> ## Content Index
> Fetch the complete content index at: https://news.theresarobotforthat.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# AMD Bets $8.2 Billion on Physical Intelligence
- URL: https://news.theresarobotforthat.com/amd-bets-82-billion-on-physical-intelligence/
- Published: 2026-09-30T10:30:00.000Z
- Updated: 2026-09-30T10:29:59.000Z
- Description: A $150M edge-AI chip round, China's patent grip on humanoids, and the Army's humanoid field test.
- Author: Ulrich Baldauf
- Tags: Newsletter

### The Big Number

AMD agreed to pay $8.2 billion, all in stock, for Fei-Fei Li's World Labs, its second-largest acquisition ever and its clearest bet yet that the next decade of AI happens in the physical world.

## Welcome back to your Robot Briefing

This week the money and the machines pulled in opposite directions. Chipmakers and investors poured billions into the silicon and models meant to run robots, while the Army published the results of its field test of commercial humanoids, and only one complete robot made its list of five winners.

That gap is the whole story. It is easy to fund a vision of autonomous machines and hard to build one that holds up outside the lab, and the companies worth watching are the ones still working on the second problem.

The 60-Second Read

AMD is buying Fei-Fei Li's World Labs for $8.2 billion to own the world models behind physical AI.

SiMa.ai raised $150 million at a $1.45 billion valuation to run robot AI on the machine instead of the cloud.

China now holds 60.8 percent of the world's humanoid robot patents, and the gap is still widening.

The U.S. Army picked five winners from 103 humanoid proposals, and only one of them is a complete robot.

News

### AMD Pays $8.2 Billion to Build the Brains Behind Robots

**Snapshot:** AMD [agreed to acquire](https://www.cnbc.com/2026/09/28/amd-fei-fei-li-world-labs.html?ref=news.theresarobotforthat.com) World Labs, the San Francisco lab founded by AI pioneer Fei-Fei Li, for $8.2 billion in an all-stock deal that hands the chipmaker a research team working on spatial intelligence, the models that teach machines to understand three-dimensional space.

Breakdown:

The transaction is all stock, expected to close by the end of 2026, and brings Li into AMD as executive vice president and chief scientist, reporting to CEO Lisa Su.

World Labs builds so-called world models, systems that let AI reason about 3D environments and physical interactions, with uses in robotics, simulation and design.

AMD had already backed World Labs as an investor before agreeing to buy it outright.

At $8.2 billion it is AMD's second-largest acquisition on record, behind the roughly $50 billion it paid for Xilinx in 2022.

**Takeaway:** The people who buy humanoid robots will not shop for world models directly, but they will feel this deal in the roadmap. When a chipmaker writes its second-largest check ever for the software that teaches machines to understand space, it is signaling that the constraint on a useful robot is no longer legs or torque, it is spatial reasoning. Expect the next factory and warehouse humanoids to be sold on how well they read a cluttered room, not how fast they can walk across an empty one.

News

### A $1.45 Billion Chip Startup Wants to Unplug Robots From the Cloud

**Snapshot:** SiMa.ai [raised](https://www.businesswire.com/news/home/20260928291227/en/SiMa.ai-Reaches-%241.45B-Valuation-with-%24500-Million-in-Total-Funding-to-Scale-Physical-AI-in-Humanoids-Automotive-and-Drones?ref=news.theresarobotforthat.com) $150 million in Series C funding at a $1.45 billion valuation, money it will spend to sell robots, drones and cars chips that run AI on the machine itself rather than shipping data to a distant data center.

Breakdown:

The round was co-led by Fidelity Management & Research and Amplify and brings total capital raised past $500 million, with AllianceBernstein, Baron Capital, J.P. Morgan and the State of Michigan joining as new backers.

Founded in 2018 by former Groq operating chief Krishna Rangasayee, the company was valued at $960 million after an $85 million Series B in July 2025, so this round lifts its worth by about half in roughly a year.

