One thing immediately stood out.
This doesn’t look like Meta trying to beat Nvidia.
It looks like Meta trying to escape dependence on Nvidia.
Meta plans to spend up to $145 billion on AI infrastructure this year.
Its first in house Iris AI chip enters production in September.
The company wants to double computing capacity from 7 gigawatts to 14 gigawatts by next year.
New chip generations are planned every six months through 2027.
That is an almost absurd buildout.
Here’s the part that made me laugh.
Even after all this spending, Meta is still buying Nvidia GPUs.
It is still working with AMD.
It is still relying on Broadcom.
It still needs TSMC to manufacture the chips.
So this isn’t replacing the existing AI supply chain.
It’s adding another layer because buying enough compute has become its own problem.
Some Reddit users called it another Zuckerberg vanity project.
Others argued every AI giant eventually has to build custom silicon because Nvidia’s margins are too high.
The more convincing argument was different.
The real shortage isn’t chip designs.
It’s manufacturing capacity.
It’s advanced packaging.
It’s high bandwidth memory.
Everyone is fighting over the same bottlenecks.
Wall Street seemed more focused on the bill than the chip.
Investors saw another $145 billion commitment and wondered where this spending finally ends.
That’s probably the bigger story.
The AI race isn’t slowing down.