I think its a bad analogy to compare AI overbuild to fiber optic

Cable has a finite distance to travel, and its still used today, even if it got bought for pennies on the dollar. Global Compute will continue to grow 10x, 100x even if it gets more cost effective. What happens faster, efficiency or growth? I can see growth moving exponentially near term as the race continues, efficiency takes time, especially to the point we literally need less data centers.

Mark Cuban thinks many data centers will turn into pickleball courts

He was on the @theallinpod making the case that AI infrastructure is being overbuilt

His comparison is the 90s when everyone raced to lay fiber-optic cable

Then technology got more efficient, and what… pic.twitter.com/w6rUCW2sYc

— Michael Burry Stock Tracker ♟ (@burrytracker) July 22, 2026

This AI boom kind of reminds me of the whole propane movement back in the late 70's to early 80's. With the OPEC embargo and fuel shortage still fresh in everyone's mind, we found ourselves looking for an alternate energy solution. Propane came up because the US is basically the Saudi Arabia of natural gas. Lot's of devices came to market to retrofit fleet and retail vehicles for propane. Unfortunately, there was little build out of the necessary infrastructure to support propane. The idea of propane-powered vehicles sort of died on the vine. Although, I think one can still find an old Impco(sp) propane carb online.

Like the idea of propane, AI requires immense infrastructure in the form of data centers. Even if the AI market has to take a breather, the technology is important enough that government will likely step in financially, on "national security" grounds, to keep it moving along. Maybe why the department of defense is asking for a few extra pennies in their budget...they see what's coming.

we shall see, but he could be correct that companies are overbuilding data centers.

The fiber optic comparison (and railroads before that) is more about thousands of redundant/unnecessary lines that were laid during the dot com boom.

They estimate about 80% of it was never activated. That's a crazy number.

To me it feels like a big disconnect between the data centers people talk about building and the ones that are actually going to BE built. I wonder if a lot of these projects are just spec projects where the builder would need to find a customer before actually breaking ground, and that may or may not ever happen

Another thought or similarities was is LNG plants.

Working at a design firm, in 2014 to 2017, there was like 150+ MT/a announced US capacity to be built. At the time, the market was like maybe 300 Mt/a.

A lot of projects were in the planning stage and died on the vine.

~egon

Burning question in my mind...

Who are the survivors, the big winners, last man standing in the AI space when private equity eventually throws the brakes on this wild show?

TTUArmy said:

---

Burning question in my mind...

Who are the survivors, the big winners, last man standing in the AI space when private equity eventually throws the brakes on this wild show?

---

I keep thinking they'll end up with different AIs that are more specialized.

Google doesnt need a high level AI to answer general search engine questions

Musk and Grok seem more focused on AI that can do science and engineering

GeorgiAg said:

---

TTUArmy said:

---

This AI boom kind of reminds me of the whole propane movement back in the late 70's to early 80's. With the OPEC embargo and fuel shortage still fresh in everyone's mind, we found ourselves looking for an alternate energy solution. Propane came up because the US is basically the Saudi Arabia of natural gas. Lot's of devices came to market to retrofit fleet and retail vehicles for propane. Unfortunately, there was little build out of the necessary infrastructure to support propane. The idea of propane-powered vehicles sort of died on the vine. Although, I think one can still find an old Impco(sp) propane carb online.

Like the idea of propane, AI requires immense infrastructure in the form of data centers. Even if the AI market has to take a breather, the technology is important enough that government will likely step in financially, on "national security" grounds, to keep it moving along. Maybe why the department of defense is asking for a few extra pennies in their budget...they see what's coming.

---

This you?

---

They are trying to offload private equity to 401ks and pensions so if it collapses the government will have no choice but to bail because that's cheaper than dealing with a retirement crisis when pension funds go bust. It will be the patriotic thing to do because we don't want retired teacher, firefighters and police officers to become homeless.

Interesting take, that China is trying to flood market with cheap (not better) AI to kill the US AI industry

Scott Galloway just explained why China doesn’t need to build better AI than America. It only needs to make American AI worthless.

Galloway: “I think China is beginning to engage in what I’ll call AI dumping.”

Not competing. Dumping.

It’s the term economists use for flooding a… pic.twitter.com/4pVK37LmOP

— Dustin (@r0ck3t23) July 23, 2026

jamey said:

---

Interesting take, that China is trying to flood market with cheap (not better) AI to kill the US AI industry

Scott Galloway just explained why China doesn’t need to build better AI than America. It only needs to make American AI worthless.

Galloway: “I think China is beginning to engage in what I’ll call AI dumping.”

Not competing. Dumping.

It’s the term economists use for flooding a… pic.twitter.com/4pVK37LmOP

— Dustin (@r0ck3t23) July 23, 2026

---

It's pretty obvious this is their current strategy.

