Latest / The Series A Show with Fexingo: Early-Stage Funding, Pitch Decks, and Startup Milestones / Infrastructure Cost Is the New Series A Conversation
Transcript
- Lucas: So Shopify's stock is up nearly thirty percent this week, and everyone's calling it an earnings beat. But the thing I keep coming back to isn't the revenue number, it's the margin story hiding underneath. Luna: You mean the fact that they're finally showing real operating leverage? Lucas: Exactly. And that's the same conversation I'm now hearing in Series A rooms. Not 'what's your gross margin' — that's table stakes. It's 'what does your infrastructure bill look like in year three if your usage doubles every quarter.' Luna: So infrastructure cost has become a headline diligence item. Lucas: It's becoming the thing. I've seen three pitch decks in the last month where founders have a slide titled 'Compute Efficiency' and they're walking through their per-request costs like they're presenting a public company's margin bridge. Luna: Give me a concrete example of what that looks like in practice. Lucas: Take a startup building an AI agent for customer support. Their old pitch said 'we automate tickets.' The new pitch says 'our agent costs four cents per interaction, and we're driving it down ten percent month over month.' That's the kind of specificity investors are suddenly demanding. Luna: Because a four-cent cost today could be a two-dollar cost at scale if the model isn't efficient. Lucas: Right. And that's why you're seeing VCs push back on the 'just throw more GPUs at it' approach. The smart ones are asking about model routing, caching, batching — all the boring stuff that determines whether your gross margin survives contact with real demand. Luna: It's interesting because a year ago, the question was 'how fast can you ship?' Now it's 'how efficiently can you scale? ' Lucas: And that's a real shift. I think part of it is the market environment — everyone's been burned by companies that grew like crazy but couldn't ever show a path to profitability. The other part is that the cost structure of ai native companies is just so visible now. Luna: Visible how? Lucas: Because the input costs are public. Everyone knows what a GPU costs, what inference per token roughly runs. So when a founder says 'we're gross margin positive,' the investor can actually sanity-check that on the back of a napkin. That wasn't true for a lot of software where the COGS were murky. Luna: So it's like the market finally has a way to hold AI companies accountable. Lucas: Exactly. And the founders who get this are building infrastructure cost into their pitch from day one. They're not waiting to be asked. Luna: Let's talk about what a good infrastructure cost slide actually looks like. What should founders include? Lucas: The best ones I've seen break it into three buckets: compute, data, and integration. Compute is obvious — model calls, hosting. Data is storage and retrieval, which gets expensive fast if you're indexing a lot of customer data. Integration is things like API costs, telemetry, third-party services. Luna: And they should show how each scales with usage. Lucas: Right, not just a static number but a sensitivity analysis. 'Here's what our unit cost looks like at ten thousand daily active users, a hundred thousand, a million.' That's the kind of modeling that separates a founder who's thought about this from one who's hoping it works out. Luna: I also love the ones who tie it back to customer lifetime value. Lucas: Oh, that's the killer move. If you can show that the lifetime value of a customer is five times your cumulative infrastructure cost, you've basically answered the unit economics question before anyone asks. Luna: And if you can't? Then it's probably not a viable business, or at least not a venture-scale one. Lucas: Which is a hard truth for a lot of AI startups. But better to face it in the pitch room than after you've raised and burned through the round. Luna: That's actually a useful framing for anyone running a startup. And if today's episode gave you a new way to think about your own business, that's exactly what we're going for here. The show stays ad-free because listeners like you support it directly. If you found value in this, you can buy us a coffee at buy me a coffee dot com slash fexingo. Lucas: Yeah, it genuinely helps keep these conversations rolling. And every little bit goes toward keeping the podcast independent and free from sponsors yelling at you. Luna: Now, back to that sensitivity analysis — what's a common mistake founders make when they build that model? Lucas: The biggest one is assuming your cost per unit stays flat as you scale. In reality, you might get volume discounts on compute, but you also add complexity — more model calls for routing, more retries, more data egress. So the curve is often U-shaped, not linear. Luna: Meaning your unit cost might dip then rise again. Lucas: Exactly. And if you don't model that, you're flying blind. I saw a pitch once where the founder had a beautiful linear chart showing costs going down forever. I asked about the complexity factor and they hadn't considered it. The whole slide fell apart. Luna: So it's not just about the numbers, it's about the thought process behind them. Lucas: Right. Investors are looking for founders who understand the levers. And that's why infrastructure cost is becoming a proxy for founder maturity. It shows you're thinking about the business as a business, not just a tech demo. Luna: Let's talk about the Cloudflare browser launch this week. Kitesurf, right? Does that change the calculus for startups building for agents? Lucas: It's a signal that agent traffic is becoming a first-class citizen on the web. If browsers are being built specifically for agents, then infrastructure costs for agent-native startups are going to be even more scrutinized. You can't just assume you'll pass those costs to the customer. Luna: So it's another layer of pressure on unit economics. Lucas: And another reason why founders who can talk about per-agent costs — compute, memory, data — will stand out. It's the same conversation, just with a new entity in the mix. Luna: I'm curious — is there a sector where this infrastructure cost conversation is most acute right now? Lucas: I'd say anything in the vertical AI agent space. Those startups are building on top of multiple models, orchestrating workflows, and the cost per completed task can vary wildly. A customer service agent that handles a complex refund might cost ten times more than one that answers a simple FAQ. Luna: So you need to price accordingly, or at least know your margin per interaction type. Lucas: And that's the sophistication level investors are now expecting. It's not enough to say 'we have AI.' You need to say 'here's our cost structure, here's how it scales, here's where we have leverage.' Luna: Have you seen any examples of founders who nailed this and it made a difference in their round? Lucas: There was a startup in the legal tech space — I won't name them — but they had a slide that broke down their cost per document review by complexity tier. They showed how they were routing simple contracts to a cheaper model and complex ones to a premium one. It was the most detailed infrastructure cost analysis I've seen in a Series A deck, and it was a big reason they got multiple term sheets. Luna: So being granular about costs can actually be a competitive advantage in fundraising. Lucas: It's a differentiator, for sure. Because it shows you're not just tech-optimistic, you're business-realistic. And in this market, that's a signal investors are hungry for. Luna: So what's the takeaway for a founder preparing their Series A pitch right now? Lucas: Build your infrastructure cost model before they ask. Have the slide ready, understand your unit economics today and how they could evolve. It's not the most glamorous part of the pitch, but it might be the one that closes the deal. Luna: And maybe it's also a sign of how the industry is maturing — we're moving past the gold rush and toward the businesses that can actually survive the long haul. Lucas: I think you're right. The next wave of Series A winners won't just be the ones with the flashiest demos, but the ones who can show a path to sustainable economics. And that starts with knowing your costs cold.