Google pays SpaceX $920 million a month to rent about 110,000 Nvidia GPUs, from this October through June 2029. $GOOGL

Pichai (Google’s CEO) said on the Big Technology podcast on Tuesday that Gemini 4 Argon used nearly 25 trillion tokens in its first 72 hours after launch—three times as many as Gemini 3.0 over the same period. Google didn’t have enough compute to serve it.

He also said Google Cloud is constantly giving up short-term revenue because compute has to go to frontier AI research and development first. Google DeepMind gets priority; whatever is left goes to the other businesses. Everyone is short on compute now—not just Google. Every company wants twice as much as it originally planned for. The compute capacity available today is the result of a decision he made in 2024.

Musk said on SpaceX’s earnings call yesterday that every hyperscaler and dozens of smaller companies are persistently asking to rent his compute, and that he could rent it out several times over. Someone said this was good news for neoclouds. He replied this morning: that assumes you can bring the compute online and keep it running reliably. At scale, that’s extremely difficult.

At $920 million a month, that’s just over $11 billion a year. Google is building its own TPUs and still has to pay this much. The contract says that after the end of this year, either party can terminate it with 90 days’ notice. In 2015, Google and Fidelity invested $1 billion in SpaceX. Now Google pays it $920 million every month.
#Spacex