Latest / The Startup Exit Podcast with Fexingo: IPOs, Acquisitions, and Founder Liquidity Events / Why AI Models Are Eating the Power Grid
Transcript
- Lucas: So we are looking at a market where the biggest risk to a startup's exit isn't just product-market fit anymore, it's the electric bill. With NVIDIA up five point four percent over the last week and ARK Genomic Innovation jumping ten percent, you can see where the money is flowing, but the underlying infrastructure cost is what really changes the exit math. Luna: Right, because if your unit economics are getting crushed by data center overhead, your valuation multiple gets compressed before you even hit the IPO window. Lucas: Exactly. And I am seeing headlines about world model companies keeping their secrets, which sounds like marketing speak for they are hiding how much energy they burn to train these things. It creates a transparency problem for investors who are trying to value the company based on growth rates that don't account for exponential power costs. Luna: It feels like the era of infinite cheap compute is over, so founders have to be more surgical with their training runs now. Lucas: That is the shift. Let me give you a concrete example from this morning. Y Combinator alum Angle Health just hit a two point seven billion dollar valuation. That is a solid number, but when you drill into insurance tech, the margin for error is thin because the compute needed for underwriting models is intense. If their energy costs spike, that two point seven billion becomes harder to defend in an acquisition talk. Luna: So the efficiency of the algorithm is now a direct driver of enterprise value, not just a technical footnote. Lucas: Precisely. In previous episodes we talked about founder vesting cliffs and dilution, but those are static numbers. Energy usage is a variable that scales with ambition. If you double your user base, you might need to triple your compute capacity depending on the model architecture. That non-linear cost curve eats into the EBITDA that acquirers look for. Luna: Which means founders who prioritize model efficiency over raw scale might actually get better acquisition terms later. Lucas: I think so. Look at the recent news about Automattic naming an interim CFO after exec departures. While that is a corporate governance story, it signals that operational discipline is becoming critical. When you are managing high burn rates due to cloud or dedicated hardware costs, every departure matters because the remaining team has to justify the spend to the board. Luna: It reminds me of how delivery costs used to shape Uber's early days, except now the delivery is terabytes of training data instead of physical meals. Lucas: That is a great analogy. The logistics are digital, but the constraints are physical. You cannot virtualize electricity. So when Disney hired a former AI startup CTO who they once accused of copying characters, it shows how entrenched these AI ops teams are becoming. They are not just building apps; they are building energy-intensive engines. Luna: And those engines require massive upfront capital expenditure that delays profitability. Lucas: Right. If you are a founder planning an exit in eighteen months, you need to know your power consumption per active user. Investors will ask for that metric alongside customer acquisition cost. If you cannot show that your marginal compute cost is decreasing, your exit valuation will suffer. Luna: Do you think smaller startups can survive this without being acquired by the big hyperscalers? Lucas: They can, but only if they niche down. General world models are too expensive. Specialized models for specific industries, like that insurance work Angle Health is doing, allow you to optimize the hardware stack. You don't need a thousand H-100s; you need a few hundred optimized for your specific data shape. Luna: So specialization is the hedge against the energy arms race. Lucas: Absolutely. And this brings us to a practical question for listeners who are building or investing right now. How do you price this risk? Because the market is still pricing in growth, not grid stability. Luna: We should probably pause here to support the show. If today was worth a coffee to you, that's the link — buy me a coffee dot com slash fexingo. Lucas: Yeah, thanks for listening. It keeps the lights on, literally and figuratively. Now, back to pricing that risk. Luna: Right, because if the grid gets congested, your server latency goes up, and your service quality drops. Lucas: Correct. So I want you to think about your own business or portfolio. Are you factoring in energy volatility? Because in September twenty twenty-six, that is a material line item on the P&L. Luna: Most people are still thinking in software margins, not utility margins. Lucas: And that mismatch is creating opportunities. Companies that have secured long-term power purchase agreements are actually more attractive acquisition targets than those burning cash on spot-market electricity. Luna: So security of supply is becoming a competitive advantage. Lucas: It is. Let's look at the stock market for a second. Shopify is down four percent, and Roblox is down almost seven percent. Those are consumer-facing platforms, but they rely heavily on stable, scalable infrastructure. If their costs rise, their guidance gets cut. Luna: Which makes their founders' locked-up shares less valuable if the earnings miss. Lucas: Exactly. So the exit strategy isn't just about hitting revenue targets; it's about controlling the cost structure. Founders who understand the physics of their business are the ones who walk away with liquidity. Luna: It seems like the best time to sell is when you can prove you are efficient, not just when you are growing fast. Lucas: I agree. Efficiency is the new growth. If you can show that your compute cost per transaction is dropping while your user base expands, you command a premium multiple. Luna: Because that proves scalability without proportionate burn. Lucas: Right. And for the founders listening, start tracking your watts per user now. It will be one of the first questions an acquirer asks during diligence. Luna: A very grounded metric for such a futuristic industry. Lucas: It is. And it separates the hype from the operation. If you are building something that requires significant AI, you are in the utility business now, whether you like it or not. Luna: So the exit is not just about the code, it's about the contract with the power plant. Lucas: Well put. We will leave it there for now. Think about your own leverage points in the coming quarter. Luna: Thanks for tuning in to the Startup Exit Podcast.