Disclosure or it didn't happen — and Google disclosed again. The grant US12067476B2 (“Mixture of experts neural networks,” issued 2024-08-20) extends the company's MoE estate with another grant naming Noam Shazeer and Azalia Mirhoseini. Assigned to GOOGLE LLC and classified CPC G06N 3/045, it is a deliberate deepening of foundational IP, not a one-off.
The pattern is the story. A single foundational grant can be designed around; a growing family of grants by the originating researchers is harder to route past. Google now holds dated MoE IP across multiple years, all aimed at the same business-critical mechanism: more model capacity without proportional inference cost.
“A system includes a neural network that includes a Mixture of Experts (MoE) subnetwork between a first neural network layer and a second neural network layer. The MoE subnetwork includes multiple expert neural networks.”— U.S. Patent No. 12,067,476 source
Why this matters to the income statement: mixture-of-experts is central to the argument that large models can be served affordably. As the technique becomes standard across the industry, the ownership of its foundational IP becomes a competitive and potentially licensing-relevant fact — one that Google's growing estate makes harder to ignore.
House rule: a grant proves invention and ownership, not revenue, and we attribute no figure. We make no infringement claim against any company using MoE — that is a separate, claim-level question we don't resolve here. The defensible point is about accumulation and date.
For the markets reader, the takeaway is to watch estates, not single patents. One foundational MoE grant is a fact; a multi-year family of them, authored by the technique's originators and assigned to Google, is a position. The 2024 grant is the latest dated entry in that position.