Discover why startups are shifting to hard tech and physical products. Learn about robotics, defense, semiconductors, and AI-driven growth trends reshaping e...
Startups Moving From Bits to Atoms: The Hard Tech Revolution
Key Takeaways
- Hard tech companies have tripled in YC batches: Increasing from 8% to 20% of accepted companies, with robotics, industrial manufacturing, and defense leading the surge
- Median YC startup revenue jumped 150%: Companies now reach $20,000 monthly recurring revenue by batch end, up from $8,000 just one year ago
- Experienced founders are thriving: Solo founders now represent 18-19% of batches (up from 5%), while founders in their 40s and 50s are building the most powerful companies
- AI is accelerating physical innovation: Code generation and advanced models are removing software talent bottlenecks, making hard tech startups viable for smaller teams
- Three macro trends are driving growth: SpaceX's success inspired space startups, geopolitical defense needs energized defense tech, and AI demand created semiconductor and power infrastructure opportunities
The Shift From Software to Atoms
For years, venture capital bet heavily on software-as-a-service (SaaS) companies. But the landscape has fundamentally changed. At YC, the data tells a striking story: hard tech companies—those building physical products that "touch atoms, not just bits"—have surged from 8% to 20% of accepted batches in just 18 months.
This isn't a minor trend. The growth reflects a profound realization: AI and modern tooling have made physical innovation more accessible than ever. Founders no longer need massive teams or unlimited capital to compete in robotics, semiconductors, or defense technology. The bottleneck that once made hard tech impossible for startups has dissolved.
The Four Pillars of Hard Tech Growth
Robotics and Industrial Manufacturing
Robotics has jumped from 1% to 6-7% of YC batches. Industrial manufacturing—"building things back in the US"—has grown even faster, from 4% to 10%. Companies like Boost Robotics are building specialized robots for data centers, while others are creating broader automation infrastructure. The industry senses a "ChatGPT moment" for robotics is near, with breakthroughs like Physical Intelligence's Astra demonstrating significant leaps in robot capabilities.
Defense and Dual-Use Technology
Defense startups have grown from 1.5% to 5% of batches. Examples include Icarus, which is building solar-powered drones for overwatch and communications, and 9 Mothers, creating anti-drone defense systems. Beyond comprehensive solutions to governments, a growing category of dual-use startups serves both private and government sectors. Companies like Nox Metals are rebuilding America's metal supply chain in places like Detroit, addressing critical infrastructure gaps that large defense contractors had ignored.
Semiconductors and Compute Infrastructure
The AI explosion has created urgent demand for compute. YC companies are building the full semiconductor stack—from custom hardware architectures to optical switches. Baud is developing hardware using ternary representation for models, recognizing that LLM architectures don't require full floating-point precision. Dipole Labs is addressing data center bottlenecks with fully optical switches that enable faster GPU communication using photons. Semiconductor and photonics companies have grown from 1% to nearly 4% of batches.
Power and Energy Infrastructure
As data centers proliferate, power becomes critical. Companies are building power infrastructure solutions, from battery systems to energy platforms. This category has grown from 1% to nearly 3% of batches, reflecting the foundational role electricity plays in scaling AI.
Why AI Actually Helps Hard Tech
A common misconception: AI is only good for software startups. The opposite is true. Advanced AI models are removing the software engineering bottleneck that once made hard tech impossible.
Historically, building hardware required hiring dozens of elite software engineers—a luxury only well-funded companies could afford. Palmer Luckey noted this challenge when building Anduril. But code generation changes everything. Modern models can now generate the complex software that hardware companies need, meaning teams can stay lean.
Experienced founders with AI expertise are particularly well-positioned. They understand the domain deeply and can now use AI to handle implementation details that once consumed months. This efficiency translates directly to revenue growth.
The Revenue Acceleration Phenomenon
YC released a remarkable statistic: the median company entering the batch has zero revenue, but by batch end, it reaches $20,000 monthly recurring revenue—up from $8,000 just one year ago.
A significant portion of this growth comes from companies building full-stack, end-to-end solutions where AI agents actually complete jobs instead of just providing tools. Insurance brokering, medical billing, clinical intake—these are genuinely difficult problems that old SaaS couldn't solve. But with agentic coding, they're now possible.
Some companies are moving from zero to seven figures in revenue during the three-month batch—a milestone that once took 18 months. This acceleration is possible because founders are building more mature products faster, running dozens of coding agent sessions to achieve product depth that would have required large teams months ago.
Data as the New Gold: The Hidden Growth Story
One of the least visible but fastest-growing categories is companies selling data and reinforcement learning environments to AI labs. In just the last two years, YC has funded more than a dozen companies in this space, each generating over $10 million annually—and several generating hundreds of millions.
Companies like AfterQuery and DataCurve represent a massive market that most founders don't know exists. The big labs are reportedly spending about a billion dollars on reinforcement learning environments. This is specialized work—RL environments for finance, egocentric data for robotics, custom datasets for specific use cases—but the business is real and growing rapidly.
The Rise of the Experienced Solo Founder
Conventional wisdom said founders needed co-founders to succeed. But the data has shifted dramatically. Solo founders now represent 18-19% of YC batches, up from just 5% one year ago.
Why? AI has changed what's valuable. Previously, founders needed a complementary skill combo: a great hustler paired with a world-class technologist. Now, knowing what to build and how to prompt AI is more valuable than the ability to code everything yourself. A founder with domain expertise and strong opinions can now move alone—initially.
Interestingly, experienced founders—those in their 40s and 50s—are performing exceptionally well. Founders like Peter Steinberger, who brings decades of engineering management experience, adapt naturally to managing coding agents. They understand problems deeply, have lived through multiple business cycles, and possess unusual leverage in the current environment.
Many successful solo-founded companies do eventually bring on co-founders, but after gaining momentum. This differs from traditional 50/50 co-founder arrangements and reflects how startups now scale.
Practical Advice for Builders Today
For anyone considering starting a company now: just start prompting. The moment you spin up a new model and realize things that weren't working a month ago suddenly work is profound. Agentic coding has reached a threshold where founders can move faster than ever before.
This applies whether you're building hardware, defense tech, or AI-powered workflows. The modern founder's superpower isn't knowing how to code—it's knowing what to build and understanding the problem deeply enough to guide AI in the right direction.
Conclusion
We're witnessing a historical inflection point. For a decade, software ate the world. Now, atoms are back—and they're building faster than before. The combination of AI-powered tooling, experienced founders, and urgent real-world problems has created a moment where hard tech startups can move at software speeds.
If you've been waiting for permission to build something physical, to solve infrastructure problems, or to compete in defense and robotics—the barriers have never been lower. The question isn't whether it's possible. It's what are you going to build?
Original source: Startups Are Moving From Bits to Atoms
powered by osmu.app