Are Tech Stocks Rolling Over?

[joesxm3]

I just started watching Jerrimy Grantham on Diary of CEO.

He started mentioning AI as railroad and great ideas getting overinvested and then the bubble pops and the leaders rise from the wreckage.

So for a nice dose of gloom and doom. busy today so posting this without much review,

 

[audreyh1]

This is exactly the concern, stated succinctly.

Instead of bottom-up, organic profits from subscription-paying customers driving growth, there is a speculative money funnel at the top of this system driving the growth of everything under it. All is well and profits don’t matter as long as the flow of investment and hype continues from IPOs, capex debt fundraising, or circular deals leveraging the same hot ball of money, same as the railroads and dot coms.

That’s my biggest beef. Until regular corporations are able to successfully leverage AI technology to significantly increase their profits, this is essentially vaporware. I don’t care how sophisticated the tools and rapidly the technology is improving, it has to be used in a meaningful and affordable way. It’s going to require insight, cleverness and finesse, and a lot of cooperation not to mention training. At some point companies will have their chief AI officers, because in most cases it’s going to take a huge amount of work to restructure each company in a way that really leverages the new technology (by that time things will already be far along). This kind of thing doesn’t happen overnight. Companies overall are slow when it comes to restructuring, especially when exactly how is not clear/well understood, and there will be a lot of expensive hit and miss.

 

[Fermion]

Still, at some point it has to have net positive earnings.

Investors let Amazon go A LONG TIME without positive earnings.

 

[street]

That’s my biggest beef. Until regular corporations are able to successfully leverage the new technology to significantly increase their profits, this is essentially vaporware. I don’t care how sophisticated the tools and rapidly the technology is improving, it has to be used in a meaningful and affordable way. It’s going to require insight, cleverness and finesse and a lot of cooperation. At some point companies will have their chief AI officers, because in most cases it’s going to take a huge amount of work to restructure each company in a way that really leverages the new technology (by that time things will already be far along). This kind of thing doesn’t happen overnight. Companies overall are slow when it comes to restructuring, especially when exactly how is not clear/well understood, and there will be a lot of expensive hit and miss.

Yes, excellent view on new tech. Time will tell but for the long haul I see positivity.

 

[audreyh1]

Yes, excellent view on new tech. Time will tell but for the long haul I see positivity.

Well sure, long term.

 

[NW-Bound]

I have to push back on the bolded part. I follow these names pretty closely because they’re core to my options wheel strategy, and that statement just doesn’t line up with current reality as I see things.

Amazon is clearly profitable at the company level today, and AWS is a major profit engine, not some marginal side business. Microsoft’s cloud segment and the overall company are also solidly profitable. Oracle’s long‑standing database and applications businesses have been good cash generators for years; OCI is still earlier‑stage, but it’s being built on top of a profitable base, and recent earnings calls are about growth and margins, not about the company bleeding red ink. Meta is the only one where you can fairly say a big chunk is “in the red,” but even there it’s specifically Reality Labs/VR/AR that burns cash, while the core Family of Apps is highly profitable and more than covers those losses.

Surely, these huge companies were profitable and established businesses, unlike the dot-coms of 2000. However, their cap ex now far exceeds their holding cash and their cash flow. They are now issuing bonds or new shares to go to war with each other, building things that do not yet generate revenues commensurate with the investments. Investors are getting more skeptical with circular financing. There are also talks that they manage to structure some of the borrowing such that it stays off their balance sheet.

I don't know a lot about corporate financing, but remember that in the Great Recession, the investment bankers all over the world managed to show good profits off the subprime business, until it proved to be a fiasco. As Buffett famously said then, "you don't know who has been swimming naked until the tide is out".

 

[NW-Bound]

"Another analogy is the fiber optic boom."

I bought quite a bit of GLW Corning about a decade ago for $20/share, because it was a good dividend stock. Apple is funding Corning's manufacturing plant to make Gorilla glass in Kentucky, and Corning is the company that makes fiber optic cable. I've been selling some stock for about a year - it's currently $205.83.

Yes, GLW survived the 2000 tech bust, same as Cisco. Both stocks took more than 20 years to recover to their 2000's high, and that's before inflation is accounted for.

