What "trend" even means has changed. The word can be read two ways. In the old paradigm, a technological trend meant the steady development of a tech stack or an industry in a given direction over a strategic horizon — at least 3-5 years. Think "the trend toward automation," "the trend toward Uber-ization." But in fashion, "trend" means something else entirely: whatever's most relevant right now, this exact season. As in, "the trend color of summer 2026 is chartreuse — a yellow-green shade."

Web3 felt this shift in meaning first. There, it's usually called a "narrative," and narratives run for about six months, often less. Catching a narrative means getting to market fast while the topic is hot, pulling in investor and user attention, and then trying to ride that momentum forward while you reinvent yourself around the next one.

The same thing is now happening to the classic Web2 venture. AI has accelerated everything so much that planning more than five years out looks like naive madness. Startups increasingly rebuild their decks and strategies around whatever's being talked about today, chasing as much money and as many users as possible on whatever has consensus right now — instead of pitching their own picture of the future. And investors are chasing "this season's" popular thing, too, because the future is accelerating so fast that forecasting it feels genuinely frightening. It's easier for the brain to accept what everyone else is already saying.

Nowhere is this clearer than in the YC and a16z Speedrun batches — the two biggest trendsetters, rewriting what they ask founders to build every single cycle. YC's last three batches listed 14, 8, and 16 categories of startups they wanted to see — zero repeats across all three. Not a single category stayed "in trend" for more than half a year. Summer '25 was all AI copilots. Spring '26 shifted toward AI-executors and agentic systems. Summer '26 shifted again to physical AI and hard tech.

Speedrun is a little more stable. It's Winter/Spring and Summer/Fall 2026 cohorts overlap because both trace back to "14 Big Ideas for 2026" — the annual memo a16z speedrun partners publish at year-end as a forecast for the coming year. The two cohorts resemble each other, though not 100%. But the overlap with the final batch thesis from the previous year, 2025, is still zero. In other words, the fund operates literally on ~1-year trend cycles: spot or manufacture category leaders while it's hot > invest over the following year > rotate into the next narrative.

How many of these trend cycles is a 10-year fund supposed to survive? An initial investment period usually runs at least 3 years. And what about an exit in another 5–7 years? So how do you build a fund thesis under these conditions? You can no longer build a strategy around something that lives for literally one season. Every GP's job just picked up an asterisk — you have to find something sturdier to anchor the fund's strategy on. Because the fund's job was never to guess which of YC's sixteen categories will still matter three years from now, it's to build a thesis that doesn't need any of them.

More and more funds are showing up with the thesis "we just follow YC and Speedrun." Some write "AI" as their thesis, telling LPs that AI is diverse enough that the important thing is to get in the fight and figure out the rest later. We at Prosto VC went a different route — investing in "Frontier tech": technology stacks and industries that aren't in the spotlight yet, but will matter in the 2030s — quantum computing, spacetech, robotics, biotech. And on every deal, we take on the job of justifying why we believe that specific portfolio company will still be alive and relevant past 2030.

LPs want to hear about what's hot right now, too. When I pitch a quantum thesis while all of Twitter is talking about AI agents, I end up spending more time explaining why I'm not in consensus than explaining why our thesis is actually right — and that, not consensus-chasing, is literally the job I'm being paid to do as a GP. By the time cab drivers are talking about it, it's already too late — you need to get into startups before they land in the big funds. Seasonal trends are the foam left behind once the wave has pulled back.

Today's trends end the moment YC announces its next batch. A GP's actual job is to hunt for the invariant — whatever doesn't depend on consensus at all: Moore's Law and its analogs in quantum systems, the physical limits of energy and materials, demographic curves, regulatory cycles that physically cannot reset faster than a few years. YC and Speedrun cannot rewrite the laws of thermodynamics in their next cohort. If all you want is to "be on trend," you don't need to hand your money to a GP to manage. But if you want actual multiples, you still need exactly what you needed before AI showed up. This summer's shade of chartreuse fades out by fall. But it's strange to watch venture capital — an industry that, by definition, is supposed to see ten years further than everyone else — start living, in some people's heads, on the podium runway's seasonal calendar.

As they say, what's the difference between a politician and a statesman? A politician, like a blogger, says and does whatever will get the most "likes," while a statesman does what needs to be done. The GP of the fund should fall into the second category because time will judge him once the fund closes, based on concrete figures rather than the number of trendy investments.