Short Bits Go Long On Atoms

For the best part of two decades, the smartest money in technology went into software. And why wouldn’t it? Software scales infinitely, costs almost nothing to copy, and once you’ve built it, throws off recurring revenue while you sleep. Hardware was for mugs. Atoms were heavy, slow, and expensive. Bits were weightless and wildly profitable. That was the orthodoxy. It held for fifteen years. It is now falling apart, and what comes next could be genuinely exciting. AI Just Killed the Software Premium Here’s what changed. Artificial intelligence can now write competent code. It can build applications, wire up integrations, and automate workflows. It can do the work that justified thousands of SaaS companies charging per-seat licence fees to millions of businesses. That matters enormously because when AI can replicate what your platform does, your moat isn’t a moat anymore. It’s a puddle. Companies across the technology sector are already shifting focus to delivering real value from AI rather than selling the tools to build it. The repricing is already underway. Software valuations have been sliding for two years, and the drawdown isn’t just about interest rates. The market has started to figure out that a huge amount of application-layer software, the middleware, the connectors, the products whose entire pitch was “we saved you the trouble of building this”, is being commoditised faster than anyone expected. The trouble is disappearing. The middle layer of the technology stack is being hollowed out from the inside. What survives sits at the extremes. At the top, frontier AI companies command enormous valuations because they own the intelligence itself. At the bottom, infrastructure endures. Databases, compute platforms, orchestration layers. The plumbing still matters. But everything sandwiched between those two layers is under real pressure, and investors still clinging to the old SaaS playbook of recurring revenue, high gross margins, and minimal capital intensity are holding a position that’s deteriorating by the quarter. What happens to the software market when AI can create almost limitless applications tailored to your needs? The Things AI Cannot Build From a Prompt So where does value go next? Down. Into the physical world. Into atoms. Into the things that are genuinely hard to replicate. Chips. Materials. Fabrication. Energy. Quantum hardware. Robotics. These are domains where scarcity is real, where barriers to entry are measured in billions of dollars and decades of accumulated know-how, and where the laws of physics set limits that no amount of clever software can get around. This isn’t nostalgia for the industrial age. It’s a recognition of something quite specific. AI, by solving so many problems in the digital domain, has simultaneously made the physical domain relatively more valuable. When software is abundant and cheap, hardware becomes the bottleneck. When anyone can build an application, what matters is what you can build in the real world. The constraint has moved. It used to be code. Now it’s concrete, silicon, copper, and kilowatt hours. We’re watching this play out in real time. NVIDIA built its entire CES 2026 narrative around “physical AI“. Investment in humanoid robotics hit $3.2 billion in 2025, more than the previous six years combined. The race to build energy infrastructure for AI data centres has created demand curves that look like the industrial buildout of the mid-twentieth century. Semiconductor fabrication has become a matter of national security. Private quantum firms like PsiQuantum are raising $1 billion rounds to build fabrication-scale quantum hardware. These aren’t disconnected stories. They’re different chapters of the same book. Capital is migrating from bits to atoms. Building hard stuff. Is the building of stuff with atoms the real defence against the rise of AI? Defensive industries look to be more in vogue as funds rotate away from software and SaaS investments. The Irony is Exquisite There’s something almost poetic about it. AI, the ultimate expression of the software era, is creating the conditions for hardware’s comeback. Every new model needs more compute. Every compute expansion needs more chips, more power, more cooling, more physical infrastructure. The intelligence lives in software, but the substrate is stubbornly, irreducibly physical. You can’t run a data centre on tokens. You run it on turbines. And as Landauer showed us decades ago, even erasing a single bit of information has a real thermodynamic cost. Information was always physical. We just forgot for a while. For investors looking at public markets, the rotation points in some clear directions. Semiconductors across the entire stack, from design through fabrication to equipment, are obvious beneficiaries. Energy infrastructure, particularly nuclear and next-generation power, is being accelerated by demand that didn’t exist five years ago. Robotics offers exposure through established players and through the hyperscalers now building physical AI platforms at scale. The quantum ecosystem alone now spans over 1000 companies worldwide, all building physical technology that can’t be replicated in software. Quantum computing sits at the sharp end of this thesis. Companies like IonQ, Rigetti, and D-Wave are building hardware that operates at the edge of what physics permits. Each has taken a distinct path to get there. Their competitive advantages aren’t lines of code. They can manipulate individual atoms, ions, and photons with extraordinary precision. These are moats that no language model can replicate, built on decades of scientific work that can’t be shortcut. Venture capital is flooding into the sector, and as the technology matures and recent progress in error correction suggests the timeline is compressing faster than the sceptics expected, the value creation will be substantial precisely because so few organisations on Earth can do what these companies do. Early investors in quantum stocks have already seen what happens when the market starts to price in that scarcity.

