Behind the Ticker · Sep 28, 2026 · 34m · 28m saved
With Kevin T. Carter, Founder, EMQQ Global and EMX ETFs
Nearly every closed, expensive AI model gets the headlines. Kevin T. Carter says the actual adoption story is happening somewhere else entirely — in open-weight Chinese models that cost up to 90% less and that most of Silicon Valley is already quietly running on.
Washington's AI debate has been about whether US labs can out-build China. Carter's answer is that American startups have already answered that question for themselves, by building on Chinese models rather than American ones.
"According to many estimates about 80% of US startups are using China models." — Kevin T. Carter
Carter has worked in the ETF industry since 1992, launched the first China tech ETF in 2010 and EMQQ in 2014, and this year launched EMX ETFs, including a fund built entirely around these Chinese AI labs.
The full episode is covered here so you can skip it. 34 minutes of audio, 6 minutes of reading.
Here are the 5 takeaways that matter.
This Scuttlebutt Summary is available in full to premium subscribers. If you're not one already, please consider subscribing to keep reading.
Key Takeaways
80% of US startups now run on Chinese open-weight AI models, Carter says, because they're customizable, run on-premises, and cost up to 90% less
His new ETF, TGRZ, targets six Chinese labs known as the "Tigers," all spun out of Tsinghua University's computer science lab
DeepSeek started inside a quantitative hedge fund, not a traditional AI lab
The real money, Carter argues, is one layer down the stack: TSMC, Samsung and SK Hynix are projected to earn $965 billion combined across 2026-2027
Power, not chips, is the bottleneck he's most worried about — and he says China already has three times the US's electricity capacity
1. Startups Chose China Models
Carter said the defining difference between the US and Chinese AI industries isn't capability, it's licensing. US labs — Anthropic, OpenAI, Google — keep their models closed. China's major labs release open weights, which lets anyone download a model, run it on their own hardware, and customize it.
"They are just as good getting you know nearly as good as our best models and they're a fraction of the cost in in some cases they're they're 90% less and they're getting used. People don't appreciate how widely used the China open models are now." — Kevin T. Carter
He said this isn't limited to startups: Airbnb runs on Chinese models, and SpaceX's recent acquisition of Cursor brought in a company also built on them, for the same reasons — cost, control over data, and performance that's close enough to the leading closed models.