These are the pick and shovel makers. The gold diggers like Global Crossing, WorldCom, 360networks, Rhythms Netconnections and many more were wiped out. Some pick and shovel makers like Lucent, Nortel were also wiped out despite being reputable established companies.

 

[Zathras]

Surely, these huge companies were profitable and established businesses, unlike the dot-coms of 2000. However, their cap ex now far exceeds their holding cash and their cash flow. They are now issuing bonds or new shares to go to war with each other, building things that do not yet generate revenues commensurate with the investments. Investors are getting more skeptical with circular financing. There are also talks that they manage to structure some of the borrowing such that it stays off their balance sheet.

I don't know a lot about corporate financing, but remember that in the Great Recession, the investment bankers all over the world managed to show good profits off the subprime business, until it proved to be a fiasco. As Buffett famously said then, "you don't know who has been swimming naked until the tide is out".

I am sure any bust will wipe out some players.
But there are also big players that will survive and absorb the market the smaller players had when they go under.

As one example, Amazon plans on spending about $200 Billion on AI. They have $143 Billion cash and $20-30 Billion free cash flow (prior to spending on AI).

Yes, it would hurt, but they would come through stronger than most.

It also depends on how it unfolds.
If AI is useless, all but the largest companies will go under.
If AI proves of some use, it won’t be as bad.
If AI proves very useful, only the over leveraged companies with no other foundation will go belly up.

In any event, I don’t know which path we are on, so I am happy to sit this one out other than the chip makers (who win unless AI is totally worthless and everyone goes belly up).

 

[Route246]

Surely, these huge companies were profitable and established businesses, unlike the dot-coms of 2000. However, their cap ex now far exceeds their holding cash and their cash flow. They are now issuing bonds or new shares to go to war with each other, building things that do not yet generate revenues commensurate with the investments. Investors are getting more skeptical with circular financing. There are also talks that they manage to structure some of the borrowing such that it stays off their balance sheet.

I don't know a lot about corporate financing, but remember that in the Great Recession, the investment bankers all over the world managed to show good profits off the subprime business, until it proved to be a fiasco. As Buffett famously said then, "you don't know who has been swimming naked until the tide is out".

I agree regarding the AI/DC capex race but the implications you're drawing don't match what I'm seeing in terms of numbers.

Yes, capex is massive and they’ve clearly leaned more on the bond market to fund it. That’s worth watching. But these companies still throw off huge operating cash flow and sit on big cash piles. They are choosing to lever up to go faster, not because they’re running out of cash, and there’s not much evidence of Enron‑style off‑balance‑sheet tricks here as there is much more transparency given the Internet news cycle.

To me this isn’t “dot‑com 2.0” or “2008 2.0.” It’s a capital‑allocation bet: if they spend hundreds of billions on AI/ML and the returns disappoint, shareholders will feel it even if the businesses remain very profitable. That’s what I’m focused on, especially since I’m running an options wheel in these names, not the idea that they’re secretly on the brink.

I also tend to focus on earnings with a "where's the beef?" type mentality. A company with no earnings is really what I try to stay away from. If you leave out TSLA the companies in play in the Mag7 are extremely profitable and generate a lot of cash from their core businesses.

These companies are looking out for the shareholders because the management has a significant piece of that action. APPL and NVDA have aggressively done stock buyback with some of their cash pile.

 

[Route246]

Yes, GLW survived the 2000 tech bust, same as Cisco. Both stocks took more than 20 years to recover to their 2000's high, and that's before inflation is accounted for.

These are the pick and shovel makers. The gold diggers like Global Crossing, WorldCom, 360networks, Rhythms Netconnections and many more were wiped out. Some pick and shovel makers like Lucent, Nortel were also wiped out despite being reputable established companies.

You left out Exodus on the down side and left out Equinix on the upside. Exodus just ran out of cash and died. Equinix stopped expansion, got a bail out from SingTel and went from a penny stock and reverse split to trading over $1000/share today and never filed for bankruptcy.

Cisco is a good example of stellar management and used the IBM playbook, seriously downsizing and then focusing on core businesses and enterprise customers and have emerged very profitable today, albeit a little boring in terms of technology.