Data centers aren't being over built because they are physically constrained in the rate they can build and power them. The backlog for data center construction is massive. They can't get the electricity they need fast enough, so these data centers are being built and onlined A LOT slower than the tech companies want.

The demand is for Gigawatts of more power. Venture capital is now funding any founder promising firm power - the number of turbine, nuclear, and geothermal startups has exploded in the last 2 years and continues to grow. Now any startup that can field a working demonstration will easily be able to raise at least a hundred million to scale production.

Elon Musk bought a turbine company.
Boom's first product and revenue will be a power turbine adapted from their engine development, not a super sonic aircraft.

It'll be interesting to see how it plays out for sure. There will be losers but it's hard to say who.

Anyone saying AI isn't moving the needle on profitability or margin is joking themselves or hasn't really looked at their organization and asked people to adopt it.

bmks270 said:

---

jamey said:

---

Interesting take, that China is trying to flood market with cheap (not better) AI to kill the US AI industry

Scott Galloway just explained why China doesn’t need to build better AI than America. It only needs to make American AI worthless.

Galloway: “I think China is beginning to engage in what I’ll call AI dumping.”

Not competing. Dumping.

It’s the term economists use for flooding a… pic.twitter.com/4pVK37LmOP

— Dustin (@r0ck3t23) July 23, 2026

---

It's pretty obvious this is their current strategy.

---

Their models require more tokens for similar output and their chips are way less efficient at producing tokens than the export controlled Nvidia chips (~2x worse than Blackwell, ~20x worse than Vera Rubin). China doesn't have the capacity to dump on U.S. AI model makers currently, because they don't have the capacity. They can skim lower value queries from AI labs, but the demand is currently saturating capacity. That leaves the higher value demand for the U.S. labs and China skimming the lower value queries.

They may be able to do it in the future, but it's not a near term reality from what I can tell. The competitive advantage is at the hardware layer, not the model layer is the takeaway, because the techniques to cut token usage/improve outputs are readily replicable.

In the future AI model makers instead of utilizing consistent models across queries will start routing queries more aggressively based upon actual need. Instead of consumers selecting models to query you'll ask a broad "Claude" for your output and it will route it to a quantized version of models to get a similar quality output from a lesser demanding model.

AI is going to get way more efficient on a token production basis and way more efficient on a token consumption basis over time.

Demand is still going to saturate every bit of computation, which is why Mark Cuban is almost certainly wrong about data center overbuild. It's like saying computer demand in the 90's was soon to be saturated, but even more wrong.

TTUArmy said:

---

Burning question in my mind...

Who are the survivors, the big winners, last man standing in the AI space when private equity eventually throws the brakes on this wild show?

---

I think it will be Anthtopic and Google.

The more recent movements at scale have been CNG (compressed natural gas- predominantly methane) rather than propane although there are lower range fleets that run off propane as well. Infrastructure (CNG filling stations) never really happened at scale tho so I'd have to say it was a failure. Most of the longer range CNG vehicles can also run on gasoline.

Propane is generally a lot more expensive than CNG.

The idea of a liquid CNG fuel tank at 3000 psi+ probably had something to do with the lack of widespread adoption.

dmart90 said:

---

TTUArmy said:

---

Burning question in my mind...

Who are the survivors, the big winners, last man standing in the AI space when private equity eventually throws the brakes on this wild show?

---

I think it will be Anthtopic and Google.

---

Unless Google has an absolutely huge mis-step, I don't see a future in which they aren't one of the last standing.

They have the organic traffic. They have leverage of being the rest of the internet's main source of traffic ("Google Zero"). They have money.

As with everything in the business world this is about timing.

• Some will be too early to the party and will have DCs that are filled with underpowered inefficient compute with too much fixed cost, too much sunk cost to effectively upgrade or optomize
• Some will be just good enough to find a niche market
• Some will hit the walk-off grans slam
• Others will still be holding a shovel with the broader market moves on.

My question becomes why is Cuban saying this? I've never viewed him as a good Samaritan that's just out using is personal time to to spread a cautionary tale for the sake of business charity. I may not be the sharpest tool in the shed, but I have to feel that if he's saying something he stands to benefit in one way or another.

I always take Cuban's input with a giant grain of salt. Partly because of the reasons you mentioned, but also because his primary success happened 30 years ago. He's not really made any kind of mark (pardon the pun) in tech since. He also got fleeced on his sale of the Mavs. He's not an idiot, but I also think he puts a pretty big front up and is not the expert he'd love folks to think he is

Agreed. I remember him on interviews saying streaming services wouldn't be a thing because of limits on broadband.

Will new technology supplant and mothball infrastructure being built today? History says certainly.

But we didn't go from horse and buggies straight to self-driving cars. There were lots of smaller steps and missteps along the way.

Say bad things are going to happen and when they happen you can point to it and say "See how smart I am?" If they don't happen you don't remind people and there's so much information out there that no one remembers you said it at all.