The VC tide has gone out. A real innovation wave is now arriving.

For years after the dotcom bubble burst, being a founder was weird and hard. Things were just starting to pick up again when the bottom fell out of the global economy.

The GFC kicked off the most sustained period of fiscal and monetary stimulus in modern times. Just as the Fed was coming off the gas COVID hit, forcing a quick return to zero interest rates plus a massive new liquidity pulse to keep the lights on after we all went home.

When borrowing is free all kinds of strange things start to happen. Easy to spot stuff like price inflation and frothy equity markets. But also less tangible shifts in the perception of risk, or the lack thereof.

For founders, the past fifteen years were a dizzying elevator ride from scarcity to abundance, with a final wild peak of absurdity. Being a founder went from gritty and unfashionable, to glamorous and well-paid, to a LARP that YOLOing college grads played with other people’s money.

Fueling it all was a headless Ponzi scheme:

– LPs reaching for yield overallocated to venture.

– Existing GPs supersized, and new GPs sprang up by the hundreds to catch the crumbs.

– More funds deploying more dollars lowered the quality bar while pushing up valuations.

– Startups booking sales to other startups laundered the money flood and called it “blitzscaling”.

– Bigcos, with inflated stock as currency, picked off the winners at eye-watering multiples.

– Liquidity flowed back to LPs, who eagerly recycled their gains on ever-shortening fund cycles.

Unlike the dotcom bubble, this last one was mostly a waste. In Carlota Perez terms, there was no messy but productive “installation phase” for an important new industrial capacity. Instead, we got a series of trivial and mostly made-up “next big things” like crypto, AR/VR, fintech and the Metaverse.

In early ‘22, the Fed finally took the punchbowl away. For the first time in a long time, money wasn’t free. Thank god.

Building a generational company from scratch is the hardest thing you can do in capitalism. It’s not, or at least it shouldn’t be, a YOLO lark with other people’s money.

But VC bubbles deflate slowly. With private markups hard to come by and liquidity drying up, it can take years for LPs to figure out the J curve is actually an L curve. But with the risk-free rate at 5% and the Ponzi played out, the illiquidity of venture just doesn’t pencil for most LPs. The tide will keep going out.

What’s the plot twist in this story? The money fountain is sputtering just as the first real innovation in a decade roars into view.

After half a generation of overhyped trivialities, Large Language Models have reminded us what real technological breakthroughs look like. Less than two years after OpenAI released the first version of ChatGPT, the user adoption rate for GenAI now exceeds that of either the Personal Computer or the Internet. 

The only thing growing faster than GenAI adoption is the capital budgets of foundation model competitors. OpenAI just raised the biggest venture round in history ($6.5B at $150B post), with Anthropic a relative bargain at just $40B. Meanwhile, Meta is either fueling the fire or salting the earth by pumping out open-source models as capable as the private ones.

As useful as they appear to be, training new foundation models is wildly expensive, well beyond the deployment capacity of even the richest VCs.

As a result, the model wars are mostly not powered by venture dollars. Instead, the big winners of the last cycle – Amazon, Google, Meta and Microsoft – are jockeying furiously for continued dominance over the next one. LLMs are compute and energy hogs, and renting state of the art bundles of compute and energy by the millisecond is what tech incumbents do best.

So what’s a founder to do?

As the easy money dries up and the capex elephants dance, the real opportunity is snapping into view. Like PCs, smartphones and the cloud before them, LLMs will transform the way knowledge work is done. But unlike prior technology waves, LLMs won’t just give humans superpowers, they will increasingly replace those human workers entirely.

As always, the LLM business cycle will be powered by corporate buyers. Homework cheaters and deepfake campaign videos are just the cultural exhaust of capitalism’s relentless quest for new levers of productivity and competitive edge.

Inevitably, billions will be wasted on wrong bets about where value will accrue in the coming wave. How can private models recoup their training costs when open source ones are just as good? Will small models and bounded datasets beat LLMs crunching the sum of human knowledge? Will selling AI tools to incumbents prove more valuable than “full-stack” competitive attacks? No one really knows, and that’s what makes it interesting.

But one thing is certain: as the money tide recedes and uncertainty spikes, being a founder is becoming gritty and hard again. LLM breakthroughs are coming too fast, and finding a place to stand that doesn’t get washed away by the next model release is too hard for most teams to handle. Real builders working backward from customer problems are finding ways to stay relevant, but the game has barely begun.

Investors and founders can wait decades for moments like this one. Let the fun begin.