Bubbly Moonshots

On stagflation, neolabs, and how to handpick not-stupid AI categories

Breakthroughs and bubbles are entangled.

What to some is the moon is a bubble to others.

The same conditions that enable transformational companies also guarantee that most attempts will fail. Excess capital doesn’t just coexist with innovation; it’s the search cost.

AI is dissolving the scarcity mindset. When you believe you have more resources to execute — capital (obviously) and capability (as the stack commoditizes) — you can afford to dream bigger. The question is whether those moonshots land, and more importantly, whether the world would know in time if they won’t.

Think of it like monetary policy. Capital is inflation; capability is real GDP. Lots of capital chasing little capability feels like stagflation. But if capability has superlinear unlock, then that’s genuine growth regardless of the speculative vibes. The problem is that capability doesn’t announce itself like macroeconomic data. For quarters or years, breakthroughs and bubbles look identical from the outside.

Indeed, you can’t call it in real time. The bubble story writes itself either way: failures confirm the narrative, and until then, successful raises just look like excess. There’s no definitive proof while you’re living through it.

The neolabs are samples of this tension. Legendary researchers spinning out companies at skyrocketing valuations. Capability is undeniable, but man, the price is not cheap. You’re watching high inflation and high real GDP simultaneously. Maximum entropy.

(One could argue the price premium accounts for what separates neolabs from median AI startups: the researchers have taste, are experienced with the frontier so know what they don’t know, and have self-worth that enforce the honesty. Being proven wrong publicly is an existential threat to their reputation, so they don’t BS. The dangerous companies are the ones whose founders are tasteless or clueless, if not intellectually cowardly.)

“Is it a bubble?” is the wrong question. It assumes you can answer it before the answer exists. The dot-com era produced Amazon. Excess capital is the Darwinian selection mechanism. Most fail; the winners transform industries. You can’t have one without the other.

But let’s not forget global and local are different games. When the market asks “are we in a bubble,” it’s really asking whether enough moonshots land to justify the excess. For a VC, though, that’s almost irrelevant. You’re not betting on humanity’s aggregate outcome, rather specific companies in specific categories. If a bubble bursts in your hand, you get fired, even if civilization is better off. Bubbles, like moonshots, are more local than you think.

So how do you spot the real moonshots?

Count the self-proclaimed Neil Armstrongs. There can’t be too many.

In a genuine moonshot, people join missions bigger than their ego. In a bubble, everyone’s busy celebrating that they supposedly first planted the flag. Researchers who are team players build things none of them could alone. Researchers more interested in value capture than creation risk killing the future. The specific failure mode “dreaming bigger” creates is it lets founders defer the question of whether their particular, individual version of the dream matters. The moonshot framing provides cover for romanticism and narcissism to unhealthily intertwine.

That’s why categories that converge around two clear winners instead of spawning a hundred unheard-of startups are the ones to watch.

The iconic examples: Cursor and Cognition grabbing coding IDE customers, Skild and Physical Intelligence monopolizing the robotics foundation model space, Surge and Mercor eating up outlier expert data supply, and Fireworks and Together carving out cloud compute services.

More recent ones? Manus and Genspark debuting AI agents, Lovable and Replit dominating app/web design, Exa and Parallel leading search, and Axiom Math and Harmonic solving AI for formal verification.

Compare this with categories where who’s who confuses even the insiders. Code review. Tabular data foundation models. Enterprise agentic workflows. RL consulting.

King-making might be the rational strategy, as Apollo projects may have been identified under the critical assumption that top-tier players in the private market have done their due diligence on talent distribution. Whether something is future-bound depends on whether the best people board the same ship. While consolidation tells you something, fragmentation tells you more.

When faced with a new category, don’t mistake abundant motion for inevitable movement.

Two winners means the best people have voted with their careers. A hundred startups chasing the same thesis? While great minds think alike, they can crumble together too.

And yes, funding no.3 player just speeds up bubble popping.