The Advertising Model Dies When Nobody is Clicking There’s another dimension to this that rarely gets discussed, and it might be the most consequential of all. A huge proportion of the software economy was never really about building useful things. It was about monetising attention. Optimising funnels. Getting people to click affiliate links. Squeezing revenue out of ad impressions and conversion rates. An enormous amount of human talent, people with engineering degrees and genuine ability, spent their careers figuring out how to make a button slightly more clickable or a notification slightly more addictive. That era is ending, and honestly, not a moment too soon. As AI agents increasingly interact with services on our behalf, the entire ad-supported model starts to unravel. Agents don’t see banner ads. They don’t click affiliate links. They don’t get nudged by push notifications or manipulated by dark patterns. When your AI is booking your flights, comparing your insurance, and managing your subscriptions, there’s no eyeball to monetise. The advertising surface that funded so much of the software economy simply ceases to exist. In a world where agents talk to agents, nobody is watching the screen. That’s genuinely good news. It means the low-hanging fruit of the attention economy dries up, and many smart people who might otherwise have spent their careers A/B testing checkout flows will need to find something else to do. Some of them won’t be happy about it. But the net effect is that talent is redirected toward what actually matters. Energy systems. Materials science. Fabrication. Neuromorphic hardware. Brain-inspired computing. Quantum-enhanced energy grids. The hard, fundamental, sustainable work of building things in the physical world. There was a time, not that long ago, when engineering meant making something real. We might be heading back there. Ad revenue drives much of the internet, aside from subscription services. What happens when human attention is no longer captured, and instead, a flurry of agents does the interaction? Is the attention economy about to die? The Picks and Shovels Are Physical This Time The practical investment takeaway isn’t complicated. Within technology, value is moving from the application layer to the physical layer. The picks-and-shovels trade of the AI era isn’t software tools for building AI products. It’s the physical infrastructure that AI needs to exist and the physical systems that AI will eventually operate. Governments are already betting heavily on this thesis, with billions flowing into quantum research at the intersection of energy, materials science, and computation, and bipartisan legislation committing $2.5 billion to quantum R&D at the Department of Energy alone. Think about the early internet. The biggest long-term winners weren’t the companies building websites. They were the companies that built the fibre-optics, servers, and networking gear that made the web possible. We’re likely at a similar inflexion. The AI application layer will produce some spectacular winners, but the durable, compounding returns may belong to the companies building the physical world that AI inhabits. The technology industry spent two decades convinced that software would eat everything. It did. And now AI is eating software. What remains, what can’t be eaten, what has to be engineered atom by atom in clean rooms, cryostats, and fabrication plants, is where the next generation of value is created. The trade is in the title. Short bits, long atoms. Disclaimer: This article is for informational and educational purposes only and does not constitute financial advice, investment advice, trading advice, or any other type of advice. The information provided should not be relied upon for making investment decisions. Always conduct your own research and consult with a qualified financial advisor before making any investment decisions. Past performance does not guarantee future results. Investing in stocks, particularly in emerging technology sectors like quantum computing, involves substantial risk including the potential loss of principal. Tags: