is for. Luna: So instead of hiring a hundred sales reps, they're signing hundred million dollar cloud contracts before they even set a price range. That's a bet — and it changes how you model the business. Lucas: Let me give you a concrete example. There's a data infrastructure company — they filed in March, planning to raise about $350 million. Their use of proceeds section dedicated $120 million to AWS reserved instances. That's more than a third of the entire raise. Luna: And that's a fixed cost. It's not like headcount you can trim in a downturn. You've committed to that compute. Lucas: Exactly. And banks underwriting the deal have to treat it as a liability. When I talked to a syndicate desk last week, they said they're now modeling cloud contracts as debt-like obligations. That's new. Luna: It also pushes out profitability. If you're spending 38 percent of your IPO cash on GPUs before you've even scaled revenue, your path to positive earnings per share gets longer. The market has started penalizing that. Lucas: We saw it with a couple of late-April IPOs. Companies that guided for breakeven in 18 months got hit harder than ones that said 24 months but had lower compute commitments. Investors are doing the math: capex-heavy companies need higher revenue growth just to cover the fixed costs. Luna: Is this concentrated in ai native companies, or is it spreading to broader software? Lucas: It's most pronounced in AI, but I'm seeing it in data infrastructure, cybersecurity — any company that needs to train or inference at scale. The old distinction between 'tech company' and 'compute company' is blurring. Luna: So what does this mean for the IPO window? We've been talking about it closing for unprofitable companies. Does this accelerate that? Lucas: I think it creates a two-tier market. One tier — companies with strong gross margins and moderate compute needs — they can IPO on traditional metrics. The other tier — heavy infrastructure spenders — they need to demonstrate that the compute is generating a return, not just enabling a demo. Luna: The market is essentially asking: 'Are you building a moat or just renting one?' Lucas: That's a great way to put it. And the S-1 is where that story gets told. I've been reading the risk factors more carefully — many now include a specific section on cloud concentration risk and vendor lock-in. That wasn't common even a year ago. Luna: We saw some headlines this week about AI coding startup Cognition raising a billion at a 25 billion valuation. That's probably going to be a compute-heavy IPO when it comes. Lucas: Absolutely. And if you look at the recent performance of some compute-heavy public companies — NVIDIA is down almost 5 percent in the last five days, despite being the ultimate beneficiary of this trend. The market is getting more discriminating. Luna: It's not enough to say 'we need GPUs.' Investors want to know what you're building with them. Lucas: Right. And that brings us back to the S-1. The use of proceeds section used to be boilerplate. Now it's one of the most scrutinized paragraphs in the document. I expect we'll see more companies breaking it down by specific cloud provider, term length, and expected utilization. Luna: Which is also a signal to competitors. If you disclose you've committed $120 million to AWS, your rivals know your cost structure. Lucas: That's the trade-off. Transparency for credibility, but at the cost of strategic disclosure. Some companies are choosing to keep it vague — 'a portion of the net proceeds will be used for infrastructure investments' — but the market is pressuring for more detail. Luna: So what's the forward look? Are we going to see more companies delaying IPOs until they have better compute efficiency? Lucas: I think so. Or we'll see more structured deals — like IPO proceeds being held in escrow and released based on compute milestones. I've heard whispers of that being discussed at a few bulge-bracket banks. Luna: That would be a fascinating innovation. 'We'll give you the capital, but only as you prove you can use it productively.' Lucas: Exactly. It's a shift from IPO as a liquidity event to IPO as a financing tool tied to operational milestones. That's a different mindset. Luna: And it changes the conversation between founders and underwriters. The banks aren't just pricing the equity anymore; they're auditing the cloud contracts. Lucas: Which means the due diligence process is getting longer and more technical. I've heard of banker teams now including a cloud architect in the diligence meetings. That would have been unthinkable five years ago. Luna: So the next time we see a big AI IPO, we should look at the S-1 not for the valuation story, but for the infrastructure story. Lucas: Absolutely. That's where the real strategic signal is. And for listeners who are building companies — maybe think twice before committing to a three-year cloud contract right before you file. Luna: Or at least model the optics. The market is watching. Lucas: It sure is. That's it for this episode — we'll keep an eye on the next few S-1s and share what we find.