 

[oiseux]

That’s my biggest beef. Until regular corporations are able to successfully leverage AI technology to significantly increase their profits, this is essentially vaporware. I don’t care how sophisticated the tools and rapidly the technology is improving, it has to be used in a meaningful and affordable way. It’s going to require insight, cleverness and finesse, and a lot of cooperation not to mention training. At some point companies will have their chief AI officers, because in most cases it’s going to take a huge amount of work to restructure each company in a way that really leverages the new technology (by that time things will already be far along). This kind of thing doesn’t happen overnight. Companies overall are slow when it comes to restructuring, especially when exactly how is not clear/well understood, and there will be a lot of expensive hit and miss.

Companies at the front edge of adoption and who are doing it smartly have realized considerable savings, and not just from reducing labor or hiring. (I'm a relatively recent early retiree). Many have fallen short of their targets, with problems associated with internal limitations like data access or unfocused, haphazard deployments with no roadmap/ROI studies before deplyment. (Studies: Bain and Orgvue via Yahoo). The interesting thing is they are all getting more focused (btw, a boom for these consulting and service firms). On the other hand, effective implementations see a return of $3.70 for every $1 returned, and the best see a 10x return, per an IDC study. The returns show are beyond employee reductions, but in increased revenue, improved time to margin, reduced maintenance, operational efficiency, etc. In some examples, IBM save $3.5B in total over two years. Amazon has cut cost of delivery by 25%. Others have accelerated time to money or mature cost by six months or more, halved maintenance costs, reduced inventory, improved pricing, etc

Per an IBM CEO study, 76% of companies surveyed have already appointed a chief AI officer. Link

 

[Route246]

That’s my biggest beef. Until regular corporations are able to successfully leverage AI technology to significantly increase their profits, this is essentially vaporware. I don’t care how sophisticated the tools and rapidly the technology is improving, it has to be used in a meaningful and affordable way. It’s going to require insight, cleverness and finesse, and a lot of cooperation not to mention training. At some point companies will have their chief AI officers, because in most cases it’s going to take a huge amount of work to restructure each company in a way that really leverages the new technology (by that time things will already be far along). This kind of thing doesn’t happen overnight. Companies overall are slow when it comes to restructuring, especially when exactly how is not clear/well understood, and there will be a lot of expensive hit and miss.

Most corporations under‑select and under‑empower genuine thinkers, then wonder why tools and incremental skill upgrades don’t translate into meaningful outcomes. Exceptional corporations deliberately cultivate and empower a minority of people with deep vision and curiosity, build systems around their thinking, and foster engagement, so that AI tools and coding improvements function as genuine multipliers rather than cosmetic upgrades. The weak get weaker and the strong get stronger.

 

[Markola]

I’m reading this morning about how OpenAI is considering delaying their IPO due to the poor performance of the SpaceX one.

They have no organic profits to feature, they probably recognize that Sam Altman is about as charismatic as a robot, and that data centers are more challenging to hype than Martian colonies.

 

[Cassius King]

Couple of questions that I have been pondering:

As the nature is with chips/memory, it has historically been a race to the bottom as they become commoditized, after this initial hysteria, this too shall level out.

In regards to AI and datacenters. There are more than a dozen horses at the gate ready to race. There can only be one winner and a couple of place and show companies. This too will have to settle out.

With all the backlash about where datacenters are built and the energy they consume, water used. Doesn't that make an orbital datacenter more sensible and economical?

The lifecycle of chips/servers/memory is short. Newer, faster, less energy reliant options will be created tomorrow and need constant replacement and upgrading. Is there a company that exists today that can dispose of and recycle/reuse these components?

Apologize for the thought dump here. Thanks.

 

[Qs Laptop]

The lifecycle of chips/servers/memory is short. Newer, faster, less energy reliant options will be created tomorrow and need constant replacement and upgrading. Is there a company that exists today that can dispose of and recycle/reuse these components?

This is why the semiconductor stocks and data center infrastructure stocks have a multi-year growth profile. The race is on to produce more energy efficient chips to reduce the amount of power and cooling needed at the data centers. The 2030's will likely be the decade of refitting and downsizing data centers with more efficient components.

 

[MRG]

This is why the semiconductor stocks and data center infrastructure stocks have a multi-year growth profile. The race is on to produce more energy efficient chips to reduce the amount of power and cooling needed at the data centers. The 2030's will likely be the decade of refitting and downsizing data centers with more efficient components.

