Now Playing
ASICs Seize 25% of AI Inference & Broadcom's $22B Record Punished
Audio is available on Spreaker — see link below.
Custom silicon is reshaping AI infrastructure as hyperscaler ASICs race toward 25% of the inference market by 2026, while Broadcom posts record $22B revenue only to see its stock drop 17%. Today's briefing covers the GPU vs ASIC economics, agentic AI compute shifts, Intel's foundry reality check, and Nvidia's TAM defence.
Audio is available on Spreaker — see link below.
Custom silicon is quietly dismantling the GPU's grip on AI infrastructure, and the numbers are sharp enough to matter. Hyperscaler-built ASICs, which held less than five percent of AI inference in twenty twenty-three, are on track to capture twenty-five percent of that market by twenty twenty-six.
The math behind this shift is cleaner than it looks. Meta's MTIA v2 is now in production.
What's accelerating the timeline is the rise of agentic AI. These aren't simple prompt-and-response systems.
Broadcom just posted record quarterly revenue of twenty-two point one nine billion dollars. AI semiconductor revenue was up one hundred and forty-three percent year over year to ten point eight billion.
Intel's stock hit above one hundred and forty dollars following the announcement of an Apple manufacturing partnership backed by policy momentum. That's a forty-dollar move on a deal with no confirmed wafer delivery timeline, no proven yield performance, and no visible financial contribution.
The one number that keeps NVIDIA's position from collapsing under this narrative is the total market size. AI inference silicon is projected at one hundred and fifty billion dollars by twenty twenty-six, up from forty billion in twenty twenty-three.
Chapter summary auto-generated from the verified script. Listen to the full episode for the complete content.