With AI Boom, Dell's Datacenter Biz Is Finally Bigger Than Its PC Biz

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.

> ...that hardly anyone wants.

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

> that hardly anyone wants.

People continues to mis-weight the benefits of AGI / LLM or in the case "graphics card maximizers". It is precisely because of this, we can continue to extend semi-conductor improvements for at least another 5 years, brings in 1200W CPU in server and liquid cooling along with new DC and Rack design 5 years earlier. Partly continues to fund Gaming GPU where the industry itself were not able to sustain it. New improvement to Networking Gear due to heavy use of LLM. HBM profits that drives current EUV DRAM innovations.

As far as I can tell, I see very little down size to all these investments where Big Tech would have spent those money on stock buy backs or dividends.

And this is assuming AGI really is a bubble and brings little to no productivity, and exclude all the improvements that is adjacent to it.

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.

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.

> 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.

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.

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).

We buy them because our experience is that they are extremely reliable and their iDrac management system is better than the alternatives, which saves us time (thus money). Maybe they aren’t the cheapest at initial purchase, but less maintenance and the ease of administration makes up for it.

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

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.

For academia, I've found that Dell seems to seek and get sole-supplier contracts with universities, so research groups are forced to use their grant money to buy from Dell, often at inflated costs.

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.

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.

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.

While it's great that speculation moved from "when will AGI come?" into "when will those companies blow?", I don't think they will blow any time soon.

If you have been following things, you may have noticed that there's no large economic sector out there consistently returning investments. And the investors money isn't just disappearing nor any such investor just deciding to spend everything in some beach vacation somewhere.

Until that changes, bubbles will keep growing. If the AI one deflates for any reason, it will only serve to inflate some other one.

Tuesday in three weeks, possibly.

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.

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

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'll give this a shot, as I spend my days managing a portfolio exposed to sector rotations on the mid-length timeframe.

______________________________________

It is not ideal to think in terms of amorphous projections where you sum up every probability. Try and consider at least some path dependence and realize that there is a risk distribution.

What you are specifically looking at is the scenario where there is a) an AI bubble in the public markets (remember, much of AI of it is marked numbers in the VC world) and b) the possible scenario where the grey swan gets big enough and the risk of violent unwind is high enough to where you see arguable value in shifting to defensive positioning in a somewhat diversified portfolio.

I'll assume you are weighted heavy to QQQ/NASDAQ in the portfolio. If you have a truly diversified portfolio with rules for systematic rebalancing, you never manually shift to defensive (obviously what I and most advisors would recommend).

The first note is that an "AI bubble" in public markets, assuming there is one, is very much tied to the overall state of high market concentration in the trillion dollar tech names. To see a violent AI bubble unwind, it will almost necessarily involve the "concentration bubble"/Nifty 50 2.0 unwinding as well. That means the AI Bubble narrative in public markets is tied to overall market liquidity just as much as the AI specific factors. The passive complex will put a large continual bid in for NVDA/GOOGL/AMZN/etc even in the case of an AI downturn, which should neutralize the violent unwind scenario unless both impulses turn negative (active money moving out of AI, and passive money pulling out of broad based ETFs entirely).

Of course the relationship above is reflexive, which is to say that NVDA gapping 20% down on earnings may well spur wider outflows (as you personally alluded to when your response would be to "go full defensive in my portfolio"), and vice versa. So what you want to be watching is specifically for signs of this contagion spreading.

If liquidity/flows strongly turns negative, the concentration bubble unwinds, NVDA and co tank, and that likely triggers a sell-off in the AI speculative names and qualifies as an "AI Bubble Burst". Therefore you want to be watching general economic weakness like recession signals, layoffs, etc. Bond yields blowing out will also likely kill the AI bubble, as it's capital intensive. Even with the fortress balance sheets of the hyperscalers, the long end blowing out (it's starting to press into dangerous territory currently) would likely kill a hypothetical "AI Bubble". So watch bond yields (and bond volatility / MOVE index).

