From the article: … … … Back when I worked at Google, they would sometimes have a “space crunch” where new data centers weren’t going to come online for a while, resulting in engineers looking for...
From the article:
Rather than signaling doubt about AI’s future, recent data center adjustments by Amazon and Microsoft reflect an industry confronting harsh realities: power grids that take years to expand, land speculators inflating prices sixfold, and utilities overwhelmed with requests for more electricity than actually exists. The question isn’t whether AI infrastructure demand is real — it’s whether the real estate market and power grid can handle what’s coming.
…
The real constraint isn’t wavering demand but basic infrastructure. Power grids across the country are struggling (or failing) to keep up with AI’s explosive energy requirements.
…
The tech industry is adjusting its buildout plans in response. “We are finding the best right sizing,” said Henrique Cecci, a Gartner (IT) analyst who tracks the data center market. Beyond infrastructure constraints, the data center sector is moderating its energy demand forecast, going from five- to six-fold growth expectations to a more realistic three- to four-times increase. “Nobody knows exactly how big AI is, but everybody agrees it’s a big thing,” Cecci said.
The supply side tells a grimmer story. Utilities ordering necessary grid technology like combustion turbines today won’t receive them until 2029, according to the Electric Power Research Institute (EPRI), an independent energy research institute. Traditional grid buildout takes four to seven years under normal circumstances, but supply chain bottlenecks have made the situation worse.
…
The pursuit of those extra dollars has fueled what JLL’s Cvengros calls the real bubble in data centers, and it has nothing to do with AI demand.
“You have everybody and their mother trying to get in this data center game,” he said. “Whether it’s a farmer, somebody who owns land, private equity, or it’s a group who puts land under contract, they slap a power study on it, and then try to sell it for six times what it’s worth.”
This frenzy has created a secondary market where developers acquire sites, add basic power studies, and flip them at massive markups. Each power study generates a formal request to utilities for grid capacity, creating an avalanche of applications. The situation has become so distorted that utilities now see 30 to 100 gigawatts of power requests in individual markets — far exceeding what’s actually available.
The power crunch is forcing utilities to require massive upfront commitments. In markets seeing 30 to 100 gigawatts of power requests, companies must now put down deposits of $10 million to $30 million just to get in line for grid connections, Cvengros said. Those massive utility deposits aren’t just about power planning — they’re designed to reduce the speculation.
“That is a bubble that will fall apart in the next 12 months, 24 months,” Cvengros said.
Meanwhile, the hyperscalers with real business models continue their strategic building, working around the speculators when possible. They’re even buying entire neighborhoods, paying residents well above market rates to relocate. Cvengros recently helped a group acquire about 30 homes in one transaction, demolishing them to make room for data center expansion.
Back when I worked at Google, they would sometimes have a “space crunch” where new data centers weren’t going to come online for a while, resulting in engineers looking for ways to optimize the software and do more with fewer machines. So, I’m not unhappy to see costs go up for a while. A focus on efficiency due to supply bottlenecks doesn’t seem like a bad thing.
That final paragraph in your block quote talking about demolishing 30 homes is a stunningly selfish choice in a tight housing market. Homelessness is already a big problem without this. Edit, I...
That final paragraph in your block quote talking about demolishing 30 homes is a stunningly selfish choice in a tight housing market. Homelessness is already a big problem without this.
Edit, I don't mean surprising necessarily but it has the same destructive vibes as when the auto industry and big oil killed most streetcar systems.
I see huge potential for small modular reactor + real estate investors. Somebody's going to make a fortune.
This frenzy has created a secondary market where developers acquire sites, add basic power studies, and flip them at massive markups. Each power study generates a formal request to utilities for grid capacity, creating an avalanche of applications. The situation has become so distorted that utilities now see 30 to 100 gigawatts of power requests in individual markets — far exceeding what’s actually available.
I see huge potential for small modular reactor + real estate investors. Somebody's going to make a fortune.
Buying land and doing a basic power study to get in the queue can be done fast enough to pay off quickly. Actually building a nuclear reactor will take years and who knows what the market will be...
