Latest / The Series A Show with Fexingo: Early-Stage Funding, Pitch Decks, and Startup Milestones / Series A Pitches Are Adding a Compute Pricing Slide
Transcript
- Lucas: So you're seeing it more and more now, Luna — the compute pricing slide. It's the slide where a founder puts a number on what it actually costs to run their AI model, and then shows how that number changes as they scale. Luna: I've definitely noticed that in recent decks. But is it really a new thing, or is it just the old unit economics slide wearing a different hat? Lucas: It's a bit of both, honestly. But the reason it's becoming its own slide is that AI compute has become such a dominant cost line. I mean, look at some of the public numbers. There was that report earlier this year that some AI startups are spending more on cloud credits than on salaries. And that's not sustainable for a Series A company. Luna: So the slide is really about showing investors that you understand your cost structure. That you're not going to blow through your Series A in six months just on GPU time. Lucas: Exactly. And it's becoming a non-negotiable for some VCs. I was talking to a partner at a growth fund last week, and she said she won't even take a meeting with an AI startup unless the deck has a slide that shows the cost per inference or per token, and how it's going to drop over time. Luna: That's a pretty high bar. What does a good compute pricing slide actually include? Lucas: Typically it's a chart showing the cost per unit — so per API call, per query, per generated image — and then a line that shows how that cost declines as you scale. You want to show that you've thought about things like model quantization, distillation, or using cheaper hardware for inference. The slide basically says: 'Here's what we spend today, here's what we'll spend at 10x volume, and here's how we get there.' Luna: And I imagine it's not just about the cost, but about the trend. If the cost per unit is flat, that's a red flag. You want to see a downward slope that convinces you the unit economics will actually work at scale. Lucas: Right. And that's where the slide gets interesting. Because it's not just a cost slide — it's a thesis slide. It's saying: 'We understand that AI is expensive today, but we have a plan to make it cheaper.' And that plan is often more important than the actual number. Luna: Has there been any pushback from founders? Like, 'this is just another thing you're asking us to put in the deck'? Lucas: Some, yeah. But the best founders are already thinking about this. They know that every dollar of compute is a dollar that could be spent on engineering or sales. And honestly, if a founder can't articulate their compute costs, that tells you a lot about their understanding of their own business. Luna: So it's a filter. Lucas: It's a filter. And it's also becoming a benchmarking tool. There's a startup that's helping Wall Street price AI compute, which is a whole other layer. The idea is to create a standard way to measure what AI compute is worth, almost like a commodity index. That would make it easuser to compare startups and to value them. Luna: That's fascinating. And I guess it's a sign that the market is maturing. We're seeing the same kind of infrastructure that grew up around cloud computing — the benchmarks, the pricing indices — now emerging for AI. Lucas: Exactly. And you know, it's not just AI startups. Even public companies are being scrutinized on this. I was looking at ARK's genomics ETF, which jumped seven percent in the last five days. That's partly driven by the idea that AI is making drug discovery cheaper and faster. But the whole sector is sensitive to compute costs. Luna: So the compute pricing slide is really about connecting the AI cost structure to the business model, and showing that you can be the one who actually makes money. Lucas: Right. And it's a conversation we're having a lot on this show, and it's one that I think is really valuable for founders who are building in this space. And that's actually a good point to bring up something about this podcast itself. We keep this show free and ad-free, and that's a deliberate choice. If you find these conversations useful for what you're building or running, you can support us at buy me a coffee dot com slash fexingo. It helps us keep going without taking ads. Luna: Yeah, and we really appreciate that. It's what lets us keep digging into these topics without any sponsor pressure. Lucas: So back to the compute pricing slide — one thing I want to stress is that it's not just a cost projection. It's also a reflection of your team's technical chops. If you can show that you've optimized your model to run on less expensive hardware, or that you've learned to use your GPUs more efficiently, that's a signal that you're a team that can execute. Luna: I've seen decks where the compute slide is just, like, one line: 'GPU costs will decrease over time.' That doesn't cut it anymore. Lucas: No, it doesn't. Investors want to see specific numbers and a credible path. And it's not just about the unit cost — it's also about the total burn. If you're spending $200,000 a month on compute at Series A, that's a huge chunk of your typical $10 million round. You need to show that you can get to a point where the cost is manageable. Luna: And I imagine the slide also has implications for the valuation. If your compute costs are high and you can't show a clear path down, that's going to put pressure on your multiple. Lucas: Absolutely. And I think we're going to see more sophisticated investors use this slide to negotiate down valuations. They'll say, 'Look, your cost structure is too heavy, so we're only willing to value you at X.' Luna: That sounds like a good thing in some ways, because it forces founders to be more disciplined from the start. Lucas: It's a discipline that pays off. I think we're moving towards a world where every AI startup will have to show a compute pricing slide, not just the ones that happen to have it. And that's going to make the whole ecosystem healthier. Luna: So if I'm a founder listening to this, what's the one takeaway? Lucas: Start building that slide now. Even if you're not raising yet, start tracking your compute costs religiously. Know your cost per unit, and understand what drives it down. Because when you do go out to raise, that slide is going to be one of the first things investors ask for. Luna: And it's not just for AI startups. Even if you're not using AI, showing that you understand your infrastructure costs is a sign of maturity. Lucas: Exactly. And that's the kind of signal that sets a good pitch apart from a great one. So, what do you think — will we see this become a standard slide in every Series A deck within the next year? Luna: I wouldn't be surprised. We've seen it go from zero to almost mandatory in the last few quarters. It's a trend that's clearly here to stay.