Why We Were Wrong About Humanoids.
We thought humanoid investment would trail off as product-market fit stayed elusive. The product question was reasonable. The capital-markets call was wrong.
In brief
We were wrong because we treated product-market fit as the near-term constraint on humanoid investment. Investors instead funded the option that foundation models could turn embodied AI into a platform: humanoid funding roughly doubled to $3 billion in 2025, and Figure alone raised over $1 billion at a $39 billion valuation.
The Prediction
Investment in humanoids trails off as companies struggle to achieve product-market fit.
What was the original call?
The skepticism was about deployment. Humanoid robots had impressive demos, expensive hardware, difficult integration, and uncertain near-term product-market fit. A UK government technology assessment reached a similarly cautious conclusion on technical maturity, applications, and public acceptance.
That deployment skepticism may still prove directionally right. The mistake was assuming that product-market fit would be the near-term constraint on capital.
What actually happened?
Instead of trailing off, humanoid investment accelerated. The State of AI scorecard marks the prediction as a miss: roughly $3 billion was invested in humanoids in 2025, up from $1.4 billion in 2024. The IEEE Robotics and Automation Society's 2025 funding report documents the wider surge in robotics capital.
One company carries most of the story. In September 2025, Figure closed more than $1 billion of Series C capital at a $39 billion post-money valuation, backed by NVIDIA, Intel Capital, Qualcomm Ventures, Brookfield, and Salesforce. Nineteen months earlier, its Series B had valued the company at $2.6 billion. A 15x markup between rounds, in a category we said would cool.
Investors were not only underwriting current deployments. They were buying option value on embodied AI becoming a platform category.
Why did we get it wrong?
We overweighted product-market fit and underweighted narrative gravity. Foundation models made general-purpose robotics feel newly plausible, and capital became willing to finance the bridge from demos to deployment.
In an AI platform cycle, financing can precede evidence by years. Sometimes that is exuberance. Sometimes it is the only way a hardware category with factories, data collection, and long integration cycles gets built. Our prediction collapsed those possibilities into a single near-term funding call.
Why does the miss matter?
The miss separates two questions that we treated as one: will humanoids work commercially, and will capital keep funding the attempt?
Those questions can have different answers for several years. A useful robotics forecast must specify whether it concerns technical capability, repeat deployment, unit economics, or financing appetite. This failure mode, being right about the product and wrong about the capital, shows up elsewhere in our track record too; see the accuracy analysis for the pattern.
What we are watching now
The evidence that matters now is repeat deployment beyond pilots: renewal, utilization, intervention rates, maintenance burden, safety, and integration cost.
We are also watching whether foundation-model progress translates into robust manipulation and autonomy in messy environments. Capital answered one question. Customers still have to answer the other. If Figure's valuation is right, the deployment data will eventually have to show it.
Frequently Asked Questions
How much was invested in humanoid robots in 2025?
Roughly $3 billion, up from $1.4 billion in 2024, per the IEEE Robotics and Automation Society's global funding report. Figure's Series C alone exceeded $1 billion at a $39 billion post-money valuation.
Why did the State of AI Report predict humanoid investment would trail off?
Because deployment evidence was thin: expensive hardware, hard integration, and no repeatable product-market fit. The deployment read may still prove right; the error was assuming capital would wait for the evidence.
Was the humanoid skepticism wrong?
The funding call was wrong, and we graded it a miss. The commercial question is still open: valuations now assume embodied AI becomes a platform, and the deployment data, renewals, utilization, and unit economics, has not yet confirmed it.