Nvidia buys into Korea
The day's biggest check came from Nvidia, which is putting roughly $1 billion into Naver, South Korea's largest internet company. The money buys newly issued shares and hands Nvidia a 4.5% stake, with the stated goal of building out South Korea's next AI factory. Investors approved immediately — the market reaction was positive out of the gate.
That deal didn't land in a vacuum. South Korea is pushing on AI from two directions at once. On the industrial side, Hyundai has outlined an expansion of its "physical AI" work in the country — physical AI being the label for artificial intelligence that acts in the real world rather than in a chat window. On the diplomatic side, per UPI, the country's president sat down with OpenAI's Sam Altman. Capital, manufacturing muscle, and political attention are converging on the same square of the map.
A smaller Korean story rhymes with it. T-Robotics, a relatively low-profile robotics firm, is aiming its hardware at two of semiconductor manufacturing's tightest chokepoints: AI chips and glass substrates. Its pitch, per a report from finance.biggo.com, is brute durability — a robot arm rated to shrug off 500°C and handle 240kg. Not glamorous, but the AI buildout runs through exactly these kinds of unglamorous tools.
Physical AI leaves the lab
The theme of the day, if there is one, is AI stepping out of software and into matter. China Daily published a first-person account on July 27 arguing that AI and robotics are moving out of research settings and into physical work. Global Times, meanwhile, gave the trend a slogan: the "next new three" — artificial intelligence, robotics, and innovative drugs — presented as the next wave of Chinese industrial upgrading.
In the US, Richtech Robotics is being profiled along the same axis. A Digital Journal piece titled "From livestream to embodied AI" frames the company as trying to move AI out of the chat window and into face-to-face encounters with actual people.
Embodiment has a bottleneck, though, and it's data. Robots need examples of physical action, and scraped video only goes so far. TechCrunch reports that the next source companies are eyeing is considerably stranger: readings from the human brain. Training robots on brain waves rather than YouTube is early-stage, but it tells you how hard the data problem has become.
Nvidia is attacking the same problem from the software end. The company expanded its Agent Toolkit with PhysicsNeMo and its CUDA-X libraries, pitching AI agents that can reason about physics and help design chips. Nvidia's own framing for the release: it changes "how the world engineers, designs and builds."
Not everyone is buying
One useful counterweight came from western New York. The Salamanca Central School District has paused a pilot that would have put AI robots in classrooms — stopped before the machines ever reached students' desks. It's a small decision with a familiar shape: enthusiasm at the top of the funnel, hesitation at the point of contact.
Smart glasses have a privacy problem
Apple's first smart glasses are running late, and reporting this week points at privacy as a major reason. Per Engadget, the team behind the product is wrestling with concerns serious enough to hold up the launch.
They have a live example to study. Futurism reports that Meta's camera-equipped AI glasses have developed a creep problem — the device that was supposed to be the next great consumer gadget is currently better known as a privacy headache. And the fallout isn't staying with Meta. Apple noticed. When your competitor's product becomes a cautionary tale, shipping late starts to look like a strategy rather than a stumble.
And the long game
Finally, SoftBank and Quantinuum published a joint white paper attempting to answer quantum computing's most stubborn question — not whether quantum machines will be useful, but when. Putting a date on it is the harder and more honest exercise, and it's notable that two players with skin in the game are willing to try.