This whole thing is insane to me. Early in my career I was an assembly developer and sometimes had to count the instructions in a particular code path of infrastructure code. You had to write code to optimize certain logic paths.

I’ve never seen the investments in workloads that from the outside don't appear to be optimal. There seems to be no end of the appetite for both CPU and memory, resulting in pushing demand for electricity, cooling, water etc... My experience was more around transactional systems but this is an workload I expect will be aided by an unconventional technology breakthrough. I saw an article, about a company trying to reduce the AI memory footprint by 30x.

I hope it's very early to implement the environmentals for these new massive datacenters because hopefully our current plans are off by orders of magnitude. I remember being in large datacenters built in the early 1990s and sized for a little growth looking like empty caves in 2010.

 

[audreyh1]

I’m reading this morning about how OpenAI is considering delaying their IPO due to the poor performance of the SpaceX one.

They have no organic profits to feature, they probably recognize that Sam Altman is about as charismatic as a robot, and that data centers are more challenging to hype than Martian colonies.

Surprise, surprise, not!

However, waiting may find an even worse situation.

 

[oiseux]

Couple of questions that I have been pondering:

As the nature is with chips/memory, it has historically been a race to the bottom as they become commoditized, after this initial hysteria, this too shall level out.

This has been true, certainly for memory and storage, but the cycles have been becoming more shallow. The number of suppliers have been dwindling, with three large suppliers with a large technology and market lead, and then a few about 3 to 4 generations behind in technology (that are benefiting from the supply shortage (The leading China one has required tremendous financial support and has been hampered by the lack of access to leading edge manufacturing equipment). The cost of a new, from ground factory to build the chips (the "fab") now costs $15 to $20B, most of which is in new equipment that is also supply constrained. That is why it still takes 3 to 4 years to get a new factory in operation. The NAND side has a few more players (less concentrated) and is less constrained by equipment, which why it remains more vulnerable. There are also some new wildcards --- customers are now signing five year min/max purchase agreements (even with deposits) and the next generation of HBM (4.5) features a logic layer that can be customized to the architecture of each logic chip.

With all the backlash about where datacenters are built and the energy they consume, water used. Doesn't that make an orbital datacenter more sensible and economical?

The lifecycle of chips/servers/memory is short. Newer, faster, less energy reliant options will be created tomorrow and need constant replacement and upgrading. Is there a company that exists today that can dispose of and recycle/reuse these components?

I think we covered this up above, but the launch costs and replacement cycles are economically challenging to justify today vs terrestrial solutions, and there are numerous technological problems due to the rigors and risks of space that need to be solved. No one is saying it is impossible, but it does appear it will take a while.

For current terrestrial use, there are recycling systems in place, although the economics can be challenging to justify. It adds to the challenges of space, with the lifecycles potentially shortened and disposal another challnege.

My experience was more around transactional systems but this is an workload I expect will be aided by an unconventional technology breakthrough. I saw an article, about a company trying to reduce the AI memory footprint by 30x.

Yes, there are have been companies chasing methods for quite a while, from software models and tool chains to alternative system architecture, CPU and memory solutions. A good part of the issue today is that the default approaches to AI use models and systems that are very memory transaction centric, and the pursuit of higher weights while retaining precision has fueled the curve. The models and usages at their heart at times boil down to 95%+ multiply-accumulate operations, which are essentially memory transactions. So different models are segmenting for different uses (cost/precision/memory footprint) while other techniques (Turboquant, for example) are being explored. There are other more revolutionary approaches as well, but these have struggled to gain traction. These things typically all involve tradeoffs, though, and the industry will begin to sort things out, and new solutions will incrementally emerge.

 

[Andy22003]

Despite being hyped to infinity and beyond, SpaceX sits below the launch price of $155. It has no earnings. OpenAI and Anthropic want to go public next. They, too, have no earnings. The massively-successful, low-debt software monopolies are floating debt to keep up with the AI capex race, with no earnings in sight. AI companies are dumping or raising their unlimited monthly subscription price models. Corporate users are dialing back their token budgets. There isn’t enough electricity to power all the data centers, whose construction pace is lagging. Data centers in space are only on the drawing boards.