SiMa pitches lower latency and cheaper silicon than Nvidia's GPUs, and says it quadrupled revenue between 2024 and 2025.

Its next chip, slated for the first half of 2028, targets 1,000 dense TOPS aimed at humanoids, drones and driver-assistance systems.

**Takeaway:** For anyone speccing a humanoid, the quiet signal here is power and latency, not the valuation. A robot that round-trips every decision to the cloud is a robot that stalls when the network does and burns budget on bandwidth. A wave of edge-AI silicon aimed squarely at humanoids means the machines you evaluate next year should be judged on what they can decide on-board, untethered from a data center.

⏱️ The Compliance Clock

As humanoids move off show floors and into homes and onto factory lines, the standard that decides how hard a robot may touch a person becomes the one buyers cannot skip. ISO/TS 15066 sets concrete force and pressure limits for collaborative robots that share space with people, and any humanoid working within arm's reach of your staff sits squarely inside its scope. Before any machine works beside your staff, ask the vendor how it measures and caps contact force against that standard.

[Read the full guide →](https://theresarobotforthat.com/blog/collaborative-robots-cobots-guide-2026/?utm%5Fsource=newsletter&utm%5Fmedium=email&utm%5Fcampaign=compliance-clock)

News

### China Now Holds 6 in 10 Humanoid Robot Patents

![Bar and pie charts comparing humanoid robot patent data by country. China leads significantly with 60.8 percent of global patents and 1,584 applications in the first half of 2026, dwarfing the United States at 21.8 percent and 113 applications, and South Korea at 4.6 percent and 35 applications.](https://res.cloudinary.com/duuood1js/image/upload/f_webp,q_auto/v1790734986/theresarobotforthat/C257-infographic-2026-09-30.png)

Image Source: There's A Robot For That

**Snapshot:** China [held](https://www.koreaherald.com/article/10886376?ref=news.theresarobotforthat.com) 60.8 percent of the world's humanoid robot patents across the five major patent offices in the first half of 2026, up from 51.8 percent in 2024, while the United States slipped to 21.8 percent and South Korea to 4.6 percent.

Breakdown:

In the first half of 2026 China filed 1,584 humanoid patent applications, against 113 from the United States and 35 from South Korea.

Its lead is wider still in robot foundation models, the AI systems that let a robot learn new tasks without being reprogrammed, where China holds 66.3 percent of patents to America's 14.1 percent.

A Korean government assessment scored China's advanced-manipulation capability at 90 against a US benchmark of 100 in 2024, ahead of Korea's 80, a reversal from 2022.

Analysts tie the gap to China's tightly integrated supply chain for actuators, joints and assembly, the same cost advantage it built in electric vehicles.

**Takeaway:** Patent share is not the same as better robots, but it does decide who pays whom. A buyer sourcing humanoids in 2027 is increasingly sourcing Chinese-held intellectual property, which raises the odds of royalty costs or supply dependence baked into any non-Chinese machine. It is worth asking a Western or Korean vendor which core components and models they actually own, and which they license.

News

### The Army Tested Commercial Humanoids. One Complete Robot Made the Cut.

![The Army Tested Commercial Humanoids. One Complete Robot Made the Cut.](https://storage.ghost.io/c/9c/a6/9ca6316d-1f3a-4d33-b3e6-87f231933bbb/content/images/2026/09/story-2026-09-30-3-1.jpg)

Image source: U.S. Army

**Snapshot:** The U.S. Army [named five winners](https://pit.army.mil/update/xtechhumanoid-winners-advance-military-exploration-of-emerging-humanoid-capabilities/?ref=news.theresarobotforthat.com) of its xTechHumanoid competition out of 103 proposals, and only one of them is a complete humanoid: Alex, built by the Florida Institute for Human and Machine Cognition, with ballistic protection and impact-resilient actuators for dangerous missions.

Breakdown:

Ten companies reached the finals, and nine demonstrated their technology on September 9 and 10 at Marine Corps Base Quantico in front of military leaders and evaluators. The Army named the winners on September 11.

