Meta's building its own AI chip backed by a $145B infrastructure budget this year. Capability win or capex concern

Meta plans to start production of its in-house AI chip, codenamed "Iris," in September (based on an internal memo - Reuters). It's part of Meta's MTIA (Meta Training and Inference Accelerator) chip line and the stated goal is straightforward, reduce dependence on Nvidia and AMD for the massive volumes of AI compute Meta needs. Meta's working with Broadcom on chip design and TSMC on manufacturing. Testing reportedly took just six weeks with no major issues found, which is notable because Meta's in-house chip effort has actually struggled for years. So this is a real reversal of fortune for the internal chip program.

Meta shares fell about 2.6% today and the reason cited is the scale of spending behind this announcement, not the chip itself. The memo confirms Meta plans to deploy 7 gigawatts of computing infrastructure this year and double that to 14 gigawatts by 2027, funded by up to $145 billion in AI infrastructure capex this year alone. That's an enormous number even by Meta's own recent standards and its happening as investors are getting sensitive to exactly this kind of spending scale across the whole sector.

One thing to note, this new chip is explicitly meant to supplement, not replace, Meta's GPU purchases from Nvidia and AMD. Meta's also simultaneously deepening ties with those same vendors elsewhere, there's a multiyear deal for up to six gigawatts of AMD Instinct GPUs, an expanded Broadcom custom chip partnership running through 2029 and deals for Google's TPUs and Amazon's Graviton5 chips too. They are diversifying the supplier mix while total demand keeps growing across all of them at once.

There's also a fact here worth noting. Meta's targeting a new MTIA chip approximately every six months through 2027, which is a notably faster than the roughly annual release cycles common across the rest of the industry. If Meta can actually execute on that pace given how long this program previously struggled, that's a real capability signal.

So this is really two separate stories layered on top of each other. One is a genuine technical and strategic milestone, a previously troubled internal chip program hitting a real production date with clean test results. The other is the market once again reacting to the sheer scale of capex commitments across the AI infrastructure buildout, which is the same tension we've seen play out with Amazon, Microsoft, and others recently good execution news getting overshadowed by spending scale concerns.

It feels like the market cared more about the $145 billion than the chip that might eventually reduce it. So does this MTIA progress actually matters more than today's price action suggests or the capex concern has increased by this announcement.