Reiner Pope of MatX on accelerating AI with transformer-optimized chips

Bullish:

• $GOOGL - Google has strong AI foundations with TPUs, transformer research origins, and talent pool from Google Brain, positioning them well for AI success.

  • “They started with the research, right? The transformers came from there. Pretty much anyone who's maybe, I don't know, over 30 and at a large lab has been at Google Brain at some point. So I think there's just, like, there was and has been a lot of talent there. TPUs are pretty good.”

    Reiner·0:52

  • “Google has carved out a meaningful advantage via Gemini just being really fast for And its level of as far as I can tell, ahead of most of the other labs, at a latency, at a fixed, you know, high level of intelligence.”

    John·14:45
• $INFRA - Power and data center infrastructure are becoming critical bottlenecks for AI buildout, creating massive demand for infrastructure development.

  • “Data centers, which I think is power primarily, a little bit of build out, but primarily power and grid infrastructure infrastructure there.”

    Reiner·19:59

  • “OpenAI and Tropic, Google are going around buying multi gigawatt clusters. They cost, tens of billions of dollars of chips, and you want to deploy all of that in, like, in a year or so. And so you just need a massive supply chain behind you.”

    Reiner·12:36
• $TSMC - TSMC maintains durable monopoly position through conservative pricing that discourages competitors and technical advantages, with strong support for startup diversity.

  • “They don't charge a lot as well. You'd think that if they're a monopoly provider, they should charge a lot of money. They don't. I think that is a big aspect of why they're so durable.”

    Reiner·37:23

  • “TSMC cares a lot about diversity of their customer pool, and so... they're great to work with from that perspective.”

    Reiner·40:13

  • “They want to encourage startups.”

    Reiner·40:24
• $AIMODELS - AI models will get significantly faster (order of magnitude) over next 3-5 years as new chip architectures combining HBM and SRAM enable both high throughput and low latency.

  • “should expect all the AIs we're using to get significantly faster over the coming three to five years. Or order of magnitude faster, would say.”

    Reiner·16:40

  • “It is actually possible to do both in the same chip. It's kind of an obvious thing. You say you take the HBM, you take the SRAM, put them together on same chip. You put the weights in SRAM, and you put all of the inference data in HBM. That is what we are doing, in fact, And I think that actually hits a really nice sweet spot where you can get low latency and also be very cheap.”

    Reiner·16:11

Bearish:

• $AILABS - AI labs stopped publishing important research around 2022, signaling a shift from open innovation to secretive development that may slow overall progress.

  • “I think why did it all happen then and not later is probably just because people stopped publishing. Mhmm. In 2022 was about the time when just Google completely stopped publishing its research. Yes. Yes. And so all the good papers are from before that as a result.”

    Reiner·3:07
• $NVDA - NVIDIA's CUDA advantage is less relevant for frontier AI labs who write custom software for each chip generation, reducing NVIDIA's moat in the highest-value market segment.

  • “There are not thousands of LLMs. There's one LLM per FrontierLab, and there's maybe five FrontierLabs or something like that. And so just the economics of that is different.”

    Reiner·35:28

  • “When a new chip comes out, software is substantially rewritten to optimize for that specific chip. And that's just the right trade off, given the relative costs of these things.”

    Reiner·36:19

  • “CUDA's way more important for the games environment, where it's just there's a lot of games... Than this top heavy AI market that we're in”

    John·36:45