Buying land and doing a basic power study to get in the queue can be done fast enough to pay off quickly. Actually building a nuclear reactor will take years and who knows what the market will be like? It’s very risky to do without having a committed buyer for the electricity.
That de-risks the question of who will buy the power, but there are other risks: can they actually build it? How long will it take? Will the profits get eaten up by cost overruns?
I wonder if that's why our brand new Prime Minister just appointed a Minister of AI? Because Canada already has several SMR projects on the go: "Notably, Ontario Power Generation (OPG) is leading...
I wonder if that's why our brand new Prime Minister just appointed a Minister of AI? Because Canada already has several SMR projects on the go:
"Notably, Ontario Power Generation (OPG) is leading the way with their Darlington New Nuclear Project (DNNP), which will include four SMRs. Other SMR projects in Canada include Global First Power's project at Chalk River Laboratories, NB Power's ARC-100 project at Point Lepreau, and SaskPower's proposed SMR project, all using different SMR technologies."
Kevin O'Leary, Canada's loudest mouthpiece on The Dragon's Den (Canada's Shark Tank) has proposed a massive AI facility in Alberta, the largest in the world. I would assume he believes AB can provide enough power for it, although here would come from natural gas and geothermal not a reactor. But then again, Mr O'Leary is well known for shooting off his mouth and making flim flam deals so who knows if he just pulled the whole thing out of his... uh, hat.
I was catching up on Computex coverage yesterday and there were a few vendors showing off single racks that are pushing towards a megawatt, which was absolutely mind blowing to me - I’m used to...
I was catching up on Computex coverage yesterday and there were a few vendors showing off single racks that are pushing towards a megawatt, which was absolutely mind blowing to me - I’m used to even a modern “big stack of AI machines” rack being in the tens of kW, whereas an aisle of MW-scale racks would literally be getting towards industrial blast furnace power system and thermal management requirements.
Seeing stuff like that being floated, it makes sense to me that we’re starting to see civil engineering as the bottleneck more than the traditional tech side of the equation. And yeah, if that squeeze forces companies to prioritise efficiency that can only be a good thing.
That's a really good point actually, I didn't think about car charging but I'd bet a decent amount of the development in HVDC hardware and infrastructure is doing double duty across that and the...
That's a really good point actually, I didn't think about car charging but I'd bet a decent amount of the development in HVDC hardware and infrastructure is doing double duty across that and the datacenter market.
Edit: just realized I completely misread. The racks demand 1-MW, not that they were producing 1-MW, oops. But the comment is still kind of relevant as yes, 50-kw takes up about an average sized...
Edit: just realized I completely misread. The racks demand 1-MW, not that they were producing 1-MW, oops. But the comment is still kind of relevant as yes, 50-kw takes up about an average sized tiny house. Scale that up by 20 for just 1-MW. Multiply by aisles, like you said, and my Satisfactory nuclear power setup might not even be enough.
Original comment:
Wow that's nuts. I used to work in broadcast engineering as a junior technician 15 years ago and we had a 50kw digital transmitter (Thales, I believe?) that took up four full sized deep racks. Twice because redundancy for 8 massive racks. Obviously half was for transmitter equipment, the other half for power. And since it was too expensive to remove, they also had the old 20kw analog transmitter in a nearby room that was almost as big, analog transmissions having ceased just two years earlier.
I know it's not exactly the same, but thinking about how insanely dangerous those 50kw cabinets were without the proper precautions, near 1-MW cabinets is terrifying lol.
Loved that job, but the station owners were complete and utter assholes who wouldn't pay a living wage, among many many other ongoing issues. At least I did gain lots of valuable knowledge and advice from my peers that have gone a long way in my professional life.
Well I appreciated the interesting little anecdote either way! There's something that feels oddly... real? about actual broadcast gear blasting a signal right to the user that you just don't get...
Well I appreciated the interesting little anecdote either way! There's something that feels oddly... real? about actual broadcast gear blasting a signal right to the user that you just don't get with a bundle of fiber going to the nearest switch.
