🤖 Humanoid startup Apptronik raises $520 million with backing from Google and Mercedes-Benz ↗
Apptronik just pulled in a massive $520M extension round, with Google and Mercedes-Benz in the mix - plus a stack of other heavyweight investors. The implied valuation floating around is… big. Like, “seriously” big.
The pitch is straightforward: scale Apollo, their humanoid robot, and push it into factories and warehouses where boring, repetitive work never ends. There’s also a deeper angle here - tighter integration with Google DeepMind so Apollo can run smarter on Gemini-based models, which is basically the “make the robot less dumb” fast lane.
🧠 Top Democrat on US House China committee open to Nvidia H200 sales ↗
Ro Khanna is signaling he’s not automatically against selling Nvidia’s older H200 chips into China - which is a noticeable shift in tone compared with the harder-line posture the committee has had. The gist: if the US stays a couple generations ahead, older chips might be tolerable for non-military use… or so it seems.
But he draws a sharper line on newer stuff - newer architectures shouldn’t be on the menu. Meanwhile, committee leadership on the other side is still waving the “civil-military fusion” flag, basically arguing that “non-military” is a polite bedtime story.
🏗️ AI inference startup Modal Labs in talks to raise at $2.5B valuation, sources say ↗
Modal Labs is reportedly in fundraising talks at around a $2.5B valuation, which - yeah - tells you how hot “inference” still is. Training gets the glamour, but inference is the part you pay for forever, like a subscription you can’t cancel.
If this lands, it’s another sign investors are treating compute delivery and model serving like the new cloud gold rush. Slightly off-kilter metaphor, but it fits - everyone wants to sell picks and shovels, even if the miners are algorithms.
📺 Reception of AI ads “sharply negative” as brand beefs … ↗
Turns out people aren’t exactly swooning over AI-heavy advertising - some of the reaction is described as “sharply negative,” and I understand it. There’s a certain uncanny slickness that makes your brain go “nope,” even when the visuals are impressive.
Also, the AI brand drama is getting openly snarky: Anthropic ran a campaign for Claude that takes swings at AI platform advertising, and the subtext feels… not subtle. It’s like a corporate pillow fight, except the pillows are made of venture capital.
⚖️ CLEAR Act Would Establish Notice Requirements for Copyrighted Works in AI Training Data ↗
A proposed US bill called the CLEAR Act is pushing for notice requirements around copyrighted works used in AI training datasets - basically “tell people what you used,” at least in a structured, formal way. It’s less “ban it” and more “prove you’re not being sneaky.”
If it moved forward, it would add another compliance layer for companies releasing generative AI commercially. Which could be fine! Or it could turn into paperwork soup, depending on how the rules get written and enforced… tiny backtrack, but that part matters.
💾 Nvidia Stock Rises. Broadcom Is Closing the AI Chips Gap, Analyst Says. ↗
Analysts are arguing Broadcom is getting closer to Nvidia in AI chips, mainly through custom silicon - especially work tied to Google’s TPUs. The more companies chase cheaper inference, the more “good enough and affordable” hardware starts to look dangerous to the premium king.
This doesn’t mean Nvidia is suddenly cooked - not even close - but it does hint at a market splitting into two lanes: elite training monsters, and cost-optimized inference workhorses. Same sport, different weight classes.
FAQ
What happened in this AI tech news roundup?
This update tracks major movement across robots, chips, compute, ads, and copyright policy. Apptronik raised substantial funding to scale its Apollo humanoid robot for factory and warehouse work. Alongside that are shifting signals on Nvidia chip exports to China, a reported high-valuation raise for inference startup Modal Labs, consumer backlash to AI-heavy ads, a proposed CLEAR Act focused on training-data notice, and indications that Broadcom is narrowing parts of the AI chips gap.
Why is Apptronik raising so much money for its Apollo humanoid robot?
The stated aim is to scale Apollo and deploy it into factories and warehouses where repetitive tasks are common. The round includes heavyweight backers, pointing to investor confidence in physical automation as a long-horizon category. The story also highlights a deeper angle: tighter integration with Google DeepMind so Apollo can run more intelligently on Gemini-based models, pushing capability forward rather than only increasing the number of units produced.
How would Google DeepMind and Gemini-based models make Apollo “smarter”?
The premise is that stronger AI models can improve how a humanoid robot interprets instructions, plans actions, and adapts to messy on-the-ground conditions. Instead of brittle, pre-scripted behaviors, a robot can become more flexible through language and perception. In many pipelines, that translates into better task generalization and faster iteration on new workflows, especially when moving between different factory or warehouse environments.
Are U.S. lawmakers really open to Nvidia H200 chip sales to China?
One notable signal is that Rep. Ro Khanna suggested he’s not automatically opposed to selling older Nvidia H200 chips into China, depending on safeguards and use cases. The framing is that staying “a couple generations ahead” could make older chips less sensitive for non-military use. He draws a sharper line on newer architectures, while other committee voices still emphasize civil-military fusion risks.
Why are AI inference startups like Modal Labs getting such high valuations?
Inference is the “forever cost” of AI: once models are deployed, serving them at scale becomes an ongoing expense. That makes compute delivery, model serving, and optimization feel like a durable business layer - closer to infrastructure than a one-time build. Modal Labs being reported in talks around a $2.5B valuation reflects investor appetite for picks-and-shovels companies that help others run AI reliably and cheaply.
What does the CLEAR Act proposal mean for copyrighted works in AI training data?
The proposal described here centers on notice requirements - more “tell people what you used” than “ban training.” If it advanced, it could add compliance steps for companies commercializing generative AI, potentially requiring structured disclosure around copyrighted inputs. Whether that becomes manageable transparency or “paperwork soup” depends on how specific the rules get, how burdens are allocated, and what enforcement looks like in practice.