The Story
For about fifteen years there was a reliable recipe for building a fast, valuable startup. Write software. Sell it as a subscription. Keep the gross margins somewhere north of 75%, because copying code costs nothing. Grow. That was the app era, and it minted an entire generation of unicorns without anyone ever having to pour concrete or touch a wafer.
For a live example of owning the recurring relationship, see Apple’s 1.5 billion paid subscriptions.
That recipe is quietly breaking. And the clearest sign is what venture money is actually chasing now.
Start with the raw numbers, because they’re stark. In 2025, global venture and growth investors deployed $425 billion, up 30% from the year before — and roughly half of it, about $212 billion, went to AI, per Crunchbase. But “AI” here doesn’t mean a clever chat app. Five companies — OpenAI, Scale AI, Anthropic, Project Prometheus, and xAI — each raised over $5 billion and together took in north of $80 billion. That’s on the order of a fifth of the entire year’s venture capital going to five names. The concentration got even more extreme in early 2026: foundational AI startups pulled in $178 billion in a single quarter, more than they raised in all of 2025.
These are not asset-light companies. OpenAI crossed a $500 billion valuation in late 2025 and reportedly ran past $800 billion in early 2026; SpaceX sat near $800 billion as the most valuable private company on earth; Anthropic ran from roughly $183 billion to a reported $965 billion in months. The unicorn board as a whole approached $7 trillion. The biggest value in private markets now sits in things that need power plants, fabs, rockets, and data centers to exist.
Someone gave this shift a name. In a July 2026 piece for Entrepreneur, writer Logan Simmons called the emerging category “civilization technology” — businesses that “support the capacity to maintain today’s level of civilization” and scale it for a growing, hungrier world. His list is telling: energy and nuclear (SMRs, fusion, the grid), AI compute and chips, advanced manufacturing and reshoring, defense, space, robotics, supply-chain resilience, biosecurity. As he puts it, “the money and the opportunity are now in the hard stuff: power plants, rockets, autonomous ships, reactors, genomes.”
The phrase is new, but the argument isn’t a one-off. At Andreessen Horowitz, Katherine Boyle has spent a few years building the firm’s “American Dynamism” practice around almost the same map — defense, manufacturing, energy, aerospace, critical infrastructure — and one of her recurring observations is that talent is now moving out of Big Tech and into hard tech. Founders Fund and Khosla have been writing the same kind of checks. Defense made the point concrete: Anduril raised at a $30.5 billion valuation in June 2025, doubled to $61 billion in May 2026, and by late July was reportedly in talks near $100 billion. A defense hardware company tripling in about a year is not how the app era worked.
So why now? Three forces are stacking.
First, the money has to build physical things. Training and running frontier models is a capex problem, not a code problem. The four biggest cloud providers spent around $131 billion in a single quarter of 2026, and Goldman Sachs has modeled cumulative AI capex in the trillions through 2030. That spending lands as concrete, transformers, and silicon — atoms, not bits.
Second, geopolitics turned “industrial base” back into an investable thesis. Reshoring, defense, energy independence, and secure supply chains stopped being policy talking points and became venture categories. I covered a version of this when Washington bought equity stakes in nine quantum computing companies — the government acting like a frontier investor is part of the same story.
Third, and least obvious: software got easier to copy, so its moat shrank. Wing Venture Capital’s Chris Zeoli made this case sharply in an essay titled “Software Ate the World. Now Hardware Is Eating Software.” Traditional SaaS multiples have compressed to roughly 6.7x revenue, down from an 18.6x peak in 2021, while AI-native apps run 50–60% gross margins instead of SaaS’s 75–90% — because every query re-runs a model, and that cost doesn’t fall with scale the way copying code did. When the open-to-closed model gap collapses and 2.2 million models sit on Hugging Face, the defensible thing isn’t the app on the screen. It’s the grid, the robot, the proprietary data, the fab. Value is moving down the stack.
Now the honest part, because this frame can be oversold. “Half of VC went to AI” is not the same as “civilization technology has won.” A huge share of that money is concentrated in a handful of foundation-model labs whose economics are still unproven — inference already eats ~23% of revenue at scaling AI companies, and nobody knows where that settles. Deep tech is genuinely hard: it’s capex-heavy, slow, and littered with fusion and space companies that burned billions and shipped nothing. Some of these valuations are a bet on a future that hasn’t arrived. And SaaS didn’t die — it repriced. “Civilization technology” is a useful lens, not a law.
The Takeaway
Here’s my read. The interesting thing isn’t the label — labels are cheap, and “civilization technology” is one writer’s coinage that happens to rhyme with what a16z, Founders Fund, and Wing have all been circling from different angles. The interesting thing is that three independent groups arrived at the same map. When a defense-focused GP, a hardware-stack VC, and a business columnist all point at power, compute, and physical infrastructure, the frame is probably tracking something real, even if the exact word doesn’t stick.
And it lines up almost perfectly with what this blog has been watching from the ground floor. The whole reason I keep coming back to Physical AI — robots moving onto the factory floor, Google DeepMind’s humanoid control models, analog chips running robot brains on microwatts — is that the value is migrating from bits to atoms. Same with the compute-foundation stories: why Penn is trying to solve AI’s energy problem with light, why OpenAI cutting GPT-5.6 prices 80% was really a story about inference economics, why a model a quarter the size matters more than a model that’s a little smarter. Those weren’t random tech stories. They were early frames of the same picture the VCs are now pricing.
If there’s one thing to take away, it’s this: the cost of copying software went to zero, so the value ran to the things that can’t be copied for free — energy, silicon, robots, the physical base under the digital world. That’s a slower, harder, more capital-hungry kind of company than the app era rewarded. Whether “civilization technology” is the name that sticks, I don’t know. But the direction it points — toward atoms, toward the foundation — is the direction I’d keep watching. And, worth saying plainly: a lot of today’s numbers are bets, not results. The map has shifted. The territory hasn’t been fully walked yet.
This article is for informational purposes only and is not investment advice.
Photo: İsmail Enes Ayhan / Unsplash
댓글 남기기