Is AI hype or Does it Work?

  • A few links/stats on the topic.
  • Github co-pilot in this study led to substantial savings (30-50% time saved or reduced in some tasks).
  • Bloomberg Odd Lots podcast episode with the CIO of Goldman Sachs on the use of AI.
  • Good overview of current revenue run rate for gen AI products (chart).
  • A rather bullish tour of AI and semiconductors by Gavin Baker.

Start-up Failures

  • Are rising sharply according to Carta (just one data source).
  • This is hurting VC funds of certain vintages “Only 9 per cent of venture funds raised in 2021 have returned any capital to their ultimate investors, according to Carta. By comparison, a quarter of 2017 funds had returned capital by the same stage.

Pricing Data

  • Fascinating read on how to price a data asset.
  • Relevant especially with the rise of AI. At first quantity matters here but “as training sets grow ever larger, it’s often more efficient to do this than to acquire the next token; beyond a certain point, data quality scales better than data quantity“.
  • So there you have it: 5000+ words on data pricing. We’ve covered use cases and users; quality and quantity; internal and external value factors; pricing axes and maturity curves; table stakes and usage rights; and much more.

GPU Semiconductor Content

  • Nice slide from KLA.
  • So I tried just to put together this chart to show how different the GPU package is between 2015 & 2024. So of course, the B100 chip, the GPU introduced a few months ago, and this is not enough because Jensen has already introduced the next generation of GPU last week” (h/t The Transcript).

Dominating Google

  • These 16 companies dominate Google search results. Odds are you haven’t heard of any of them.
  • Across 10,000 terms where affiliates are ranking, which cover products in every niche you can think of (home, beauty, tech, automotive, cooking, travel, sports, education and many more), these 16 companies ranked on the first page of 8,574 (or 85%) of them.
  • Detailed write-up here.

Altman’s $7 Trillion

  • Back-of-the-envelope analysis of why Altman wants this sum to build semiconductor capacity and why it isn’t such a crazy number.
  • It’s a useful reminder of what it will take for AI to scale in the coming years.
  • The article also links some more serious analysis of the trend in the cost of training AI (like this).

Semiconductor Manufacturing

  • Interesting analysis on staying competitive in semiconductor manufacturing.
  • The dotted black lines toward the bottom show the estimated cost of building a leading edge fab (the lower line) and a line showing double that number (the upper line). Our thesis is that companies whose annual revenue fall between those two lines are at risk of falling off the Moore’s Law treadmill.
  • TSMC came close once. Samsung looks close now (though this doesn’t include the rest of the group subsidising the fab). It also shows that Intel’s plans to offer fab services need to succeed.
  • Source.
WordPress Cookie Notice by Real Cookie Banner