Meta’s chip plan says “why not both?”

Not replacing Nvidia — just adding more seats at the table

Meta’s latest AI move is basically the corporate version of buying a bigger garage instead of selling your old car. Reuters says the company plans to start manufacturing its in-house Iris AI chip in September, with Broadcom helping develop it and TSMC handling production.

That matters because the immediate market read could’ve been: “Uh oh, Meta is going full DIY and ditching Nvidia and AMD.” Daniel Newman’s take was the opposite — Meta is augmenting its chip stack, not swapping out the incumbents. And honestly, that fits the scale of the spend.

The capex monster is still hungry

Meta is reportedly targeting about 7 gigawatts of computing infrastructure this year, then doubling that to 14 gigawatts next year. That’s not a side project. That’s a full-blown AI land grab.

And when a company says it may spend as much as $145 billion on AI infrastructure in a single year, the message to suppliers is pretty simple:

• more custom silicon work for Broadcom
• more foundry business for TSMC
• more GPUs still needed from Nvidia and AMD

So who wins here?

If you own Meta, the pitch is that custom chips help it control costs and scale faster without depending on one supplier for everything. If you own the semiconductor names, the bigger reality is probably less dramatic and more lucrative: Meta’s AI ambitions are so large that it may end up buying both custom chips and off-the-shelf accelerators.

In other words, this isn’t a sequel where the hero replaces the whole cast. It’s a bigger ensemble. Big picture: Meta’s not saying “goodbye” to Nvidia and AMD — it’s saying “we need a bigger checkbook.”