The company that delivers dinner in China just built one of the biggest AI models on Earth. And it did it without a single American chip.
Meituan is a food delivery app. Think DoorDash. This week they open-sourced LongCat-2.0, a coding model with 1.6 trillion parameters, trained end to end on roughly 50,000 Chinese-made chips. Zero Nvidia. Not in training, not in running it.
The entire point of American export controls since 2022 was to stop exactly this. Keep the fastest silicon out of China so nobody there could train a model this size.
A food company just walked around the wall using hardware they built at home.
It scores 59.5 on the hardest coding test out there. Beats GPT-5.5. Still trails Claude Opus 4.8. Competitive, not king.
But nobody was supposed to get this far without the good chips. Someone did.
What do you guys think?
$MU CEO, in effect:
"For > decade $AAPL has been buying our chips for $5, gluing it inside a metal box, & selling it to consumers for $99 upgrades & laughing at our attempts to get $7.
Now we're charging them $50 & they turned around & raised prices on their customers $250."
$MU, we were significantly above consensus for next two qtrs. As the previews are rolling in, nice to see street coming in close to ours. It also appears my FY28 numbers are conservative, so we will have to do something about it. FYI. DB raised MU PT to $1,500 from $1,000. We still do not see $160 in EPS due to Flash prices correcting in 2027-28. But I guess with contracts in place, anything is possible.
NO Financial Advice. Please do your DD and do not trade on this info.