I Said We Needed a Billion Robots Last Year. Goldman Just Put a Price Tag On It.
- Rich Washburn
- 23 minutes ago
- 5 min read
Fourteen months ago I wrote a piece called "Why We NEED a Billion Robots." The thesis was simple: the AI conversation was obsessed with the wrong half of the workforce. Everyone was watching chatbots write emails while ignoring the fact that nobody had figured out how to automate a plumber, an electrician, or a warehouse picker. I argued that solving machine intelligence was the easy part — manufacturing that intelligence into a body that could actually do physical work was the real, decades-long bottleneck. Goldman Sachs just published a report confirming it, then attached a number to it that's bigger than most countries' GDP.
The Number That Reframes Everything
Roughly $7.6 trillion will be invested globally in AI infrastructure between 2026 and 2031 — compute, data centers, power, the whole build-out everyone's been arguing about for two years. That's real money. It's also, according to Goldman, the wrong thing to be impressed by.
Here's the sentence that matters: software currently accounts for less than 0.5% of global GDP. Mark Sorrell, Goldman's global head of industrials, told Axios flatly that we're "only just beginning" to see AI's impact on industrial businesses. Read that again. Every large language model, every AI startup, every SaaS company riding this boom — collectively, that's less than half a percent of the world's economic output. Which leaves 99.5% of the global economy sitting untouched by everything we've spent the last three years talking about.
That's not a rounding error. That's the entire thesis. The AI industry has spent its first act disrupting the cheapest, fastest layer to disrupt — software, because it's pure code, no factories required. Goldman's argument is that act two is the physical economy: factories, mines, utilities, oil rigs. The stuff that has to be built, wired, and physically operated, not just deployed to a server.
Why the Bottleneck Was Never Intelligence
I made this argument in May 2025 and nothing about it has changed, it's just gotten more expensive to ignore. The benchmark saturation of cognitive tasks — math, logic, code, language — was already well underway a year ago. Digital agents doing knowledge work faster and cheaper than humans was never the hard problem. The hard problem was this: replacing skilled manual labor globally requires something like 400 million to 1.5 billion humanoid robots, and building a robot is not like shipping a software update.
The actual constraint is actuators — the "muscles" of a robot, the servos and motors and linear drives, most of which need rare earth magnets. It's batteries, which need better chemistry and more of those same rare earths. It's sensors. It's the fact that global industrial robot production runs around 500,000 units a year, and even scaling that at 10x every three years — a growth curve faster than TVs, cars, or smartphones ever managed — doesn't get you to a billion units until sometime between the 2040s and 2060s. Goldman's new numbers don't contradict any of that. They confirm it, and then show you the capital response to it.
Goldman's Real Argument Isn't About Robots. It's About Money.
Here's what's genuinely interesting about Goldman's report, "Harnessing AI for the Real Economy" — it's not really a capability story. It's a capital structure story. The report spends most of its time on how you finance an industrial transformation this large, not whether the technology works.
Bringing a single 250-megawatt AI data center online costs roughly $12 billion, all-in, equipment included. That number alone explains why Goldman is building out new financing structures — private credit, structured credit instruments, sovereign capital — because traditional balance-sheet financing was never built to fund infrastructure at this scale. Their equity research separately estimates $5.3 trillion in combined data center capex from large tech companies between 2025 and 2030. Data center power consumption is projected to rise 165% from 2023 to 2030. The build-out of the digital layer alone requires financial engineering nobody needed for the last several tech cycles.
Now overlay the physical layer on top of that. Goldman's own humanoid robotics forecast just got revised upward six times over — from a $6 billion 2035 projection to $38 billion. Tesla is converting Fremont production lines, ending Model S and X manufacturing, specifically to scale Optimus. Figure AI, backed by OpenAI, Microsoft, and NVIDIA among others, is pushing toward five-figure unit deployments. NVIDIA's Jensen Huang has been saying the quiet part loud for a while now: "every industrial company will become a robotics company."
None of these are small bets. They're bets that the capital, not the intelligence, is the current constraint — which is exactly where I left the argument fourteen months ago, just without the trillion-dollar backing.
The Demographic Math Nobody's Talking About
There's a second force compounding all of this that has nothing to do with AI hype cycles: the people who currently do this physical work are retiring.
Twenty-six percent of US manufacturing workers are over 55. Only 8% are between 16 and 24 — against a national average of 12% for that age cohort across all industries. This isn't a future problem. It's a current one. The physical economy needs automation not because AI made it fashionable, but because the humans who've been running factories, utilities, and industrial operations for the last thirty years are aging out faster than anyone is replacing them. That's the part of this story that makes it inevitable rather than speculative. Even if the AI hype cycle cooled off tomorrow, the labor math alone would force this capital toward physical automation.
Where This Actually Goes
The agentic AI conversation — the loops, the orchestration, the self-improving scaffolds I've written about recently — is the proof that the cognitive layer is basically solved. Models that can plan, execute, correct themselves, and operate with minimal supervision are no longer the hard problem. They're the finished ingredient sitting on the shelf, waiting for a body.
What Goldman's report actually documents is the moment the money figured that out too. $7.6 trillion isn't being spent because anyone doubts AI can think. It's being spent because the industry now has to solve manufacturing scale, materials science, and capital structure — problems that don't yield to a software update, no matter how good the model behind it gets. I said this would take thirty to fifty years, not two. That hasn't changed. What's changed is that the largest bank on Wall Street just agreed the other 99.5% of the economy is where this goes next, and started building the financial infrastructure to fund it.
The robots were never the hard part. The money finally showing up is.
Rich Washburn is a technologist and strategist working at the intersection of AI, infrastructure, and capital. He is Managing Partner and Chief AI Officer at Eliakim Capital and CIO of Data Power Supply.