Finally or unfortunately? We (customers, hackers) get further and further from owning hardware and running software on our terms.

They are still selling desktops, laptops, and ordinary servers (and some quite extraordinary ones). It's just that their AI systems are selling like hot cakes.

Which is more or less everyone's AI systems. The appetite for AI training and inference seems to be exploding. My best guess is that it'll be like the dot-com bubble, but this time we'll get cheap Nvidia DGX rack machines being sold by asset recovery services.

In a few years when all these rack servers are outdated for AI workloads, there will be a glut in the market of cheap only-slightly-dated servers.

My JellyFin setup in 5 years is going to be fucking awesome.

Same, after getting into homelabbing, "post-lease" became my favorite word.

When this bubble bursts, we'll have a lot of number crunching capability being sold for peanuts.

Looks like Jaron Lanier's predictions about servers being the new form of power are coming true. Companies are predictably moving to remove the power from the hands of their chattel.

https://en.wikipedia.org/wiki/Who_Owns_the_Future%3F

I'm still not sure why anyone would buy a dell server. Supposedly xAI buys from them, which probably accounts for this, but it is generally a much more sensible choice to buy from companies that are lower-cost and less focused on selling you "enterprise support" (supermicro et al).

Nobody ever got fired for buying Dell with ProSupport. I buy Dell servers at work because it's not my money but it is my ass on the line if something goes wrong. The quotes you get from your rep look nothing at all like the retail pricing on Dell's website.

Is this the support I've heard about from Dell? I've always heard about it from a consumer perspective, but I've been told that they'll come out to you same or next day for hardware support? If that's the case, that's pretty damn impressive.

Yes, they have both next-day and same-day on-site support plans. For hardware failures they will bring the part to your office/data center and perform the replacement. We're in a third party colocation facility and Dell is happy to dispatch techs there; we don't have to be present.

Had this for a work laptop a few years ago... That said, I think it was a design flaw in that model, had to have the MB replaced twice under warranty, the third time it was out of warranty and I just got assigned a new laptop.

My daughter took her Alienware laptop to school with her. The keyboard broke and we contacted Dell support hoping to just buy a new keyboard. Instead, they sent a technician to her dorm next day who fixed it on-site.

What really impressed me was that we bought the laptop in the US where we live and she was going to school in Canada.

That’s really cool! Was this just with the standard warranty, or did you have to get some sort of extended support plan?

It was the standard warranty which makes me think there was a known flaw in those keyboards.

Interestingly, I got nowhere with phone support. The support person told me there were no keyboards available and didn't know when it would be back in stock. So then I contacted Dell support on Twitter (as it was known back then) and they immediately got back to me to find out what happened and arranged the repair.

At scale, the savings of slightly cheaper hardware is often dwarfed by supply chain management costs, both direct and indirect. How much is it worth to your business to get the hardware deployed six weeks sooner and/or with a high probability of timely delivery? I imagine xAI wanted to move with maximum speed.

This is a big part of the decision process. North American supply chains are slightly more expensive but also fast and responsive. The advantage of someone like Dell is also that they can do almost everything in-house, which avoids the overhead (to the business) of coordinating integration of components manufactured in Taiwan, the US, etc into the final build. I’ve done it both ways. There are real tradeoffs so context matters.

In todays market I think there is a lot more consideration to speed of delivery over price. Dell has historically been NVIDIAs #1 OEM partner so it would not surprise me if Dell has supply chain advantages over other server vendors.

My experience with Dell is that they are not that focused on selling enterprise support (at least compared to HPE), at most they will push for bundling hardware (cables, cable trays, front covers, PERC...) that you do not really need in order to get better volume discount.

Price-wise I don't see a meaningful difference between Dell and SuperMicro (or even "non-traditional" server vendors like Asus and Gigabyte).

To me the OOB/BMC interface of the server is the main tangible difference not the support contract implementation.

IDRAC is miles ahead of SMC/MEGARAC. It's not even close.

Probably iDrac. AI servers are mostly just white boxed nvidia reference designs. Not much markup.

HPE is finally moving away from their 'Cray' systems which are just whitebox AMI systems. You get none of their iLO etc even though they are 'HPE'

> buy from companies that are lower-cost and less focused on selling you "enterprise support" (supermicro et al).

From past experience, vendors like SuperMicro aren't much cheaper at scale because Dell can give 80% discounts on volume, and most Dell server sales are done via Channel with an MSP who will manage and administer the system.

