Michael Dempsey(@mhdempsey):A bit of a rambling idea space on AI but: increasingly i do feel like these types of things (style/bias tweet below is written in, latest podcast with dwarkesh + dylan, other content) from a subset of insiders are a rorschach test for if you believe hyper narrow but incredibly deep insights win out in how one predicts the future versus one layer up but more well-rounded/mutli-disciplinary understandings. A lot of the takes kind of bridge the gap of understanding capitalism shifts very well but don’t understand how capitalism drives competition which drives margin compression and cost structure change which drives trajectory. It's why you see people like @alexolegimas or @stevehou quickly weighing in to lend some thinking in many cases, or more famously, Cowen's takes on diffusion. Admittedly, I have bias due to crypto, in which a subset of the industry got one thing very right (btc or eth mostly) and then extrapolated the learnings of believing in the first-order long-tail outcome to then get a bunch of things very wrong, or at a minimum, getting the timelines of change very wrong, while telling everyone how they just didn't get it. Obviously these are very different dynamics because the economic value and productivity from AI is so much larger than crypto today, but the level of hot takes and centralization of "those in the know" and how they think feels actually quite similar, and the dogmatic nature does as well. Perhaps this is the academic versus applied thinking, in which edges of academic theory can get direction of progress very right vs consensus but as that runs up against real-world complexities and human complexities, the applied thinking can get the second leg up of trajectory more correct. This shift is perhaps the heart of building an applied research organization. I know nothing about AMD internal politics, and so many companies in times of rapid technological change run into the problem of "we must try to do it our way, and then fail, in order to change our mind", and AMD definitely feels like one of those. But also this sounds very similar to many of the people who have now boomerang'd from "my bigco sucks, time to start my own co or join newco that is cooler" back to "my bigco was right/is good/turns out they had their reasons, newcos are not great." Anyways, am sure people will have lots of feels here and generally the tension across all of these is what makes company building, analysis, and investing so fun right now, especially because there are so many vectors in which one can be wrong, but still be very right, and vice versa. Just love that we get to see it all play it pretty publicly.

2026.08.27 14:00

A bit of a rambling idea space on AI but: increasingly i do feel like these types of things (style/bias tweet below is written in, latest podcast with dwarkesh + dylan, other content) from a subset of insiders are a rorschach test for if you believe hyper narrow but incredibly deep insights win out in how one predicts the future versus one layer up but more well-rounded/mutli-disciplinary understandings. A lot of the takes kind of bridge the gap of understanding capitalism shifts very well but don’t understand how capitalism drives competition which drives margin compression and cost structure change which drives trajectory. It's why you see people like @alexolegimas or @stevehou quickly weighing in to lend some thinking in many cases, or more famously, Cowen's takes on diffusion. Admittedly, I have bias due to crypto, in which a subset of the industry got one thing very right (btc or eth mostly) and then extrapolated the learnings of believing in the first-order long-tail outcome to then get a bunch of things very wrong, or at a minimum, getting the timelines of change very wrong, while telling everyone how they just didn't get it. Obviously these are very different dynamics because the economic value and productivity from AI is so much larger than crypto today, but the level of hot takes and centralization of "those in the know" and how they think feels actually quite similar, and the dogmatic nature does as well. Perhaps this is the academic versus applied thinking, in which edges of academic theory can get direction of progress very right vs consensus but as that runs up against real-world complexities and human complexities, the applied thinking can get the second leg up of trajectory more correct. This shift is perhaps the heart of building an applied research organization. I know nothing about AMD internal politics, and so many companies in times of rapid technological change run into the problem of "we must try to do it our way, and then fail, in order to change our mind", and AMD definitely feels like one of those. But also this sounds very similar to many of the people who have now boomerang'd from "my bigco sucks, time to start my own co or join newco that is cooler" back to "my bigco was right/is good/turns out they had their reasons, newcos are not great." Anyways, am sure people will have lots of feels here and generally the tension across all of these is what makes company building, analysis, and investing so fun right now, especially because there are so many vectors in which one can be wrong, but still be very right, and vice versa. Just love that we get to see it all play it pretty publicly.

SemiAnalysis@SemiAnalysis_

2026.08.27 13:01

BREAKING NEWS: AMD middle managers are wasting over 20+ engineers’ time trying to grow their fiefdoms & get promotions on a useless side project called their “ATOM Inference Engine,” instead of spending that effort on inference engines that actual customers like Meta/xAI actually use, such as SGLang/vLLM. Dozens of AMD engineers, including principal engineers and AMD Fellows, have reached out to us privately about how, due to office politics and backing from a subset of middle managers at AMD, they are prevented from talking internally about their concerns over middle management wasting 20+ full-time engineers’ time on a side project with basically no customers. The engineers on this side project are 10x engineers and instead should be working on customer engines like vLLM/SGLang. The middle managers will try to cope by referencing the singular customer that uses it in prod, Alibaba on Qwen, but when digging deeper, it is just a side-business enterprise unit deploying ATOM in a limited capacity instead of the main Qwen business unit. It is incredibly disrespectful to AMD engineers for AMD management to pull resources & limited internal GPU R&D dev clusters away from engineers supporting customers on production engines like vLLM/SGLang to spend them on growing their fiefdoms & playing office politics.