Kevin T. Carter, EMX ETFs

Kevin Carter got his first job at Robertson Stephens in 1992. They told him he could start Monday. He said he didn't know anything. They handed him A Random Walk Down Wall Street and told him to read it over the weekend. A few years later he was building companies with the man who wrote it.

With Burt Malkiel he built eInvesting, the first fractional share brokerage, and sold it to E*Trade. Then Active Index Advisors, which was direct indexing before anyone called it that, sold to Natixis in 2004. Around then some of their investors asked for China exposure. Kevin said he'd figure it out. That was twenty years ago, and he's still at it: the first China tech ETF with Guggenheim in 2010, an emerging markets internet fund in 2014 after he noticed Mercado Libre looked a lot like the China internet names. Now a new brand, EMX ETFs, built with Tidal, and a new fund that went live on August 26.

He framed it with the five-layer stack Jensen Huang uses for AI. Power at the bottom. Then chips, which covers semis, memory, optics, racks, cooling, all of it. Then the data center. Those three layers are the infrastructure, and the companies selling picks and shovels into it are mostly in Korea and Taiwan. That's where emerging markets come in. Layer four is the models, which Huang calls the magic layer. That's where this fund sits.

Kevin spent a lot of time on open weights. The big US labs keep their models closed. The Chinese labs publish theirs. You download the model, run it on your own machines, tune it, and keep your data in house. He puts the cost at as much as 90% below the US models, and says the quality is close. By many estimates about 80% of US startups now build on Chinese models. He named Airbnb, and Cursor, the business SpaceX bought a couple of months ago.

The Tigers is a term used in China for six labs, and every one came out of the same computer science lab at Tsinghua University in Beijing. Z.ai was the first large language model company to go public anywhere, in Hong Kong this year. MiniMax listed second, also in Hong Kong. When Moonshot's Kimi K3 launched in July, it was nearly as good as the best US models at a fraction of the price, and it was the story for about a week. Moonshot filed for a Hong Kong IPO right before we recorded. The other three are StepFun, Baichuan, which is focused on healthcare, and 01.AI, founded by Kai-Fu Lee, who used to run Google in China. The fund can also own names outside the six, like Alibaba, whose Qwen model has been the most downloaded model in the world for several months.

DeepSeek isn't one of the Tigers because a hedge fund started it. Its founder was a math prodigy running quant strategies on used Nvidia gaming cards while still in college. He started a hedge fund, which meant a lot of Nvidia chips, and a small team built the model on the side. DeepSeek just raised about $7 billion, and Kevin says the founder put in $3 billion of it himself. It's actively pursuing an IPO too. The prospectus lets the fund put up to 15% in private companies. Kevin says it's early, but pre-IPO access to Moonshot and DeepSeek is part of why he launched the fund this narrow and this soon. Twenty years ago he hired about ten Chinese MBA students in the Bay Area as analysts. Some of them are now senior people in Chinese investment banking. He knows who to call.

I've been testing the Kimi models myself, and I told him what I've noticed. You hand them a task and they go do it. The US models are often better, but they tend to keep you in the chat longer.

Then I asked the value question. Kevin calls himself an Omaha guy first. How does a Buffett lens fit a fund of AI labs? He said he doesn't know that it does. None of these companies make money, and that includes Anthropic and OpenAI. He expects a lot of commoditization in the models, and he thinks the closed labs trying to build a moat are the ones most at risk. His own words for this part of the market were highly speculative.

He sees value further down the stack. TSMC, Samsung and SK Hynix are now about a third of the MSCI emerging markets index, maybe the most concentrated broad market index ever. Those three are expected to earn $965 billion across 2026 and 2027. That's more than Berkshire has earned in its whole history, and about three times what Amazon has made in thirty years. He asked me to guess the P/E on Hynix. It's three and a half. The three trade under five times earnings as a group, partly because the market still prices memory as a boom and bust business. Kevin thinks it's secular now. He also pointed out that the FTSE emerging markets index doesn't hold Korea at all, and that one difference has opened a wide gap between funds that track FTSE and funds that track MSCI.

Memory is already a bottleneck. Its cost is up several fold, and small companies are going under because they can't get it. The bigger one he sees is power. Turbines and generators are sold out for years, and people are pulling engines off retired airplanes to make electricity. China has about three times the power the US has, and added more in the last four years than the US has in total. He also flagged optics, where copper is giving way to optical links and China is the dominant supplier.

How Should an Adviser Use It?

My last question was whether an adviser should treat this as thematic exposure. He said it's absolutely thematic. The single-layer funds, this one included, are for more active investors. For strategic allocations, he thinks these companies belong inside the broad emerging markets AI strategy EMX plans to launch soon, which owns the whole stack.