Ofc, this continues to reinforce my belief that the "AI boom" in it's current form (and once you remove the scooby doo monster mask) is basically a "Datacenter/Telecom Bubble" 2.0 like back in 2000-01. The multiples, messaging, vendors, and margins are almost the exact same.

During the Dot-com Boom, Telecom and hardware companies were also expanding DC and telecom capacity massively (to surf and host a website on the Internet, you kinda need Internet and web hosting), and when the Dotbomb happened, the Telco bubble collapsed subsequently as well (remember WorldCom/MCI?)

That's why the 2000s were horrible for anyone with a CS/CE/CSE/EE degree, because both software AND hardware industries collapsed. Imagine a world in the late 2020s where you cannot land a job as a Fullstack Engineer OR an ML Infra Engineer - that was the 2000s except with older stacks.

That's also why I'd be optimistic if a bust happens - the overcapacity in compute that arose from the Telecom and Dot-Com Busts both helped usher the Cloud, SaaS, E-Commerce, and Social Media boom because the infra has been laid and became cost effective. It is also in this context that Paul G's "cockroach" and "ramen profitability" essay came to the fore.

> That's why the 2000s were horrible for anyone with a CS/CE/CSE/EE degree, because both software AND hardware industries collapsed

Odd, anecdotally I remember people saying that, but had no issue getting a job in 2004 as a fresh CS grad.

Also anecdotally: by 2006 I also don't think it was that hard, in SF. 2008 of course kicked off a new cycle of capital scarcity, but, what it seemed like to me was that two gold rushes kicked off in earnest around that time which in tech seemed to mitigate any major difficulties for software people: First the online/Facebook gaming goldrush (Zynga being the poster child, but there were dozens more publishers, advertising and monetization firms drinking from that trough) and of course the App Store goldrush.

In 2008 I remember being in tech was able to largely ignore the recession that everyone else was getting reamed by because mobile, Facebook games and generally social media, and general Web 2.0 was getting big investment across the board. The low hanging fruits still had not all been plucked.

If I had been smarter I would have bought up property at what I now know were the lowest prices I’d ever see, but alas, I did not

Having started my career in 1999 (and if you could spell C++, you were pretty well guaranteed at job at a good salary), I remember two things...

- 2000-20001 - "small" recession, along with the dot-com bubble bursting. Lasted through 2003 or so, though the bulk of job loss (across industries) was 2001 into 2002.

- 2008-2010 - housing market collapse - world-wide impact. Most of us probably remember that period. It as brutal for everybody, not just STEM grads. Too big to fail and all that hocus-pocus.

I was RIF'ed in Dec 2001, took a few months to find a job I wanted, but wasn't all that bad given I was pretty darn junior at the time.

Yeah, I was/am in Phoenix where it seemed to trail about 8 months... it totally dropped here after 9/11 though. I remember effectively having the plague looking for work through 2002. Was working pretty regularly again by the end of 2003 though... those were a couple rough years though.

The thing is, the bust froze everything for 1 or 2 years, which sucked for regular IT folks plus this time there also a huge oversupply of IT graduates. I wouldn't be surprised if there are 10x more people working in IT then back then.

Mind you, I was just a kid when it happened so I'm basing it off of 2nd hand stories, but my understanding was anyone with a pulse (ie. A hs grad or a college dropout) could get hired into the industry at the time (eg. The KOTH joke that if Peggy was laid off from being a substitute teacher she'd learn to design software).

At least ime in the 2010s and 2020s, while you had some aspect of that with bootcamp grads, it wasn't to the same degree. Or was it? Like I said, I was a kid when the busts happened so I have no frame of reference.

I had a pulse and got into this industry in the 90s through the back door.

It’s been a long run and I’ve seen busts come and go. It’s worked out really well for me, even while I’ve watched a surprising number of highly-credentialed new hires flop out over the years. I continue to appreciate people with oddball resumes.

Similar here, no formal higher education starting in the early-mid 90's, just a lot of reading/learning and experimenting along the way... Even while working, I've spent 15-20 hours a week most of my career on learning activities.

Somewhat hardware oriented and worked a few tech support/helpdesk and internal IT jobs in the early-mid 90's when I got more into programming as a side-step from design/art.

