A graphite and gold humanoid factory worker holding a crate.

2030 GROWTH PROJECTION

$7–15B

Factories and warehouses lead early use.

$7B base case · $15B optimistic [1]

Three graphite and gold humanoids representing wider deployment.

2035 GROWTH PROJECTION

~$138B

~6.5M unit sales in Goldman’s updated forecast.

Goldman Sachs · reported Aug 2026 [2]

A dexterous robot hand manipulating a cube.

2050 GROWTH PROJECTION

~$5T / year

~1B humanoids in use, in Morgan Stanley’s outlook.

Long-term adoption scenario [3]

RWA / REAL-WORLD ACTION

The next wave
moves.

Humanoids bring intelligence into the spaces where people live and work.

As the market grows, so does the need for experience grounded in the physical world.
From industrial workflows to everyday tasks.

Explore real-world data

Different research forecasts and scenarios; assumptions and definitions vary. These figures describe possible future markets.

BEYOND INTERNET TEXT

From reading to doing.

The next wave needs a different kind of experience.

A finite text supply

Websites, books, code and papers powered the first wave. The stock of useful public human-written text is finite.

More isn’t always better

Fresh web content can be repetitive, AI-generated, restricted or noisy. More scraping alone does not guarantee better training data.

Experience has to be made

A robot’s camera views, joint states, contact forces and actions have to be captured through physical interaction.

2026–2032

Epoch AI projects that public human text could be fully utilized within this window if scaling trends continue. This is a constraint on one source of data, not the end of AI progress.

WHY COMPANIES ARE PIVOTING

Language describes.
Experience teaches.

Text alone does not teach reliable physical control. Labs are investing in teleoperation, human demonstrations and robot fleets to connect perception with action.

Each physical task can produce fresh evidence about what works, what fails and how the world responds.

Inside the data opportunity

Context & sources

The 2030 band is Policy2050’s base-to-optimistic scenario; its conservative case is $3.5B. Goldman’s updated $138B and 6.48M unit-sales figures are attributed to August 2026 reporting. Morgan Stanley’s revenue estimate is approximately $4.7T annually, rounded to $5T, with more than 1B humanoids in use by 2050. These are not one continuous forecast series.