From the article:

AI data centers' thirst for energy is so severe it can wreck gear on and off site, including the dirty workhorses powering the artificial intelligence boom — thermal turbine generators.

[...]

The AI power loads "have the potential to change their demand almost instantly," FERC Chair Laura Swett said. "This rapid fluctuation causes voltage stability issues that threaten grid reliability."

[...]

Their scramble's cracking eggs. It's one reason why Tesla helped foot the bill for mechanical engineers at Pennsylvania State University to investigate particular elements of gear wreckage. They specifically dug into how the rapid fluctuations in power consumption can, in extreme cases, violently twist and break turbine generator shafts. It so happens that large scale battery storage systems sold by Tesla, and other firms like TerraFlow, provide one way to cushion unwieldy loads.

[...]

Thermal turbine systems have multiple rotating shafts coupled together. None of these shafts are perfectly rigid.

"When everything is in normal condition, every turbine section is rotating exactly at the same speed. But the shaft section will have a constant twist angle" as it transmits power, Chaudhuri said. "When there is a disturbance, the rotor masses will start rotating at different speeds. And the shaft will get twisted in clockwise and anticlockwise directions opposing each other."

That twisting, Chaudhuri explained, generates excessive strain.

[...]

AI data centers have chill periods with relatively steady power consumption at around 10%-30% of max levels, Chaudhuri says. But during model training sessions, the power consumption fluctuates.

This is where AI loads are particularly unique, he added. "We have never seen anything like this in our history."

According to Chaudhuri, the fluctuations can disturb turbine rotors so they "start oscillating against each other." This can trigger a "torsional interaction" — a sort of twisting force that can feed on itself. The effect risks "catastrophic damage of the turbine shaft because you're twisting violently with significant oscillation amplitude," he said.

Chaudhuri said he has heard of some generator shafts, including one of at least 50 megawatts, snapping due to these fluctuations.

But even without shaft breakage, "significant fluctuations and twists can lead to degradation of fatigue life," Chaudhuri said. He called shortened lifespans for gear "another major concern among the OEMs."

The upshot is that the lifespan of equipment that is backlogged for years in advance is, at least in some cases, being significantly shortened by swings in AI data center power demand.

[...]

It's not just Big Tech's on-site gear that's at risk. "If you have a (grid-based) generator which is electrically close to the data center, then that also runs the risk of shaft breakage," said Chaudhuri.

[...]

Study-backer Tesla operates its own data centers, including to train its driver-assist tech. Tesla also provides solutions to help smooth loads for AI training. The company sells its biggest battery systems, Megapacks, for this purpose. TerraFlow Energy likewise sells energy storage gear with flow battery chemistries designed to "respond instantly to load changes."

While these types of solutions are already out in the field, data on the scope of this particular turbine threat is so far scarce. This is thanks to the secretive nature of data center operators and the newness of the phenomenon. But researcher and study co-author Chaudhuri argues the risks are there and so is the degradation potential.