A startup in Munich a few months ago was constructing modular cooling systems for AI data centers in an old industrial building. A custom heat exchanger the size of a dinner table was indicated by the founder, a former aerospace engineer. He declared, “This is the bottleneck.” “Not graphics processing units. not the bandwidth. Metal, heat, and power.
Software dominated the investment scene for many years. Startups received funding on the basis of their promises of quick distribution, clean interfaces, and infinite scalability. Pitching was simple. Millions could be raised with a good deck and a few engineers. However, that edge has waned recently. Applications are now remarkably efficient—sometimes too efficient—thanks to generative AI. A high school student can now replicate last week’s popular app using ChatGPT and a Figma template.
| Factor | Details |
|---|---|
| VC Realignment | Investors are increasingly backing physical infrastructure, hardware, and manufacturing |
| Why It’s Happening | Software has become commoditized; hard tech creates defensible moats |
| Key Sectors | Semiconductors, robotics, AI hardware, medical devices, defense, climate tech |
| Investment Surge | Deep tech and hardware now account for ~30% of global VC activity |
| Strategic Advantage | Hardware + software integration is seen as particularly innovative and highly defensible |
| Market Pressure | AI infrastructure demands drive interest in power delivery, cooling, and fabrication tech |
When anyone can construct anything, what will happen? Venture capital seeks out things that cannot be replicated.
Long written off as slow, costly, and difficult to scale, the physical economy is turning out to be remarkably resilient. And more and more attractive. Once thought of as a bug, friction is now being reconsidered as a feature. The trend of scarcity has returned.
We no longer fund slide decks, a Zurich fund manager told me bluntly during a recent conversation. We provide funding for machines. She wasn’t using metaphors in her speech. Her company recently invested in a robotics company that makes high-precision arms for micro-manufacturing, which involves micron-level calibration. She remarked, “It’s built to last, but it’s not hype-proof.”
Venture capitalists have become exceptionally adept at identifying software patterns and consumer trends over the last ten years. They are now at a disadvantage because of that pattern-matching. It’s hard to tell who is creating something unique when every transaction resembles the last unicorn. Fundamentally, the return to hardware is a return to substance.
Hardware is now necessary rather than optional, especially when it comes to AI infrastructure. Power grids need to be upgraded, cooling systems need to be redesigned, and chip packaging is becoming more complicated than the chips themselves due to the demand for high-performance compute. By 2030, infrastructure to support AI needs is expected to cost $5.2 trillion, according to McKinsey. It is not a theoretical number. It is necessary for the market to operate.
The definition of a “startup” is already changing as a result of this change. For predictive maintenance, a young team in Helsinki is developing electric drivetrain prototypes with integrated sensors. A medtech entrepreneur in Boston is developing wearable glucose monitors that use body heat to self-calibrate. These projects aren’t for a science fair. These are early-stage businesses that are raising large sums of money because they address issues with actual material constraints rather than because they are ostentatious.
Additionally, the economics have significantly improved, even though the adage “hardware is hard” is still frequently used in pitch meetings. The modularity of supply chains has increased. Surprisingly, prototyping is now inexpensive. Additionally, hardware margins have significantly increased in many industries while software margins are shrinking.
The longevity of these businesses is especially advantageous. Anything that takes five years to construct usually doesn’t vanish overnight. Additionally, it is extremely difficult to replicate when factory-level integration, specialized talent, and custom materials are involved. That is more important now than it was in the past for investors seeking long-term value.
I met a founder in Lisbon earlier this year, and he told me about his experience creating a waterless textile dyeing machine. Just getting the pressure system stable under test conditions took him two years. He chuckled and said, “We would have been on version 19 by now if I were doing SaaS.” “No one else can do what we’ve done—because no one else wanted to take the risk,” he continued.
It’s becoming more typical to have that quiet confidence.
Major tech companies have also embraced this reality. Nvidia’s control over manufacturing, distribution, and chip-level design, rather than software, is what secures their position. Tesla won by creating what Elon Musk referred to as “the machine that builds the machine,” not just by writing autopilot code. From chips to case design, hardware integration has always been Apple’s long-term advantage.
Businesses are using manufacturing capabilities to build platforms that gather data, enable edge AI, and produce stickiness that is impossible to remove with a browser plugin. This convergence is especially creative because it combines the durability of carbon fiber, steel, and copper with the adaptability of code.
Venture timelines have also changed as a result of this change. Once allergic to slow burn rates, investors are reconsidering what sustainable growth entails. Top-tier funding is now drawn to deep tech and dual-use technologies, particularly in industrial AI and defense. These are machine shops with government contracts and long order books, not moonshots.
Furthermore, it goes beyond geopolitical pressure. The demand is structural. Automated logistics, edge computing, and cooling technologies are no longer fringe industries. They are rapidly expanding and represent the next layer of infrastructure.
“If it takes power to run and metal to make, we want in,” an investor told me last week.
The founders of tomorrow are paying attention. Supply chain thinking is something they are learning how to incorporate into their product design. In addition to backend developers, they are hiring mechanical engineers. Additionally, they want to know how to create something incredibly durable as well as how to scale.
Software has not been abandoned by venture capital; rather, it is no longer viewed as the sole source of advancement. A hybrid model is emerging in which hardware is enhanced by software and physical products serve as platforms for intelligent systems.
There is no trend here. It’s a return to the origins of innovation.