Also anecdotally: back in my sysadmin days, the Supermicros we had had so many more problems than the Dells. Granted this was 2012-2015 era hardware being run in 2018, so I can't say for certain if that's even remotely still true, but yeah.

The 'Micro's ethernets were all shot by the time we closed up shop so each had an expansion card for 10G ethernet at that point, and one we had to run VM management on 10G because the 1G we used elsewhere had shit the bed in that unit somewhere along the way and we couldn't be bothered to buy yet another card and tear down the server again for it. Plus management traffic was negligible.

I don't miss that job. Fuck being on-call.

For the international hardware warranty transfer.

You want enterprise support. GPU quality is atrocious, you will have your Dell tech in there replacing GPUs and fans all the time.

Can second this, the amount of GPU failures we have with Lenovo systems on just <50 nodes is significantly higher than we expected. Having a Lenovo support person at least twice a month on premise at the middle of the bathtub curve is probably also costing them (and implicitly us) a good chunk of money.

Interesting. I work in higher ed and we have thousands of GPUs under my team. Rarely ever seen a failure. Mostly when we put consumer grade GPUs in servers (Nvidia doesn't like this). True server-grade GPUs never have any problems.

IS this for some kind of HPC cluster? What kind of utilization are you at? For an AI company these GPUs are going to be at near 100% utilization 24/7. These kinds of loads destroy hardware quick.

Every site I've worked at has plenty of GPU failures. Not consumer grade either, H100/A100

The paperclip maximizer story, where an AGI directs all economic resources towards manufacturing paperclips, was wrong. We have the the graphics card maximizers, where humans voluntarily redirect vast economic resources towards generating a maximal stream of tokens that hardly anyone wants. Or perhaps call them token maximizers.

There used to be the notion that "talk is cheap". Now we are spending $trillions on generating idle talk.

I always found this a bit of an omission in Bostrom’s argument in Superintelligence - on one hand he’d say we should avoid giving AI a goal like paperclip maximisation while overlooking that the method of transistor maximisation used to attempt to build the Superintelligence was the same kind of goal. Same goes for whatever system he fantastically proposed to run his fanciful “mind uploading” Cartesian scenario.

The difference is the demand for paperclips is already met. Each additional paperclip is almost pure waste of resources.

The demand for silicon, even outside AI is not, and advances in silicon are going to benefit pretty much every industry in the long run, even if during the short term they are distorted by the AI demand.

> ...that hardly anyone wants.

Meanwhile, in reality, ChatGPT is the fastest growing consumer product ever and LLM provider revenues are like superexponential.

Threads broke the record within 5 days of launching, surpassing ChatGPT adoption, has less pressure from competitors, and Threads very likely makes money for Meta.

On the other hand, OpenAI’s ChatGPT has more intense pressure from competitors, isn’t making any money, and costs are still rising for its operation.

I don’t know where you’re getting the idea that it’s the “fastest growing consumer product ever” and “revenues are like super exponential” as it’s demonstrated repeatedly that OpenAI has yet to turn a profit or even meaningfully dent their burn rate

The first artificial intelligence was the market. Just like the rest, it produces some wonderful tools, but putting it in charge of everything is a sure path to some kind of paperclip optimizer.

> Now we are spending $trillions on generating idle talk.

We were spending $trillions on generating idle talk before AI as well. It was just done by meatbags.

How long do you think we have before the AI Bust? This isn’t a rhetorical question, I’m looking for estimates before I have to exit all financial markets and go full defensive in my portfolio. Can it make to 2026? Or will the bust be several years from now? I know no one really knows, but estimates can be very helpful. Sometimes you can feel the wind change.

If you are feeling the wind change, then why not go into defensive mode today?

Tuesday in three weeks, possibly.

The kind of person who can time market crashes is not the kind of person who comments on HN on a Wednesday morning. Market booms and crashes come and go - such is life.

But HN is the kind of place where someone notices that less gpu orders are coming in or that Donna from HR said the CEO said X or “we can’t get anyone to listen to our AI pitch deck”.

Share your thoughts if you are a decision maker or have any insider knowledge? What do you think about the bubble?

Bubbles don't pop the moment critics start calling them bubbles, they tend to have a couple more years in them

The day they do pop, there is no further warning. News comes out, and the market reacts billions of microseconds before retail learns about it, an eternity.

The dot com bubble took about 9 months, iirc. Of course, that’s my memory. What do others think?

