China bans Nvidia while GPT-5 destroys humans at coding
📖6 min read
Hey Snackers,
China just dropped the hammer on Nvidia harder than a GPU mining rig during a power outage. Beijing's internet regulator officially banned domestic tech giants like ByteDance and Alibaba from buying Nvidia's AI chips, forcing them to use homegrown alternatives from Huawei instead. The move deals a massive blow to Nvidia's revenue stream and accelerates the global AI chip war into overdrive.
Meanwhile, OpenAI's GPT-5 just dunked on every human programmer at the ICPC World Finals, solving all 12 problems while the best human team could only manage 11. We're officially living in the timeline where robots are better at coding than the people who created them.
Speaking of creating our own demise, Meta unveiled smart glasses with a "Neural Band" that reads the electrical signals in your wrist muscles, letting you control devices with finger movements you haven't even made yet. Because apparently voice commands weren't creepy enough.
SILICON CURTAIN
China tells Nvidia to take its chips and shove them
In what might be the tech equivalent of a messy divorce, China just made official what everyone saw coming: a complete ban on its tech companies buying Nvidia's AI chips.
The Middle Kingdom's internet regulator dropped the ban hammer on companies like ByteDance (TikTok's parent) and Alibaba, telling them they can't purchase Nvidia's AI chips or even the special servers Nvidia designed specifically for the Chinese market. Talk about getting ghosted.
- Beijing had been not-so-subtly nudging its companies toward locally made alternatives from Huawei for months, but now they're making it mandatory.
- This forces China's entire tech ecosystem to rely on potentially less advanced hardware while they scramble to catch up.
- Huawei, sensing opportunity like a shark smelling blood, immediately unveiled its roadmap to build AI chips that can compete with Nvidia, including three new Ascend chip series over the next three years and proprietary high-bandwidth memory.
The timing couldn't be more awkward for Nvidia, which has been printing money faster than the Fed during a crisis.
THE TAKEAWAY
This isn't just about chips - it's about the future of AI supremacy. China is essentially saying "we'll build our own AI infrastructure, with blackjack and... actually just the infrastructure." The real question is whether Huawei can deliver silicon that doesn't make Chinese AI models run like they're on dial-up. Given that DeepSeek claims it trained its R1 model for just $294,000 (compared to the hundreds of millions US companies spend), maybe they're onto something. Or maybe they're just really good at creative accounting.
ROBOT OVERLORDS ACHIEVED
OpenAI's GPT-5 just made every CS student question their life choices
Remember when we thought AI would just write our emails and make weird art? Well, OpenAI's GPT-5 just solved every single problem at the ICPC World Finals - that's the Olympics of collegiate programming - while 139 human university teams struggled to keep up.
The AI achieved a perfect 12 out of 12 score, beating the best human team which managed 11 problems. Google's Gemini 2.5 also showed up to flex, earning gold-medal status. It's like watching a calculator beat humans at mental math, except this time the calculator can also explain its reasoning in perfect English.
- This isn't just about writing fizzbuzz solutions anymore - we're talking about complex logic and reasoning that would make most senior developers sweat.
- The milestone suggests AI has effectively "solved" competitive programming, adding it to the growing list of human activities we're no longer the best at.
- Meanwhile, the Periscope founders (remember livestreaming?) launched Macroscope with $30 million in funding to help developers understand the code that AI is now better at writing than them.
The irony of needing AI to understand AI-generated code is not lost on us.
THE TAKEAWAY
We've reached the point where AI doesn't just assist programmers - it outperforms them. The good news? Someone still needs to tell the AI what to build. The bad news? That's probably next on the disruption list. At least we'll always have... actually, never mind.
MIND GAMES
Meta's new glasses can read your thoughts before you think them
Meta just unveiled smart glasses paired with a "Neural Band" that detects electrical signals in your wrist muscles, allowing you to control devices with finger movements you haven't consciously made yet.
Because apparently shouting "Hey Siri" in public wasn't embarrassing enough, now we need glasses that respond to our pre-conscious muscle twitches. The Ray-Ban branded specs work with a wristband that essentially reads your mind through your arm, which is totally not creepy at all.
- The band can detect subtle finger movements - sometimes before you even consciously decide to move.
- This moves us beyond voice commands toward devices that respond to our intentions, creating what Meta calls an "intuitive interface."
- Zoom, not to be outdone in the dystopian tech race, rolled out photorealistic AI avatars for when you're not camera-ready, plus AI that suggests which meetings you can skip (spoiler: probably all of them).
Nothing says "the future is here" like glasses that know what you want before you do.
THE TAKEAWAY
We're speedrunning toward a future where our devices know us better than we know ourselves. The Neural Band is essentially a mind-reading device with extra steps, and Meta's betting we'll trade our privacy for the ability to scroll Instagram with imperceptible finger twitches. At least when the robots take over, they'll already know exactly what we're thinking.
THE BEST THING WE READ TODAY
Oracle's $300 billion house of cards gets a reality check
Credit rating agency Moody's just looked at Oracle's $300 billion worth of AI contracts (mostly with OpenAI) and basically said "this seems fine" while nervously sweating. The agency flagged massive financial risks in Oracle betting its entire data center expansion on just a few AI companies that may or may not exist in five years. It's like watching someone bet their mortgage on a startup that promises to revolutionize the paperclip industry, except the stakes are measured in GDP percentages.
What else we're Snackin'
- Italy became the first EU country with comprehensive AI laws, requiring human oversight and limiting AI access for kids under 14 (source)
- Irregular raised $80 million to make sure frontier AI models don't accidentally end humanity (source)
- Meta is in talks with Fox and News Corp to license content for AI training, because nothing says "quality data" like cable news (source)
- Google and PayPal announced an AI partnership to make shopping even more addictive (source)
- Keplar raised $3.4 million to replace market researchers with voice AI that never gets tired of asking about your shopping habits (source)
- SoftBank and OpenAI's Japan venture is running late, proving that even AI giants can't escape project management hell (source)