The AI bubble today is bigger than the IT bubble in the 1990s

Have you looked at a graph of Twitter revenue and profit recently (hint: it is very low). Nobody has ever claimed that you cannot fire all the employees and keep Twitter online with a skeleton crew; the claim has always been that you can't do that _and remain a viable business_.

That was because Twitter was massively overstaffed. Are all these other businesses also overstaffed? The result of cheap money? Are they all atu

Anyone that ever worked in a sufficiently large tech company knows that it can keep the lights on with a skeleton crew. It doesn't mean the results will be good or desirable, but it can be done.

Over time cracks will start to show. New featurs are seldom delivered and half-assed. Bugs will take a long time to fix or just become part of the landscape. That sort of thing.

Twitter is a special case that Musk did not buy it to make it into a profitable company, or to make it more valuable, or to improve it in any way. He just bought it because he wanted to rid it of the "woke mind virus" (and factually tried to get out of the deal and was sued into actually buying it).

CEOs are doing layoffs for a variety of reasons, but mostly to improve profits by reducing costs. These cycles happen, and what it actually points to is that those companies are not seeing a viable path for growth at the moment that justifies their valuation. Layoffs with some vague gestures towards AI is a good smokescreen for the time being.

NVDA has a P/E of 55, which is definitely elevated, but nowhere near the 230+ that CSCO had at that time. TO say nothing of SUNW.

The big AI labs are definitely losing money, but they're doing it on the back of tens of (rapidly growing) billions of dollars in ARR, versus the dot com e-commerce and portal flameouts who would go public on (maybe) a million in revenue, at best.

We also have large AI teams at FAANG who are being funded directly by the fat margins of these companies, whose funding is not dependent on the whims of VC, PE or public markets.

These times are not really comparable.

With the internet, there was a clear value proposition for the vast majority of use cases. Even if some of the specific businesses were poorly-conceived or overly optimistic, the underlying technology was very obviously a) growing organically, b) going to be something everyone used & wanted, and c) a commodity.

All three of those parts are vital for a massive boom like that.

Generative AI is growing some, yes, but a lot of the growth is being pushed by the companies creating or otherwise massively invested in gen-AI. And yes, many people try out ChatGPT's webapp, but that's mostly a gimmick—and frankly, many of the cases where people are attempting to use it for more are fairly awful cautionary tales (eg, the people trying to use it as a therapist, and instead getting a cheerleader that confirms their worst impulses).

Gen-AI may be useful to some people, but it's not going to be a central feature of most people's lives—at least not in the forms it exists in today, or what can be clearly extrapolated from them. Yes, it can help some with coding—with mixed results—but not everyone's a programmer. Not everyone's even an office worker. The internet has obvious useful applications for a plumber or a lawyer; if I hired one of those and they said they were using generative AI to help them in their work, I'd fire them instantly. There are already a bunch of (both amusing and harrowing) stories of lawyers getting reamed out in court for using gen-AI to help them write legal filings.

OpenAI may or may not have a robust moat—I've seen people arguing both ways; personally I suspect lean slightly toward the "not" side—but generative AI as a whole is not something that's an interchangeable commodity the way internet access, or even hosting, is. First of all, in order to use the models that are touted as being advanced enough to actually look like more than spicy autocorrect, you need a serious GPU farm. Second of all, AFAIK, those models are being kept private by the big players like Google and OpenAI. That means that if you build your business on generative AI, unless you're able to both fork out for a massive hardware investment and do your own training to match what the big boys are already doing, you're going to be 100% dependent on another specific for-profit company for your entire business model. That's not a sound business decision, especially during this time where both the technology and the legal aspect of generative AI are still so much in flux.

Generative AI may be here to stay, but it's not going to take over the world the way the internet did.

Curiously, nvidia's P/E ratio is lower than it was two years ago!

Most of the LLM applications are either entirely useless or trivially reproduced with much simpler free models (or even entirely non-“AI” methods).

I would love to know how much of the inference I pay for is being paid for by VC cash: I suspect a lot of it.

https://www.youtube.com/watch?v=IUK6zjtUj00

This is not remotely true. I mean it's so incredibly not true I wonder how you came to believe this.

Haven't you ever heard of how hedge funds pay for cellular data to understand retail store traffic, or how satellite photos help them estimate the fullness of gas tanks at ports to predict pricing?

Or how data about predicted electrical pricing based on usage helps factories schedule energy-intensive production during times of low pricing?

Or how aircraft maintenance companies like AAR rely on "big data" to position replacement parts in a globally distributed system of warehouses to reduce the time it takes to repair aircraft (their contracts are based on airline uptime), thereby reducing passenger delays due to mechanical issues?

Or how farms use weather and satellite data to deal with droughts, identify areas to spray, and estimate competitor yields for the purposes of planning?

Or how governments now conduct surveillance of pathogens and drug use through sewer water data?

Or how semiconductor companies use massive amounts of data collected from production line sensors to massively increase yields and reduce chip prices, despite the complexity of chip production having increased massively?

You benefit directly or indirectly from companies using data all the time.

Bubbles looks very impressive until they pop, most fall for them, that is why they are bubbles.

During peak crypto madness vagueposting was an extremely effective market manipulation tool. I know people who made a lot of money on unconfirmed rumors in hours but of course it was just zero-sum gambling - the ”early adopters” made their money at the expense of the latecomers. No value was generated.

People don’t even need to be convinced that AGI/ASI is near, just ”but what if there’s a chance?”. It’s similar psychological tricks as selling lottery tickets.

https://www.quiverquant.com/insiders/1780458/Julie%20Wainwri...

Really, someone, just show me how you vibecode that seemlingly simple feature https://github.com/JaneySprings/DotRush/issues/89 without having some deep knowledge of the codebase. As of now, I don't believe this works.

or the "humans make mistakes too" crowd

or the "just wait, we are at an inflection point in the sigmoid curve" crowd