It wouldn’t matter, except that a small handful of these companies’ stocks drive US stock market returns. What am I missing? If you don’t like my entirely unoriginal outline, what is your bull case instead?

I am of the opinion that the serious rotation out of AI tech is fast approaching. Of course, I could be completely off but I am moving a good chunk of my 457b away from serious NASDAQ exposure, tech. I am betting on the rebuilding of the US. So, I have a set up in place that will be implemented Monday (6/29). This is my 457b at ~$400K, About half with be in a Schwab US Large Cap fund: FDNX. The rest will be divided between PAVE, AIRR, GRID (lots of infrastructure and energy) and a smaller bit in a momentum fund; FMTM. That has some tech, but also financial, health, industrial. I don't need 60%+ growth and the risk and roller coaster ride that goes along with it? While not without risk, the new allocation has a less volatile record but a decent history, so I will see where we stand at year's end.

 

[audreyh1]

It’s hard to know when and if it’s coming anytime soon. These cycles often take way longer than anyone expects.

 

[Route246]

Couple of questions that I have been pondering:

As the nature is with chips/memory, it has historically been a race to the bottom as they become commoditized, after this initial hysteria, this too shall level out.

In regards to AI and datacenters. There are more than a dozen horses at the gate ready to race. There can only be one winner and a couple of place and show companies. This too will have to settle out.

With all the backlash about where datacenters are built and the energy they consume, water used. Doesn't that make an orbital datacenter more sensible and economical?

The lifecycle of chips/servers/memory is short. Newer, faster, less energy reliant options will be created tomorrow and need constant replacement and upgrading. Is there a company that exists today that can dispose of and recycle/reuse these components?

Apologize for the thought dump here. Thanks.

Regarding data centers in space I am extremely dubious. Nobody has explained to me how you can put two-ton systems in space along with the staff to do the smart hands touches required in every data center known to man. Anyone, and I mean anyone, who has spent time in a modern data center knows what a crash cart is and knows that there are spare parts cabinets full of cables, testers, tools and supplies. I have no clue what these people are thinking when they talk about data centers in space. The other issue is cooling. I don't see how you can remove the massive amount of heat rendered with these systems.

 

[joesxm3]

How are they pushing the crash cart up to service the 10,000 Starlink satellites in orbit now?

Elon did a video a while ago explaining how the "AI1" satellites will work. He said that the AI1 is basically the same tech as Starlink, except actually easier to do. AI1 will have a larger solar panel facing the sun than Starlinkj and will have a radiative cooling panel knife edge 90 degrees to the solar panel. They are easier because they do not need all the complicarted radio antennas that a Starlinkj has, just about one rack's worth of NVDIA GPU's and some laser comm links to connect to the Starlink network.

There will be a large number of small nodes, so when one satellite fails it will just be deorbited and replaced with another. If they weigh 2000 pounds, that means one Starship with 100 ton payload capacity could easily launch them, probably limited by the "pez dispenser" that can hold up to 60 satellites.

 

[NW-Bound]

I dunno.

I remember an interview about the Hyperloop some years ago. The interviewer asked about the difficulty of maintaining a near vacuum in a long big tube, plus how passengers could quickly enter the pod inside this tube. Musk said "It's not that hard".

And in 2017, Musk promised that in 10 years, SpaceX will take people by rockets from NYC to Shanghai in 40 min.

And the Las Vegas Loop looks more like "Teslas driven by chauffeurs in a storm drain" than the grand network of tunnels for high-speed cars that were promised.

 

[MichaelB]

It’s hard to know when and if it’s coming anytime soon. These cycles often take way longer than anyone expects.

Yeah, they do take a long time. And when they happen there’s often so much noise it’s hard to see what’s happening in real time. It’s only looking back that we see all the market signals with clarity.

 

[Markola]

NW-Bound, I agree the Key Person Risk here is, eh, “unique.” 🫪

In my critique of these AI companies, I could be entirely wrong. Someday, somehow, this transformational AI tech will be monetized, or else it will cease to exist, and I don’t see that happening. How and who gets from here to profitable is the murky part that doesn’t pencil for me currently.

I’m also in no position to gloat, as I’m worrying about a different tech investment with its own Key Person Risk currently, which I won’t go into here.