[adastra22]

AI, while undeniably powerful and transformative, is in the midst of the biggest, most insane tech bubble we have ever seen. And nearly all of that money is ending up, directly or indirectly, in GPU data centers. And NVIDIA is the largest cost (profit maker) there.

When that investment firehouse gets turned off, the AI providers will stop building new data centers. Likely for some years. That revenue stream for NVIDIA will go to zero so fast…

The unknown, as with any bubble, is timing.

[dtagames]

Demand for chips has only increased since their invention and never gone down, much less "to zero." Chips are a critical part of the tech business for the foreseeable future, regardless of what happens with AI or any other use case for them. They're raw material for computing, and computing use only goes up.

[adastra22]

NVIDIA growth in data center sales the last 4 years: 2022: from $6.7 billion +58.5% to $10.61 billion; 2023: +41.4% to $15.01 billion; 2024: +216.7% to $47.53 billion; 2025: +142.4% to $115.19 billion

NVIDIA isn’t a startup. It isn’t disrupting a market. It is the ESTABLISHMENT. Low double-digit growth numbers for market leader in established industries would be, by itself tremendously remarkable. Apple was 6% last year, for example. That’s doing great.

NVIDIA grew 142% this year and 217% the year before. That’s… that’s f%#£ing unbelievable is what that is.

The entire consumer market for NVIDIA is less than 10% of their data center market. NVIDIA is a ln AI company with a side hustle in computer graphics. Oh and a nontrivial amount of that is researchers and small companies buying consumer chips for non-LLM AI training and inference, so real numbers are even smaller.

“Zero”, while not mathematically accurate, is indistinguishable here. Elimination of most of the data center sales would immediately move market valuations by trillions of dollars.

[vel0city]

Demand for networking equipment has only increased since their invention and never gone down, much less "to zero".

I'm sure that same phrase was echoed at Nortel and more offices in the 90s.

It's all hot stuff until you have a few billion dollars worth of inventory manufactured that you can barely give away for a million dollars one day. Sure it's not zero, but you're still pretty fucked in the end.

[adastra22]

NVIDIA doesn’t separate their networking revenue, but at time of acquisition mellanox had less than a billion dollars on sales. Less a half a percent of NVIDIA’s current data center sales. That has undoubtedly grown, but I would be surprised if the networking share of their data center business was more than a rounding error. Keep in mind they sell GPUs for $50k, 2-8GPUs per box, and even a state of the art Infiniband card to put in that machine is only a few thousand bucks.

[DSingularity]

You are missing his point.

[adastra22]

I think I replied to the wrong comment, thank you.

[fun444555]

Man, I feel old. I remember feeling this way during dot-com bubble. The few months of unemployment grounded me. Anyway, better positioned this time. History is a good teacher.

[mikkupikku]

This time the bubble has companies firing tens of thousands of employees because they think AI has made then redundant. When the bubble popped the first time, the internet thing stuck around and was a permanent change, popping the bubble wasn't a return to the status quo. I wonder how it will work out this time.

[signatoremo]

Feeling which way? Demand for chips are much higher than it was at the peak of the dot-com bubble. That was a painful period, but only a blip in the history. Everything considered “hype” back then has become reality, at a scale that was beyond any dot-com era expectation.

[chii]

> most insane tech bubble we have ever seen.

the current "bubble" hasn't surpassed the dot-com boom yet.

[adastra22]

The total aggregate loss of value from top to bottom of the dotcom crash was about $5 trillion. That’s the current market cap of NVIDIA alone, to say nothing of other AI companies. So yeah, it has.

[chii]

that number back in 2000 is about $9.7 trillion adjusted for inflation. You can't merely just compare the numbers at that time with a number today. It's meaningless.

You have to compare forward earnings to share-price ratios.

[adastra22]

I think you missed that I'm comparing a single company to the entire market crash in 2000.

What about OpenAI, Anthropic, xAI and the other foundation labs that have collectively raised trillions?

What Microsoft, Amazon, Google, and Meta, which would likely survive but maybe lose up to a trillion each in valuation?

What about the very long tail of venture-backed AI companies that will go bust? You might complain that the dotcom number was just public companies, but back in 2000 everything was a public company. A company with thousands of high-earning employees going bust matters to the greater economy whether or not it is Nasdaq listed or not.

If a single company represents the entire dollar amount of the dotcom bust, or half when inflation-adjusted, and that valuation is entirely predicated on that growth continuing at historically unprecedented rates.. yeah we're in a bubble, and the damage when it bursts is going to be big.

That was the point I was making, and I fail to see how forward earnings to share-price ratios has any relevance here. The whole point of a bubble popping is that the market suddenly finds out those forward revenues were a mirage, a house of cards, and are very much made up.

[chii]

> fail to see how forward earnings to share-price ratios has any relevance here.

the relevance is that these earnings expectations are lower than when the dotcom bubble happened.

The fact that a single company can have a market cap today that is greater than the losses from the dotcom bust is irrelevant. We have more wealth today than back in 2000, and these market caps reflect that.

[WastedCucumber]

>the relevance is that these earnings expectations are lower than when the dotcom bubble happened.

[citation needed]

[Libidinalecon]

You also can't just leave out the bond market.The dotcom bubble sat on top of the 10 year bond at 5%. The US had a balanced budget and 10X less debt than now. It was the peak of the unipolar moment. Part of the AI bubble is the lack of safe investments to act as a flight to safety. I just don't think they are really comparable. The AI bubble could certainly be more destructive given the circumstances and concentration when it pops, even if not bigger. The other main difference is we actually had the internet in place. So much of what is fueling the AI bubble doesn't even exist.

[positron26]

Maybe the most insane tech bubble you've seen. We'll have to wait and see how the superbowl goes.