Crazy, and Rational: Venture's Cognitive Dissonance Moment

Having spent 27 years in and around private markets, including 22 years at SVB and First Republic where I had a front row seat to the dot-com era and everything after, I’ve seen my share of cycles.

And yes, I think we are showing every sign of late cycle market behavior when it comes to AI. However, I also think we are living in a duality right now, and if we only hold one side of it, we are going to miss the other side of bubbles.

First, the signs of froth are unmistakable.

The rhymes with 1999 and 2021 are unmistakable. Too much capital. Valuation insensitivity. FOMO at every stage. Incentives that push prices higher because nobody in the chain gets paid to sit out (and in reality, the incentives are to participate more heavily during these times & principal-agent issues get worse).

A few things I’m seeing right now:

1. Companies are raising every 5-6 months again. Cursor went from roughly $2.5B in December 2024 to $9.9B in June 2025 to $29.3B in November, was in talks above $50B by April, and just got acquired by SpaceX for $60B. That’s a 24x move in about 18 months. To be fair, the revenue growth underneath is real and unlike anything we’ve ever seen. But when a company reprices every two quarters, the round isn’t simply pricing the business today. .

2. $100M+ seed rounds. Call them whatever you want, but these aren’t seed rounds. Reflection AI went from $545M to a reported $25B in about a year, before shipping its flagship model. The “lab” companies are the most prominent in getting mega rounds early, but these rounds are simply a high strike price option premium.

3. Funds are getting bigger and buying anything in AI that shows rapid growth. Bigger vehicles need bigger checks, and the only companies that can absorb them are the ones already marked up. So everyone piles into the same names at whatever the price is that week. It’s catching the tiger by the tail. You can’t let go because the marks fuel your next fundraise, but you can’t really steer either.

4. The access market has gotten sloppy, and in some cases predatory. This is the part that worries me most because it’s where the least sophisticated capital gets hurt. Double and triple layered SPVs, each stacking fees and carry on the layer below (even worse is the suspected fraud that is happening with those claiming to have access to “hot names”). Even if not fraud, run the math on a three layer structure with high upfront fees and stacked carry. And in plenty of cases, if it’s a secondary, the underlying transfer was never authorized in the first place. OpenAI, Anduril, and Anthropic both publicly declared the intention of cracking down on unauthorized indirect share transfers.

So yes, valuations at nearly every round are dissociating further and further from where the companies actually are, and the market is choosing to ignore the macro and micro risks. AI is still new. Things are changing extremely fast. It’s genuinely uncertain which of these businesses survive long term, let alone grow into their marks.

And high valuations can be a potential poison pill that, if everything doesn’t break perfectly, will impair a company’s path to future capital, risk turning employee equity upside down, and force companies into desperation.

And yet

Every supercycle has produced bigger winners than the one before it, and the gap keeps widening. 1999, for all its flameouts, gave us Google, Amazon, and eBay. Google went public in 2004 at roughly $23B and people called it expensive. The mobile and cloud wave gave us Uber, Airbnb, and Snap, companies that went out at valuations that would have been unthinkable a decade earlier. Each cycle, the survivors weren’t just somewhat bigger. They were an order of magnitude bigger.

SpaceX just completed a roughly $1.77 trillion IPO and briefly touched $2 trillion on day one. Anthropic went from about $1B to a reported $47B revenue run rate in roughly fifteen months and raised at $965B. OpenAI raised at $852B. Thomas Laffont at Coatue made the point recently that these three listings alone could exceed the total exit value of all venture backed companies over the past decade. Coatue’s data shows that companies reaching $100B have a 31% chance of getting to $1 trillion.

Why do the winners keep getting bigger? Speed of distribution and adoption, and TAMs that have expanded from consumer internet to essentially all knowledge work. Google took years to reach the revenue Cursor hit in months. Software now deploys globally in a weekend, and AI compresses the distance between product and revenue in a way we’ve simply never seen.

The reality is this is something the VC universe knows. They know that the vast majority of companies will not make it as the true themes of AI winners becomes more visible. But they also know overpaying for the chance at getting a top 5% company is worth it as getting one can mint a firm, not just a fund. But the difficulty of getting one of these mega scale companies will prove to be harder than it seems now. Additionally, if investors are paying premium prices, the impact of the winners will be muted unless the scale is enormous ($25B+).

Where I think this lands

I don’t know when this period of froth ends. Six months, twelve months, two years. Anyone giving you a date is guessing or selling. What I do know is that gravity will be back at some point.

But here’s what I suspect we’ll say when we look back at 2026 a few years from now:

1. The winners of this cycle will be the biggest we’ve ever seen, continuing a 30 year trend, driven by distribution speed and TAM expansion that prior cycles didn’t have.

2. The % of real winners will be FAR smaller than the % of companies people believe are going to be winners. Also, it’s important to separate capital markets from technology, as some of these companies may win from a business standpoint, but still turn out not to be wins for investors that came in late.

3. After a period of disillusionment, we will see another wave of slope of enlightenment, and that period will likely see far more expansive growth than we have ever seen from technology.

So, I expect to see this period result in the widest dispersion between winners and losers venture has ever produced. Which leads to the fund level consequence: portfolios without at least one top 5-10% asset of this cycle are going to significantly underperform, full stop. Average exposure to the AI theme will be worth close to nothing.

Those most aggressively calling this the top of the market will be right about most companies. The maximalists will be right about a handful.

For allocators, a few practical implications. The question isn’t whether to have exposure to this cycle, it’s how. Manager selection and access matter more than timing. I don’t think just being top-quartile in this world will necessarily mean much. It’ll be the top 10% of funds and then everyone else.

We’re somewhere in the late innings of this phase, and a lot of today’s marks will turn out to have been fiction. When the tide goes out this time, the shore will be littered with companies that were priced for a future they never had a real path to. But standing among them will be the largest companies ever created. Both eras of every supercycle wash up together. The flameouts just outnumber the giants. The giants might just be big enough to make it not matter.