SoftBank's Latest Quest to Lose Money With Style

When Masayoshi Son gets that gleam in his eye, you know billions are about to be allocated with the precision of a drunk archer. After the spectacular flameouts of WeWork, Uber (which only became profitable a decade later), and a graveyard of robot-themed startups, SoftBank is back for more. This time, it's Skild AI—a company promising to build the "foundational model" for robotics. Because what the world needs, apparently, is a $14 billion software layer for machines that still can't reliably fold a towel.

The "General Purpose" Promise: A Euphemism for "We Don't Know Either"

"General-purpose robotic software" is the new "synergy"—a phrase that sounds impressive but is utterly meaningless. In the tech industry, "general purpose" translates to: "We haven't figured out what specific problem to solve, so we're going to solve all of them, poorly." It's the business model equivalent of showing up to a potluck with a bag of flour, eggs, and sugar, declaring you've brought the "foundational ingredients for dessert."

Remember Boston Dynamics? Their robots do backflips and parkour for viral videos, while in real factories, the most advanced "robotics" is often a glorified conveyor belt. Skild's bet is that a single, massive AI model can teach a robot to do anything, from welding a car to making a soufflé. The only catch is that this requires the model to understand physics, causality, material science, and common sense—things that even the latest LLMs struggle with in a purely digital world. Adding gravity and friction to the mix should go smoothly.

The Valuation: Where Math Meets Magic

A $14 billion valuation for a pre-revenue, pre-product robotics software company isn't just optimistic; it's a form of abstract art. Let's put this in perspective. This values Skild AI at roughly:

• 3x the market cap of BlackBerry.
• Half the value of General Motors' entire Cruise self-driving unit (which has actual cars on the road, however poorly they perform).
• Enough money to give every one of their employees a $280 million bonus (assuming they have 50 people).

This valuation isn't based on discounted cash flows; it's based on discounted reality flows. It assumes a future where every home has a helpful robot butler, every factory is fully automated, and Skild's software is the brains inside all of them. It's a future that's been "five years away" for the past fifty years.

The Real Product: The Hype Cycle Itself

Let's be honest. The first product Skild AI has successfully shipped is a compelling funding narrative. The ingredients are classic:

1. The Foundational Model Buzzword: Attach your startup to the one thing (LLMs) that VCs are currently throwing money at.
2. The Moonshot Narrative: "We're not building a feature; we're building the platform for the next computing paradigm." This justifies any valuation.
3. The Talent Trifecta: Founders are ex-CMU and Meta AI. In tech, a pedigree is often more valuable than a prototype.

The actual work—making a robot understand that a glass is fragile, that a door must be pulled not pushed, that a cat should not be used as a doorstop—is left for the engineers to figure out after the champagne cork pops at the funding announcement.

What's Next? A Symphony of Stumbling

So what does $1.4 billion buy you in robotics? It buys you time. Time to hire hundreds of the world's best AI researchers. Time to buy massive amounts of GPU compute to train models. Time to build and break thousands of expensive robot bodies. And, most importantly, time to weather the inevitable "trough of disillusionment" when everyone realizes that general-purpose AI in the physical world is harder than making a chatbot sound like a therapist.

The roadmap is predictable: years of technical papers, carefully curated demo videos (filmed under perfect, controlled conditions), partnerships with legacy manufacturers desperate for a AI story, and a gradual pivot towards a narrower, actually useful application when the "general purpose" dream meets the hard floor (repeatedly).