Real Alpha
Try Pro free

Prof G Markets · 4 min · Jan 16, 2025

AI infrastructure may have a stronger moat than language models

Andrew Ross Sorkin · Editor-at-Large of DealBook, The New York Times

Andrew Ross Sorkin sees sustained demand for chips and energy, but questions whether language models are defensible when competitors can build them quickly.

The episode’s best minutes

Full analysis
  1. 28:10Private equity is stuck until firms can exit investments4 min
  2. 35:12AI infrastructure may have a stronger moat than language models4 min
  3. 23:53The bond market could constrain Trump’s agendaPro pick 3 min
  4. One more moment in the full episode
“And what does that say about the defensibility and the sort of the moat around the large language models?”
Andrew Ross Sorkinat 37:30 · Prof G Markets
37:30

Takeaways

  1. He sees a long runway for AI infrastructure and energy.
  2. Fast followers raise questions about model defensibility.
  3. If models are powerful, other businesses may lose their moats.

About this segment

Andrew Ross Sorkin sees sustained demand for chips and energy, but questions whether language models are defensible when competitors can build them quickly. That puts software valuations and business moats in doubt.

Andrew Ross Sorkin

Andrew Ross Sorkin is editor-at-large of DealBook at The New York Times and co-anchor of CNBC’s Squawk Box.

Founded DealBook and co-anchors CNBC’s Squawk Box; wrote the bestselling books 1929 and Too Big to Fail.

From “LA Fires, Trump Trades, TikTok, and Legacy Media’s Fate — ft. Andrew Ross Sorkin | Prof G Markets”, Jan 16, 2025. Watch the full episode

Answers “What will AI look like in the next few years?” See all 169 answers

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

  1. 35:25

    AI and valuations. Consumer and retail companies trade at 0.5 to three times revenues. SaaS trade at six times revenues. And OpenAI and Anthropic are trading at somewhere, you know, know, between 60 and 80 times revenues. And while no doubt it's intoxicating, exciting, your mind spins with potential when you use these LLMs, I mean, there's a lot of expectation around revenue growth. built into these valuations. What's your sense of the AI market? Are you a bull here, cautiously pessimistic? Do you think it's overvalued? What are your thoughts on AI and valuations right now? So on the valuation front, I still think the chip makers, anybody in the world of infrastructure, frankly, anyone in the world of energy, I think there's a long road ahead.

  2. 36:17

    I think from a directional perspective, just the path is that there's going to be more and more AI, if you will, and more and more need for chips, energy, and everything else. You could make the argument maybe that from a technological perspective, at some point, we'll hopefully figure out the energy piece of this, meaning we'll figure out a way to make chips that require less energy. My bigger question is really on the software side when it comes to valuation.

  3. 36:47

    I wonder whether all of these large language models become somewhat commoditized. I would argue that some of them seem relatively similar to each other. You know, every month one's better than the other, but they're all sort of playing in the same place. It's shocking to me even that when you think about what Elon Musk is doing with XAI, that in a year, if you have enough money and enough NVIDIA chips, you too can create a large language model that is relatively competitive with others. And look, I give their team an enormous amount of credit. I think there's some very smart folks there. But if you had told me that Google and Microsoft and OpenAI and everybody else was working on these things for years and years and years and years and years, and all of of a sudden, literally, I mean, literally what Elon's done is spun this up in 12 months.

  4. 37:42

    It's incredible. And what does that say about the defensibility and the sort of the moat around the large language models? So then the question is, what are these large language models really worth? And are they just a feature? Is it just a feature of everything else? I also think it raises some real questions. If it's so great, talking about valuations, what kind of moat is around every other business? Meaning you either have to take the position that the large language models are amazing, in which case they should have crazy valuations and everything else should be relatively destroyed and worthless because you and I will be able to make an app that competes with any of these things at any moment in time. Or you have to I believe that the large language models are really not where the action is, and it's going to be everything else.

  5. 38:28

    It's probably somewhere in between. Stay with us.