Meta Just Built The Chip NVIDIA Feared the Most

This month a new AI chip was scheduled to go into production at TSMC. It wasn't designed by NVIDIA, AMD or Intel. It's Meta's own chip, from the same company that signed a deal in February to deploy millions of NVIDIA GPUs. Its code name is Iris, and according to an internal memo seen by Reuters, it cleared six weeks of testing without any major issues. So why would one of NVIDIA's biggest customers build its own alternative to the very thing it keeps buying? And why now? Meta's engineers already gave away part of the answer in a research paper, with one number: 44%. That's how much their previous custom chip cut the total cost of ownership compared to GPUs, on average, across the models they moved into production. And here is the part that most people miss. Meta is not trying to beat NVIDIA at training the biggest models. It's going after the part of AI that runs every single day for 3.6 billion people. If this works, NVIDIA doesn't lose a benchmark. It loses its pricing power with one of its largest customers. Thank you for watching: ⁠Meta Just Built The Chip NVIDIA Feared the Most #Meta #NVIDIA #AIChips __ Exploring AI isn’t about having all the answers, it’s about asking better questions. As E.Y. put it, “the greatest danger is thinking you’ve already understood it.” This channel is where curiosity meets clarity, with updates, insights, and deep dives that keep you ahead of the noise. The future isn’t waiting, why should you?