Longer than that... I remember the stories when F-d company posts started coming up, recession camp activities and it started to spread around the later part of 2001. A lot of projects outside CA/WA were still holding on, but funding completely dried up and "future work" was all but dropped after 9/11 as the final tipping point. IMO the tipping point was that September in 2001, but the signs were there upwards of a year sooner.. and a lot of recovery didn't really get rolling until 2003.

Just my own observations and memory.

The tipping point on the dotcom bubble itself was the NASDAQ market peak on March 10, 2000, the tipping point on the broader economy that the bubble bursting helped contribute to was the economic peak in March 2001 that marked the start of the March-November 2001 recession.

A number of government policies adopted before the bubble burst under Clinton and before the bubble bursting had worked its way into the public consciousness under Bush (safett net cutbacks, downward tax burden shifts, etc.) made the November 2001-December 2007 aggregate expansion after the 2001 recession and before the Great Recession have very poor distributional characteristics, such that much of it, especially the first couple years, felt like a recession for most outside of a narrow elite, which probably combines with the emotional impact of 9/11 to fix that late date as the beginning of the downturn, even though it was near the end of the downturn in aggregate economic terms.

> > Bubbles don't pop the moment critics start calling them bubbles, they tend to have a couple more years in them

> The dot com bubble took about 9 months, iirc

It took 39 months and 5 days from Fed Chair Alan Greenspan’s “irrational exuberance" description (1996-12-05) until the bubble popped (NASDAQ market peak on 2000-03-10). Greenspan wasn’t the first critic, either, though he was obviously a very high profile one.

A bubble that pops is caused by an event that triggers a panic wherein people are looking to sell and are happy to reduce their price of selling sharply in order to dispose of an asset.

The question we have to ask is - what event will trigger this? To the extent that marginal investors (who who are price makers) will seek to exit.

The thing is, the “event” can be as simple as normal market fluctuations, because once idea of a potential bubble becomes sufficiently widespread, any momentary downturn in any indicator will be seen by some segment of the market as a sign that the burst is here, and if that segment is large enough, it becomes a self-fulfilling prophecy.

The bubble pops way earlier than the news comes out. You can see it in the investment companies as their analysts start heavily promoting whatever is about to pop to keep stock moving while divesting at the same time. Last thing they want is to be holding the bag.

Often there's some kind of event that starts the pop, like when Elon Musk slashed the headcount at Twitter and everyone else saw that as a signal that the hiring frenzy was over and it was safe to start cutting staff. There was lots of discussion before that day about how out of control tech hiring had become but the money was flowing and everyone wanted to look like a great place to park it.

My guess as to what happens here is one of the big players showing that they're cutting back on capacity in a significant manner, which will spook the investors who know a bubble pop is coming but want to time their exit as near the top as possible. By the time the pebbles notice, the avalanche will be in full roll.

That's part of it... I would say the real end is closer when you start seeing F*ckedCompany style articles about $Startup closing doors on employees, or hardware repossessions. It was pretty common through later 2000-2001, the finial pop imo was following 9/11 well after a lot of the startup market fizzled, but a lot of more traditional employers started cutting projects/staff too.

Can you define what a bubble bursting means? whose market cap do you mean when you talk about this? I don't think OpenAI or Claude's market cap is going down by more than 20-30% anytime soon.

I'm hoping we get some good competition for AWS that comes out of this. This is probably the largest data center build out push in a long time.

It would behoove a lot of these companies. There again say $13 billion series f to make an AWS competition stack.

I mean you have all the infrastructure to magically write the code for you right?

I'm skeptical, it's not like the competitive offerings we already have don't bring anything to the table. If anything, GCP is a lot more pleasant to use. Yet AWS still maintains its dominance, imho out of inertia. No org wants to make "Migrate to [Other Cloud]" its main project for the year for some small savings or "because we've heard great things about it." The opportunity cost is too high. And new firms who spin up go with the cloud the founding engineers know (have used before) because they want to ship asap, not to learn another cloud provider. So I predict AWS continues on like this for decades.

Alternate title: Dell's PC business keeps shrinking.

From TFA:

"Don’t blame the PC business. Dell’s Client Solutions Group had $12.5 billion in sales in Q2 F2026, up seven-tenths of a point, but net income was up 4.7 percent to $803 million, representing 6.4 percent of revenues and marking the highest level seen in the past seven quarters"

Was the word 'biz' used for a character limit?