The other four winners build parts, not robots: an integrated power system from ACEs Group, mobile power generation from Hydroplane, fingertip tactile sensors from RAS Labs and command-and-control software from Velocity Explorations.

Alex runs on what the Army calls hybrid autonomy, combining artificial intelligence with human control.

The Humanoid Community of Collaboration, which coordinates humanoid work across the Joint Force, says its next lines of effort include the power and energy limits that could shape how humanoid and other robots are used in the field.

**Takeaway:** Read the winner list as a map of where humanoids still fall short. Two of five prizes went to power, one to touch and one to the software that turns orders into actions, which puts the hard problems in energy, dexterity and control rather than in the walking body. The same limits show up on a factory floor, only with lower stakes. Before you trust a humanoid pitch, ask how long it runs on one charge, how it senses grip and slip, and how much of the job still needs a person at the controls.

▶ Operator's Playbook

Before you sign any humanoid pilot this quarter, ask the vendor for two numbers in writing. First, the maximum contact force the robot can exert on a person, measured against ISO/TS 15066, which tells you whether it can safely work beside your staff. Second, the share of tasks it completes without a remote operator, which tells you whether you are buying automation or renting teleoperation. A vendor that cannot put either on paper is not ready for your floor.

### Other Top Robot Stories

**Maven Robotics** [pulled in $100 million](https://www.zawya.com/en/press-release/companies-news/shorooq-and-presight-back-maven-robotics-in-100mln-series-a-bringing-gulf-capital-to-the-frontier-of-physical-ai-1298027?ref=news.theresarobotforthat.com) in a Series A from Abu Dhabi's Shorooq and Presight, a G42 company, bringing Gulf capital into general-purpose industrial robotics.

**IPercept** [raised $16.5 million](https://techfundingnews.com/ipercept-16-5m-series-a-cnc-diagnostics-2150-isogon/?ref=news.theresarobotforthat.com) in a Stockholm Series A co-led by Isogon Ventures and 2150 to put physical AI on factory-floor CNC machines for customers including Airbus, Bosch and Volvo.

**DYNA Robotics** [launched DYNA 2.1](https://www.prnewswire.com/news-releases/dyna-robotics-launches-dyna-2-1-physical-agent-a-semi-humanoid-robot-that-completes-full-workflows-such-as-a-commercial-laundry-shift-302892411.html?ref=news.theresarobotforthat.com) , a semi-humanoid physical agent it says can autonomously complete an entire commercial laundry shift end to end.

**Neubility** [unveiled Billy](https://en.sedaily.com/technology/2026/09/29/neubility-unveils-semi-humanoid-robot-billy-in-push-for?ref=news.theresarobotforthat.com) , a two-armed semi-humanoid on a four-wheeled base that the Korean delivery-robot maker says it aims to turn into the country's first commercialized humanoid.

**NASA** [profiled its dexterous robotics team](https://www.nasa.gov/general/reliable-robots-meet-johnsons-dexterous-robotics-team/?ref=news.theresarobotforthat.com) at Johnson Space Center, the engineers building robots reliable enough to handle tools and hardware in space.

Compliance tool

Where does your robot documentation stand?

The ISO 10218:2025 Gap Analysis Checklist finds every gap in 62 audit questions.

[Get the checklist ($49) →](https://theresarobotforthat.lemonsqueezy.com/checkout/buy/e01e6740-edbc-41f2-b845-8822e7258aa7?ref=news.theresarobotforthat.com)

**🤖 Your robotics thought for today:**

This week the money and the machines told different stories. The money said the future is settled, with billions moving into the chips and models meant to run robots, as if the hard part were already solved. The machines said something quieter. When the Army put commercial humanoids through a field test this month, only one complete robot made the winners' list, and the rest were batteries, sensors and software. That is not a failure, it is where the work actually is. It is easy to fund a vision and hard to build a machine that keeps going in the mud, and the companies that win will be the ones still paying attention to the second problem long after the headlines move on.

Until Friday,  
Uli