Blog

Physical AI raised more in six months than in three years — and 52% went to four companies

Robotics, drones, sensors, and autonomous vehicles pulled in $47.4 billion in the first half of 2026, more than 2022, 2023, and 2024 combined. But four rounds account for half of that — and the part that actually reaches your business is something else entirely.

August 20, 2026 · Agência Primeira Página

Physical AI raised more in six months than in three years — and 52% went to four companies

In six months, so-called physical AI raised more money than in three full years. Robotics, autonomous vehicle, aerospace, drone, industrial automation, and sensor startups pulled in $47.4 billion in the first half of 2026, across 521 rounds, according to Crunchbase. 2022, 2023, and 2024 combined totaled $41.9 billion.

The chart is breathtaking: $12 billion in the second half of 2025, $26.4 billion in the first half of that year, and now nearly double that. But there's a detail buried in the numbers that completely changes the reading — and it's the one that matters to anyone running a business.

Half the money went to four companies

Four rounds account for more than half of the half-year total:

  • Waymo — $16 billion in February, at a $126 billion valuation;
  • Anduril — $5 billion in May, valued at $61 billion;
  • Shield AI — $2 billion in March;
  • Saronic — $1.75 billion in March.

That's nearly $25 billion, or roughly 52% of the total. Three of these four operate in defense or autonomous vehicles — markets with state clients, long cycles, and extremely high barriers to entry. And the number of rounds grew far less than the volume: 436 in the first half of 2025 versus 521 now, a 19% increase in deals while the money nearly doubled.

This isn't a distributed wave. It's concentration. When volume rises much faster than the number of deals, it means the same few names are getting bigger — not that funding showed up for everyone.

What actually changed, according to investors

Investors' explanation is less glamorous than the headline and more useful. Ryan Ziegler, of Edison Partners, sums it up this way: the cost of building this kind of company has dropped, and AI infrastructure is now readily available. Joe Fath, of Eclipse Capital, adds that technical barriers are disappearing and that experienced people are migrating into the sector.

Translated for anyone who isn't about to raise a billion dollars: components got cheap and off-the-shelf models became accessible. A regular camera with computer vision, a presence sensor, automatic label reading, part counting, visual inspection — all of this stopped being an engineering project and became integration of parts that already exist.

What reaches your business (and what doesn't)

What doesn't reach you: a humanoid robot sweeping the store, an autonomous delivery fleet, a robotic arm replacing your team. That's still expensive and still the domain of those with industrial scale.

What does reach you, and is already cheap:

  1. Image-based counting and verification. A camera pointed at the conveyor belt, the dock, or the shelf, with a simple vision model counting what passed by.
  2. Visual quality inspection. Comparing a part against the standard and flagging what's off — the same task that today depends on someone checking at the end of the shift.
  3. Sensors with automatic alerts. Cold storage temperature, tank level, a door left open after hours: a cheap sensor plus a simple rule, alerting on your phone.
  4. Physical document reading. An invoice, a packing list, a handwritten form turning into a system record without manual typing.

None of these four is "physical AI" in the sense Crunchbase measures. They're the accessible version of the same technology — and that's exactly where the cost dropped.

The test before spending a dollar

Automation in the physical world fails more often than automation in the digital world: lighting changes, the part comes in crooked, someone bumps the camera. That's why the bar is different:

  • Does the task repeat every day and have one correct way of being done? If it depends on case-by-case judgment, don't automate it yet.
  • Is the error cheap? A miscounted box can be corrected. Discarding a good product cannot. Start with what's fixable.
  • Is there someone to check it during the first week? Physical automation without initial supervision accumulates silent errors.
  • Have you measured how much time that task takes today? Without that number, there's no way to know if it was worth it.

The honest reading

The real headline isn't "physical AI exploded." It's: venture capital concentrated a giant bet on a handful of defense and autonomous vehicle companies, while component costs dropped for everyone else. The first half of that sentence changes nothing for your business. The second half does — and it's the part nobody is announcing.

If you've ever considered automating a repetitive physical task and stalled on the budget, now is the time to run the numbers again. We wrote about this when robotics started becoming accessible in the iPhone moment for robotics, and it's the kind of project we handle in AI implementation for businesses.

Source: Crunchbase News, physical AI investment survey for the first half of 2026.

Perguntas frequentes

What is physical AI?

It covers artificial intelligence applications that act in the real world rather than on screen: robotics, autonomous vehicles, aerospace, drones, industrial automation and sensors. It is the category Crunchbase uses in its funding report.

How much was invested in physical AI in 2026?

US$47.4 billion in the first half of 2026 across 521 rounds, according to Crunchbase — more than the US$41.9 billion raised across 2022, 2023 and 2024 combined. The second half of 2025 saw US$12 billion; the first half of that year, US$26.4 billion.

Is that money spread across many companies?

No. Four rounds account for roughly 52% of the total: Waymo (US$16 billion), Anduril (US$5 billion), Shield AI (US$2 billion) and Saronic (US$1.75 billion). Deal count rose 19% while volume nearly doubled, which points to concentration rather than plentiful funding for everyone.

What does physical AI change for a small business?

What actually changed is the cost of components and access to off-the-shelf models. In practice that makes image-based counting and checking, visual quality inspection, sensors with automatic alerts and paper-document capture without retyping all affordable — not humanoid robots or autonomous fleets.

Why does physical automation fail more often than digital automation?

Because the environment shifts: lighting, part position, a dirty lens, someone knocking the equipment. That is why it pays to start with repetitive tasks that have one correct way of being done, where an error is cheap and reversible, and to keep a human checking during the first week.

How do I know whether a task is worth automating?

Measure first how much time the task consumes today and how often it repeats. Without that number there is no way to weigh the project cost against the gain. Then check whether the task follows a clear pattern and whether an occasional error can be corrected without serious loss.