GOOGL
NVDA
SPCX
New boss energy, but make it urgent
SpaceX’s AI ambitions are moving fast enough to make your browser tabs sweat — and now the company is cleaning up its data-center team after engineering and reliability problems surfaced in Tennessee and Mississippi.
Jake Palmer, who ran physical infrastructure for SpaceXAI, reportedly left in late July, along with other data-center executives. In their place, SpaceX veterans from the rocket and Starlink worlds are taking bigger roles. Translation: when the specialists leave, the firefighters show up.
The not-so-fun part: uptime
According to the report, some facilities went months without proper backup cooling and power systems. That’s a little awkward for a business that’s supposed to keep giant AI workloads humming, not playing power-grid roulette.
A separate May report said one Macrohard facility was leaning on 100+ mobile chillers and was running below an internal 99.9% uptime target. Temporary power and cooling setups reportedly contributed to outages that interrupted model training. In other words: the AI stack can be as smart as it wants, but if the AC quits, the whole thing gets the digital equivalent of a tantrum.
Why Google and Nvidia are in the mix
This isn’t just a SpaceX housekeeping story. Google’s $920 million-a-month compute deal depends on SpaceX delivering roughly 110,000 Nvidia GPUs by September 30th. After a one-month grace period, Google can walk away or scale payments down if SpaceX misses the mark.
That makes the reliability hiccups more than an operations story. They’re now a direct stress test on Musk’s “we can bring compute online better than anyone else” pitch.
Big picture
No one’s saying the Google deal is dead. But the clock is ticking, the cooling bills are apparently enormous, and SpaceX is learning that AI infrastructure is less rocket launch, more very expensive plumbing. That’s not exactly the kind of metaphor investors love to hear.