Investors are genuinely concerned that these companies are in the midst of destroying some of the most successful, scalable, investor-friendly business models the equity market has ever seen.


Investors are genuinely concerned that these companies are in the midst of destroying some of the most successful, scalable, investor-friendly business models the equity market has ever seen.
You must be logged in to post a comment.
I do wonder if this will play out similarly to Meta’s ill-fated metaverse venture (except now we are dealing with a real product). What would happen if these companies decided to significantly scale back their data center ambitions and let the cashflows become the free-flowing river systems of yesteryear?
I think the distinction is that, then Facebook, decided the Metaverse was going to be a big deal before validating that theory. Conversely, there isn’t a business on the planet that isn’t trying to figure out how to harness the capability of LLM’s to give them an advantage. Because the scale of the demand is so great, the capacity required to meet that demand is consequentially impacting balance sheets. If the demand wanes, this will end as badly as everyone expects. If the demand continues or grows (owing to the ease with which anyone can stand up a digital AI native product) the investment will seem minuscule in comparison to the returns.
I’m in the demand will continue to grow camp. LLM’s for all of their faults, enable humans to scale beyond their own limitations in ways unimaginable 5 years ago. For businesses, harnessing the capability in a reliable way is proving complicated, but the drive to achieve this outcome will likely make it a reality.
This one Google could find:
Current AI Share of Data Centers
• Workloads and Usage: AI makes up roughly 15% to 25% of active data center usage and workloads, with the rest driven by traditional cloud computing, enterprise databases, email, and streaming services like Netflix or social media.
The constant Claude errors customers are screaming at AWS about argue that there is not enough.
I’d wager that much of the reported demand is to run Agents. I’ve seen estimates comparing token use for LLM queries versus running an agent. Estimates ran from 10 times higher to 100 times higher for the agents. I recall Goldman pegged it at 50x in a piece supporting the continued build-out of datacenters.
I was bewildered by this at first, but it’s because an agent never sleeps. It keeps refreshing it’s own database 24/7 to stay up to date. (I’ve wondered why you just cannot make an agent refresh itself on demand but I’ve never gotten a clear answer back except that it would slow the agents down when you had a task for it.)
Agents have not yet shown that they can be trusted with access to important information, such as the keys to your Binance account.
Even worse, at this time the authors and vendors of AI Agents enjoy a liability shield. If my company adopts an agent that divulges sensitive client data, only we can be sued. Given this, the promised returns from AI agents allowing for smaller headcounts had damn well better be very financallly compelling. A SME might take the risk, but larger enterprises are hesitant until both of those risks can be significantly be reduced.
I don’t know whixh AI Yahoo contributors use, but I found three very funny Freudian slips in an obvious AI post this AM.
Yeah, I should’ve clarified, but I’m fully bought into the idea that AI is a game changer and I personally get a ton of value out of it.
I mean it more in the sense that these companies could dial back the investment a bit and watch their stock pop just by virtue of dialing it down a bit rather than creating an existential risk (in the financial sense) for themselves.
I also expect the technology (both AI and semis) will improve dramatically in the coming years and we’ll get a flood of semi supply that’ll drop costs and make these companies question the decisions to commit to massive debt.
The flip side to that is they’ll probably be able to refinance at rock bottom rates in a few years when the Fed has to go back to ZIRP.
I’ve long tired reading about how AI is going to deliver an endless parade of medical breakthroughs. Try to do an LLM search for the amount of datacenter capacity is being used by biomedical researchers. (Don’t forget that if you can only find figures for the broader healthcare sector, those include insurers like UHC who are adopting AI to reject more claims.)
Ah, lookie here!
• Scientific & Specialized Computing Share: Non-GenAI scientific simulations, weather forecasting, energy modeling, and pharmaceutical/biomedical research make up a small fraction of total hyperscaler compute workloads—generally estimated by industry analysts to sit well below 1% to low single digits (1%–3%) of total global data center footprints.
We might be looking at a one big price earnings reset
One emerging problem is these companies have no moat. For many users, one AI is almost as good as another. (Heck, new companies can even use another company’s advanced AI to train their AI.) If that is the case (no moat I mean), the only way most will be able to retain customers is by lowering prices, which is not a great long-term strategy considering the capex these pioneers are currently paying. Apple and Coke have legions of loyal, lifetime users, for example. Will AI users develop the same affinity for Claude or Chat GPT if other good AI models are eventually available at a fraction of the cost? That makes AI more like a commodity then, or a service. Certainly various AI companies will develop specialties or niches for the medical field, lawyers, graphic artists, engineers, etc. Coding itself is perhaps the best example. But that is not nearly everyone. In my humble opinion, AI doesn’t yet have a “killer app” that the majority will find well worth the cost. Meta’s Muse is perhaps a fair attempt to address that, but the jury is still out on that one. Additionally, burning cash to buy the latest generation of chips — all of which have an eventual expiration date on them — means a good portion of your capex will remain constant.
You’ve given a very useful analogy as I see it. In very general terms, think o AI as a programming language because companies are competing now to build apps from it. With that in mind, how many app companies will be needed. Will there be enough computing demand to support all the capacity buildout. Has anyone tried to game that out, even in vague terms.
Metaverse 2.0?? I read this week that someone paid $450,000 to “buy” a house next to Snoop Dogg in the Metaverse….. It’s now worth $1,024.