From the article:
…
…
…
Back when I worked at Google, they would sometimes have a “space crunch” where new data centers weren’t going to come online for a while, resulting in engineers looking for ways to optimize the software and do more with fewer machines. So, I’m not unhappy to see costs go up for a while. A focus on efficiency due to supply bottlenecks doesn’t seem like a bad thing.
That final paragraph in your block quote talking about demolishing 30 homes is a stunningly selfish choice in a tight housing market. Homelessness is already a big problem without this.
Edit, I don't mean surprising necessarily but it has the same destructive vibes as when the auto industry and big oil killed most streetcar systems.
I see huge potential for small modular reactor + real estate investors. Somebody's going to make a fortune.
Buying land and doing a basic power study to get in the queue can be done fast enough to pay off quickly. Actually building a nuclear reactor will take years and who knows what the market will be like? It’s very risky to do without having a committed buyer for the electricity.
So, we’re seeing deals like this one:
Google inks deal to develop 1.8 GW of advanced nuclear power
That de-risks the question of who will buy the power, but there are other risks: can they actually build it? How long will it take? Will the profits get eaten up by cost overruns?
I wonder if that's why our brand new Prime Minister just appointed a Minister of AI? Because Canada already has several SMR projects on the go:
"Notably, Ontario Power Generation (OPG) is leading the way with their Darlington New Nuclear Project (DNNP), which will include four SMRs. Other SMR projects in Canada include Global First Power's project at Chalk River Laboratories, NB Power's ARC-100 project at Point Lepreau, and SaskPower's proposed SMR project, all using different SMR technologies."
Kevin O'Leary, Canada's loudest mouthpiece on The Dragon's Den (Canada's Shark Tank) has proposed a massive AI facility in Alberta, the largest in the world. I would assume he believes AB can provide enough power for it, although here would come from natural gas and geothermal not a reactor. But then again, Mr O'Leary is well known for shooting off his mouth and making flim flam deals so who knows if he just pulled the whole thing out of his... uh, hat.
Mirror: https://archive.is/T9TJj
I was catching up on Computex coverage yesterday and there were a few vendors showing off single racks that are pushing towards a megawatt, which was absolutely mind blowing to me - I’m used to even a modern “big stack of AI machines” rack being in the tens of kW, whereas an aisle of MW-scale racks would literally be getting towards industrial blast furnace power system and thermal management requirements.
Seeing stuff like that being floated, it makes sense to me that we’re starting to see civil engineering as the bottleneck more than the traditional tech side of the equation. And yeah, if that squeeze forces companies to prioritise efficiency that can only be a good thing.
We already have electric sports cars that can charge at 367kw. Interesting to hear there’s a rack that’s a few multiples of that.
That's a really good point actually, I didn't think about car charging but I'd bet a decent amount of the development in HVDC hardware and infrastructure is doing double duty across that and the datacenter market.
Edit: just realized I completely misread. The racks demand 1-MW, not that they were producing 1-MW, oops. But the comment is still kind of relevant as yes, 50-kw takes up about an average sized tiny house. Scale that up by 20 for just 1-MW. Multiply by aisles, like you said, and my Satisfactory nuclear power setup might not even be enough.
Original comment:
Wow that's nuts. I used to work in broadcast engineering as a junior technician 15 years ago and we had a 50kw digital transmitter (Thales, I believe?) that took up four full sized deep racks. Twice because redundancy for 8 massive racks. Obviously half was for transmitter equipment, the other half for power. And since it was too expensive to remove, they also had the old 20kw analog transmitter in a nearby room that was almost as big, analog transmissions having ceased just two years earlier.
I know it's not exactly the same, but thinking about how insanely dangerous those 50kw cabinets were without the proper precautions, near 1-MW cabinets is terrifying lol.
Loved that job, but the station owners were complete and utter assholes who wouldn't pay a living wage, among many many other ongoing issues. At least I did gain lots of valuable knowledge and advice from my peers that have gone a long way in my professional life.
Well I appreciated the interesting little anecdote either way! There's something that feels oddly... real? about actual broadcast gear blasting a signal right to the user that you just don't get with a bundle of fiber going to the nearest switch.