Elliott says Nvidia is in a 'bubble' and AI is 'overhyped'

https://web.archive.org/web/20240802162730if_/https://www.ft...

https://webcache.googleusercontent.com/search?q=cache:https:...

https://cc.bingj.com/cache.aspx?d=3560540516523&w=cpSY8kDyA5...

Programmers have gotten pretty good at hiding latency nowadays and Nvidia makes the most popular high-throughput devices, and has an utterly dominant software ecosystem for those types of devices.

Meanwhile Intel continues to fumble. AMD actually doing pretty ok on the hardware side, but their software ecosystem is not great. Who’s going to compete with Nvidia? Apple, lol?

Just like they thought that the price of real estate would always go up, until it didn't in 2006 or so.

We're in for a correction, a strong correction, in hardware demand. Someone will figure out a better algorithm, a better way to do transfer learning, or maybe a better architecture for AI chips that greatly increases performance.

At the rate things are going, we'll be using 2 bit arithmetic (0,1,infinity,-1) for models and all of the huge multiply-accumulate arrays won't be quite so useful.

I don't see how "line goes up" continues forever.

Sober assessment in my opinion. It is obviously hard to make predictions about the future, but AI, as it exists now, has not revolutionized the world. It definitely makes me some percentage points more productive, but while the overall effect of a widely spread small productivity enhancement might be large, that doesn't means the technology lives up to the revolution it promised.

I feel like we are in a very rude awakening for a crash very soon, we have already seen multiple AI products come out and fail horribly.

In my view there are 4 possible outcomes:

- It is not overhyped but we realize that the current pace of things is un-sustainable for a variety of reasons and investment is decreased.

- It is mildly overhyped but we realize that the vast majority of what these startups are doing can be done by Google, Microsoft, or Apple with just a minor change to their system prompt. And its moved to on device processing threatening Nvidia.

- It is seriously overhyped, beyond AI startups failing we have a major issue like someone dieing or some other major scandal caused by AI doing something and no human verified it. We go back to recognizing what the technology actually is and what its limitations are.

- We hit a wall with the current implementation and it does a decent job at things, but progress just stalls and hype dies down.

I just don't see a positive outcome given the many issues that have come up and it being shoved into places as a source of authority and time and time again, it fails.

Edit: Thank you for the archive. Yeah that aligns.

AI overhyped? What?!?! Has this person used LLMs at all?

Without a doubt, LLM AI is the most useful thing to happen to me. Probably on-par with than the entire internet due to how quickly I can get a response and how detailed/nuanced/niche it can be.

I expect AI to eliminate the need for most diagnosing doctors and pharmacists. If we can't get that back in GDP, its a medical corruption problem. That example alone should be proof that AI is not overhyped.

btw. yesterday I came across withpretzel.com, this is a revolution for working interactively with supercomputers and jupyer.

How quickly or effectively will model creators be able to make their vertically integrated custom training hardware is an open question and determines how you should bet imo.

Nvidia is trying to make themselves a vertically integrated product with a necessary software layer on top, but it’s not obvious if that will succeed or if it’s enough of a moat to hold on to the monopoly long term with the incentives as they are.

Intelligent software is such an obvious capability shift that its dismissal as overhyped is dumb. The web was known to be a big deal in the 90s but it was still hard to bet on and in 99 you would have lost a lot of money in most cases. Does that mean the web was overhyped?

Enormous value will be created and captured - Nvidia has a monopoly now, if you think it’s resilient you should buy but it’s unclear how resilient it really is and it could drop in the short term either way. Major shifts are hard to bet on.

Even obvious bets like Amazon in 99 you’d have to have waited a long time to 10x your initial investment.

My personal take was I bought Nvidia when this seemed obvious a year ago and sold it when it hit 3T. It seemed clear it’d capture all the value on training in the short term, but I’m less certain about the future and 3T is a lot.

Uhh... I was with them on the "bubble" and "overhyped" from a valuation perspective. Without having done the proper math of a valuation model, it does seem (based on gut/experience) that the expected value here is baking in too much certainty when this is still early innings...

But I wouldn't go as far as saying products will never work. It's really hard to make predictions, especially about the future.

  "In the short run, the market is a voting machine but in the long run, it is a weighing machine." 
  - Warren Buffet

Might be a 45-day delay for its public disclosure, although.

Problem is that big tech has no idea how to capitalize on AI! So far their biggest "idea" was how to censor it!