Walk into any tech incubator at two o'clock in the morning, and you will hear the exact same sound. It is not the hum of the servers, nor is it the click of mechanical keyboards. It is the rhythmic, collective sigh of twenty-somethings staring at monitors, waiting for a piece of software to compile. For twenty-five years, that glow was the warmest place in the global economy. If you could write code, you could build a kingdom. If you owned the platform where the code ran, you owned the world.
We built an entire civilization on the assumption that this trajectory was permanent. Software was eating the world, we were told, and the appetites of the giants—Microsoft, Alphabet, Apple, Meta—were infinite.
Then James Anderson stood up and turned off the lights.
Anderson is not a sensationalist blogger looking for clicks. He is the quiet investor who spent decades at Baillie Gifford, anchoring the early, lonely bets on Amazon and Tesla when the rest of Wall Street called them reckless gambles. He watched the software era rise from its cradle. Now, he is announcing its funeral.
His thesis is simple, brutal, and entirely disruptive to everything we think we know about the future: the age of pure software dominance is over. The trillions of dollars minted by writing lines of code and scaling them across the cloud have hit a wall.
To understand why, we have to look past the stock charts and peer into the physical reality of what comes next.
The Mirage of the Infinite Cloud
We were promised that software was weightless. That was the great lie of the Silicon Valley boom. We believed that a company could grow to a valuation of three trillion dollars while remaining essentially ethereal. You write a program once, you duplicate it a billion times for zero marginal cost, and you collect the rent.
Consider a hypothetical engineer named Sarah. In 2015, Sarah could launch an app from her bedroom using a standard cloud computing credit. If her user base doubled overnight, the infrastructure scaled automatically. The physical world—the copper wires, the cooling fans, the concrete warehouses—was someone else's problem. Sarah’s only constraint was the elegance of her logic.
That era felt like magic. It minted a class of billionaires who believed they were artists, unburdened by the gritty laws of physics.
But the magic trick depended on a hidden variable. Software only felt weightless because it was riding on the back of hardware that was predictable, cheap, and impossibly efficient. Moore’s Law guaranteed that chips would get faster and cheaper every two years like clockwork. The software companies took that charity and ran with it, writing increasingly bloated code because they knew the hardware would always bail them out.
Now, the hardware is fighting back.
The shift from traditional software to artificial intelligence has shattered the illusion of weightlessness. AI is not software in the traditional sense. It does not run smoothly on standard architecture. It devours energy. It demands bespoke, wildly expensive silicon. It requires infrastructure so massive that it can no longer be hidden behind the clean, abstract metaphor of "the cloud."
When Anderson points out that the software era is ending, he is pointing at the balance sheets of the giants. The capital expenditure required just to stay in the game is no longer measured in millions, or even billions. We are talking about hundreds of billions of dollars poured into physical infrastructure. Power grids are buckling under the strain of data centers. Rivers are being diverted to cool the clusters.
The weightless world is suddenly very, very heavy.
The Iron Trap of the Microchip
The power dynamic has flipped completely. For decades, the chipmakers were the servants, and the software platforms were the masters. Today, the masters are begging for scraps at the table of a single company in Taiwan.
Without massive, specialized hardware, the grandest AI models are just expensive statistics. The companies that won the software wars by being nimble, asset-light, and hyper-scalable are suddenly forced to behave like twentieth-century industrial titans. They are building power plants. They are securing supply chains for rare earth metals. They are terrified of shipping delays.
This is the irony that Anderson identifies. The very companies that achieved godlike status by escaping the constraints of the physical world are now trapped by them.
Let us look at the math, because the math is unforgiving. To train a next-generation frontier model requires computational power that doubles every few months. The software efficiency gains cannot keep pace with this exponential demand. If you want a smarter model, you cannot simply write better code; you must build a bigger warehouse of chips. You must consume more megawatts.
This means the profit margins that made Big Tech the darling of the financial world are under existential threat. The days of 80% gross margins on pure software delivery are evaporating. In their place is a grinding, capital-intensive arms race where the winner is determined not by who has the cleverest algorithm, but by who can secure the most electricity.
The Human Cost of the Transition
What happens to the people who grew up inside the old system?
For a generation, the career advice given to every ambitious youth was identical: learn to code. It was the ultimate leverage. It was a golden ticket to the upper middle class, a shield against outsourcing, a guarantee of relevance.
But the end of the software era means the devaluation of the pure coder. When the baseline architecture shifts from writing explicit instructions to training neural networks that write their own code, the premium on standard software engineering drops precipitously. The industry is already feeling the tremors. The rolling layoffs across Silicon Valley are not just a temporary post-pandemic hangover. They are a structural realignment.
The system is shedding weight because the nature of the work has changed.
The tragedy is that we are still training thousands of young people for a world that ceased to exist while they were midway through their degrees. They are entering a market where the entry-level software job is being squeezed out from two sides: above by the crippling capital costs of infrastructure, and below by automation tools that can generate standard React components in three seconds.
The stakes are not merely financial. They are deeply cultural. The tech industry fostered a specific mythology of the solo founder—the college dropout who changes the world with nothing but a laptop and a vision. That myth is dying. You cannot disrupt an industry from a dorm room if the entry fee for the necessary computational power is fifty billion dollars. The gatekeepers have returned, and their walls are made of steel and silicon, not code.
The Capital Realignment
Investors are notoriously slow to accept that the party is over. They tend to dance until the music has stopped and the janitors are sweeping the floor. For the past few years, the market has treated AI as simply the next chapter in the software playbook. Valuations were calculated using the old software metrics.
Anderson’s warning is a bucket of ice water to that complacency.
If the software era is over, the investment strategies that worked from 2010 to 2021 are worse than useless; they are dangerous. Capital can no longer look for the next asset-light SaaS company that can scale to infinity on a shoestring budget. Those companies are becoming features, not platforms. They are built on top of foundations they do not own and cannot afford to maintain if the providers raise prices.
The real value is migrating elsewhere. It is moving down the stack to the material world. It is moving toward the energy providers, the thermal management engineers, the advanced optics manufacturers, and the architects of specialized hardware.
This transition is terrifying for Wall Street because it requires patience. Software could be built, tested, and shipped in a weekend. A semiconductor fabrication plant takes five years and twenty billion dollars before it outputs a single wafer. The feedback loops are slowing down. The velocity of innovation is hitting the hard ceiling of physical construction.
The Final Chord
We are left standing on a shoreline, watching the tide go out on an era that defined the modern world. The software boom gave us unimaginable convenience, connected the globe in a chaotic web of instantaneous communication, and concentrated wealth on a scale unseen since the Gilded Age. It felt permanent. It felt like the destination of human progress.
But history does not stop. It merely shifts its weight.
The glowing screen is no longer the frontier. The frontier has moved into the dark, humming halls of northern Virginia, the sun-baked plains of Iowa, and the tightly guarded cleanrooms of Hsinchu, where massive turbines spin and transformers groan under the load of a new epoch. The programmers will still write their lines, but they will do so as custodians of a machine they do not truly control, operating within limits set not by human imagination, but by the stubborn, unyielding laws of thermodynamics.