More specific areas I would be watching is Hyperscaler Capex projections in earnings calls (watch it like a hawk), and some aspects of the trade war such as global digital services taxes in particular (because profit shocks for hyperscalers will likely result in reduced capex to preserve their earnings targets). I would also be watching competition for NVIDIA in the more grey swan areas, like Huwawei revealing a chip with twice the performance per watt as NVDA chips for half the price, and a surprise software stack that kicks ass (so the low single digit outcome areas regarding Chinese competition particularly, which wouldn't actually be bad for AI, but would likely result in a shock move of money leaving US related AI names on public markets, which would be a repeat of the February to April move we saw earlier this year in response to DeepSeek. Markets have memory.)

So let's sketch out some risks over the next few years. Now we get to the part that usually adds little to no value (making predictions about where markets will go). Hint: nobody can do this broadly. Sometimes one may have specific value-adding strong convictions picked out of the universe of possibilities, but those are the exceptions.

I'll even put a % chance guess since that's what you're looking for, but it adds literally no value.

Bonds blowing out: low/moderate likelihood. 15-20% over the next few years for a major bond event imo, and I'd carry that directly to a 20% chance of an AI unwind.

Heavy economic downturn: moderately likely. 25% or so.

And since the scenarios overlap somewhat (stagflation would involve both, and it's a leading wider risk currently), let's move the risk to 30%. But remember that this is a risk scenario that will involve market cap weighted passive indexes as well, so we're really talking about market downturn risk rather than just the AI Bubble popping concurrently, since the latter is less meaningful in comparison even when we're focusing on portfolios.

Now on to the AI specific scenarios, which would be AI falling out of favor with consumers. I personally put this under 5%, so I have strong conviction this won't be the case. I'm mostly disregarding this risk. Most would probably put it closer to 20%, and some would be 50/50 if they think it's a fad. You will find a wide array of predictions here.

How about tech industry specific shocks like global digital services taxes causing a sharp drop in AI investment from the huge balance sheets driving it all? I think it's fairly unlikely as well, maybe 5-10% or so. I have high certainty that Google and Apple in particular are going to have great cashflow for the next few years. There are some side-lines to watch here though. Ex: the US government itself is strongly fending off digital services taxes in trade negotiations and using a big stick, but if the tariffs are confirmed to be nullified (recent court ruling), and tariffs are unwound, the world may use the opportunity to tax the multinational American tech companies while the US is stunned and weak, so to speak, especially if some of the other lines are playing out like bond yields getting dangerous, which puts pressure on global govs to raise revenue. That's an example of the sort of thinking that is required to really position a portfolio for things like esoteric crashes in specific sectors and have positive expectancy. You have to take into account all of these self-reinforcing feedback loops and get creative/contrarian to some degree, and find out where the market may be underpriced/offer value hedges and hidden synergistic correlations. Or you can pay up for generic tail protection/put options, and just pay the tax to cover the risk. In your case, perhaps that would be a good option? Just pay up for a put spread on NVIDIA that has a 10 to 1 payout ratio, spend 0.5% of your portfolio, and hey, you get a 5% payout in the case of an AI unwind (you don't need to go overboard, a 5% cash payout is a huge gift in a market crash while most are in panic).

On NVIDIA in particular. I do think there's a moderate risk that NVIDIA is in an outright bubble currently. I could easily see a scenario where guidance comes in weak, margins come down for some reason (even something like a string of bad tape-outs at TSMC... could be anything), and NVIDIA re-rates lower. That would be painful for the entire market, and this is the risk I would be paying up to hedge (I'm currently hedging this a bit). I think there's a good 1/3 chance that NVIDIA tanks sometime over the next few years, but half of these scenarios would just be contained to NVIDIA and more to do with margins coming down to earth a bit (wow, this is quite nice just to be able to conjure up BS % numbers like that).

____________________________________

So in closing, just worry about having a properly diversified portfolio that self-rebalances according to proven rules. Ask yourself if there was an AI unwind, how would my portfolio look, and does this aligns with my financial goals and risk tolerance? Also ask yourself what would happen if it further picked up steam, since upside risk/opportunity cost is important as well. As previously mentioned, you are concerned with a scenario that closely correlates to index concentration as well, so also ask yourself how it looks if the current market leaders perform badly, and other sectors/market factors drive performance going forwards.

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.

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.

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 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.

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"