Figure AI went from 1 robot per day to 1 per hour — in 4 months. Japan Airlines signed a 3-year humanoid contract. $37 billion in VC landed this year alone. Physical AI moved fast this week. Here is everything your industry missed, and why it matters more than the headlines suggested.

---

Think about the last time an airline signed a 3-year contract on technology that was still experimental. You cannot, because they do not. Airlines operate in one of the most regulated, liability-conscious industries on the planet. When Japan Airlines committed to a humanoid robot program at Haneda Airport in May 2026, they were not running a pilot. They were making a procurement decision — the same way they procure ground equipment, check-in systems, or gate management software.

That is the moment Physical AI changed categories. Not from a product demo, not from a VC funding round, but from a boring, bureaucratic, multi-year service contract at an airport most of you have probably transited through.

And JAL is not alone. This week told a consistent story across four separate industries. Here is what happened — and what it means beyond the headlines.

The numbers first, so we have the same starting point

When the Supply Chain moves, the Market is real

The strongest signal I track is not what robot companies say — it is what component suppliers do. This week, Khgears International, a Taiwanese manufacturer of precision gearboxes for industrial robots, announced a full pivot into humanoid-specific components: joints, drive mechanisms, actuator assemblies. They are seeking a strategic alliance with a Japanese Tier 1 automotive supplier to do it.

Khgears does not make this move on speculation. Gearbox manufacturers pivot when they have seen enough confirmed purchase orders to justify retooling their factory. When the supply chain moves, it means the demand is real, not projected.

The same signal comes from production lines. Figure AI went from producing one humanoid robot per day in January 2026 to one per hour by May — a 24x acceleration in four months. Their BotQ factory has already delivered 350+ units. For context: that is not a startup proving a concept. That is a manufacturer ramping toward industrial scale.

Why this matters: Component suppliers and production lines are lagging indicators — they follow confirmed demand. When they move at the same time, the market inflection has already happened. You are reading about the aftermath, not the prediction.

Robots are getting smarter faster than the hardware can keep up

Here is something that gets lost in the factory-and-funding coverage: the AI inside these robots is improving on a completely separate, faster curve. XPeng rolled the first mass-produced L4 robotaxi off its Guangzhou line on May 18 — and the AI model powering that car is the exact same model running their Iron humanoid robot. One model trained once, deployed in two completely different physical systems.

This is a bigger deal than it looks. Training a frontier vision-language-action model costs tens of millions of dollars. If that cost is shared between an autonomous vehicle fleet and a humanoid workforce, the unit economics of Physical AI become dramatically better than analysts currently model. NVIDIA made the same bet in a different direction — their Isaac GR00T models are now open source, meaning any robotics company can build on a foundation instead of starting from scratch.

DARPA is already asking what comes after this architecture entirely. Their May 2026 research call imagines robots where the material itself computes — no central processor, no cloud, no latency. That is a 10-year horizon, but DARPA's early bets have a habit of becoming everyone's reality.

Why this matters: Shared AI models across platforms mean the cost of building capable robots is falling faster than the hardware suggests. The gap between "what robots can do in a lab" and "what they cost to deploy at scale" is closing from both ends simultaneously.

What this actually means if you are not an Investor

A direct note to everyone who works in logistics, manufacturing, aviation, or any field with structured physical tasks: the companies in this article are not running experiments in your industry. They are operating under multi-year service contracts. The question is no longer whether robots will enter your workplace. It is which tasks they take first, and how fast.

Barclays Research framed the macro picture in their May 2026 report: the humanoid robot market could reach $200 billion by 2035, and for China specifically, robots may offset up to 60% of the demographic workforce decline projected over the next decade. That last number is not a technology story — it is a labor economics story, and it will play out in every aging economy, not just China's.

The honest answer to "will robots take my job?" is still nuanced. Agility Robotics' agreement with Toyota covers logistics tasks in a manufacturing plant — moving parts, not assembling them. The Vodafone pilot in Duisburg had robots detecting misplaced products and unsafe pallet stacking, not replacing warehouse managers. The pattern so far is robots handling the physically repetitive and physically risky parts of jobs humans already find exhausting. But the category is expanding, and the speed of expansion is the variable to watch.

Why this matters: The Barclays report title is "Robots roll out, economies rewire." That word — rewire — is the honest one. Not replace, not eliminate. Rewire. The people who will navigate this best are the ones who start paying attention now, not when the robot is already at the next workstation.

What to watch next

• Figure AI's BotQ throughput in Q3 — sustaining 1 robot/hour would make them the first humanoid manufacturer at genuine industrial scale by year-end.
• XPeng Iron deployment update — the first real test of whether one AI model can actually run both a robotaxi fleet and a humanoid workforce in production.
• Physical Intelligence's $1B raise — if it closes at the reported $11B valuation, it resets comparables for the entire sector and triggers a new wave of raises.
• Khgears' Tier 1 alliance — whoever they partner with signals which Japanese industrial giant is moving seriously into humanoid supply chains.
• The 85% China concentration risk — one country accounting for 85% of global deployments is a geopolitical variable that no analyst is pricing correctly yet.

---

FAQ

Q: What is Physical AI and why does 2026 matter?

A: Physical AI is artificial intelligence that operates in the real, physical world — humanoid robots, autonomous vehicles, robotic arms that reason in real time. 2026 matters because procurement replaced experimentation: Japan Airlines, Toyota, Amazon, and Vodafone are signing multi-year service contracts, not running pilots. Figure AI is producing one humanoid per hour. Barclays forecasts a $200 billion market by 2035. The phase shift from R&D to deployment happened this year.

Q: Will humanoid robots replace human workers?

A: The current deployment pattern is task replacement, not job replacement — robots are taking over physically repetitive, dangerous, or high-precision tasks within jobs that remain human-managed. Agility Robotics at Toyota handles parts logistics; humans still run the line. The Barclays framing is more accurate: economies will "rewire" rather than simply lose jobs. The speed of that rewiring, however, is accelerating significantly in 2026, and the category of tasks robots can handle is expanding rapidly.

Q: Which companies should I be watching in Physical AI right now?

A: Figure AI for production velocity and deployment scale. XPeng for the shared AI model strategy across robotaxi and humanoid. Physical Intelligence for foundation model development (their $1B raise at $11B valuation is a sector bellwether). NVIDIA as infrastructure — Isaac GR00T is becoming the Linux of robotics AI. And watch Khgears and other component suppliers: they tell you what the demand actually is, not what companies claim it will be.

Physical AI Digest is a weekly briefing produced by xBerry — a tech company based in Poland building tools at the intersection of AI and operations.