Every few weeks, a press release lands in my inbox claiming some AI company just posted triple-digit growth numbers. Usually I skim it, roll my eyes, and move on. But occasionally the numbers are wild enough that they demand actual scrutiny—not celebration, not dismissal, just honest analysis. A 578% revenue increase qualifies as wild enough.

Big Percentage, Small Base

Here's the first thing any engineer worth their salary asks when they see a growth percentage that high: what's the denominator? Tripling a $10 million revenue base is impressive. Growing 578% off a $2 million baseline? That's a rounding error in enterprise software terms. The math isn't wrong—it's just conveniently framed.

Small AI companies can post explosive percentage growth for reasons that have nothing to do with product-market fit or sustainable competitive advantage. A single enterprise contract. A pilot program that converts. One lucky partnership announcement that pulls forward a bunch of deals. These are real revenue events, but they don't tell you whether the underlying business actually works at scale.

What Actually Matters Beyond the Headline Number

If you're trying to figure out whether a tiny AI company is the real deal or a well-timed story, here's what I'd actually look at:

• Net Revenue Retention (NRR): Are existing customers spending more over time, or are you constantly churning and replacing them? An NRR above 120% tells you customers are expanding usage. Below 100% means you're leaking faster than you're filling.
• Gross Margins: AI inference costs money—real money in compute bills. If a company is posting 50% gross margins while claiming to be an AI-native platform, either they've solved something genuinely interesting on the infrastructure side, or they're reselling someone else's API at thin margins and calling it a product.
• Customer Concentration: One client driving 60% of revenue isn't a business—it's a consulting engagement with a press strategy.
• Burn Rate vs. Revenue Quality: Explosive growth funded entirely by VC cash at negative margins is just subsidized market share. It's not a business model; it's a bet that someone bigger will acquire you before the runway ends.

The AI Tax That Nobody Talks About

Here's something the investor-facing narrative almost never mentions: AI companies face a cost structure that traditional SaaS doesn't. Every query costs compute. Every model update requires retraining runs. Every new customer potentially increases your inference bill faster than your revenue grows—especially if you're in a competitive market where you can't price aggressively.

The companies that survive long-term are the ones figuring out how to make inference cheaper (through model distillation, quantization, smarter batching) while simultaneously building switching costs into their product layer. That's the actual technical moat. Not the model weights themselves—those get commoditized faster than anyone wants to admit—but the data flywheel, the workflow integrations, the switching pain that makes customers stay.

Healthy Skepticism Isn't Pessimism

None of this means a 578% growth story is automatically a fraud or a fluke. Early-stage companies legitimately find product-market fit fast sometimes, especially in a market where enterprises are actively looking for AI solutions and willing to sign contracts before the product is fully baked. That's a real dynamic right now.

But it does mean you should want to see the second and third acts before writing the hero's journey. What does the growth curve look like in 18 months when the low-hanging enterprise pilots have been signed? Can the team actually deliver on the implementations they're promising? Is the tech defensible, or is it a wrapper around an OpenAI API endpoint dressed up in a nice dashboard?

The companies worth paying attention to aren't the ones with the biggest percentage headlines. They're the ones who can explain, clearly and without buzzwords, exactly why their customers can't leave.

The Bottom Line for Builders

If you're building in AI right now, the lesson from these high-growth small companies isn't envy—it's pattern recognition. The ones that convert early momentum into durable businesses tend to share a few traits: they own a specific workflow deeply rather than spreading thin across use cases, they've figured out a cost structure that doesn't bleed them dry as they scale, and they've built enough integration depth that switching costs are real.

A 578% growth number is a conversation starter. The actual story—the one worth caring about—is everything that number doesn't tell you.