Disclosure or it didn't happen, and this disclosure has provenance. Google's grant US11790214B2 (“Mixture of experts neural networks,” issued 2023-10-17) names Noam Shazeer and Azalia Mirhoseini as inventors — the researchers behind the modern mixture-of-experts technique. Assigned to GOOGLE LLC and classified CPC G06N 3/045, this is foundational IP, not a peripheral filing.
The mechanism is why MoE matters to the business. A dense model uses all its parameters on every input, so making it bigger makes every inference more expensive. A mixture-of-experts model routes each input to only a few “expert” sub-networks, so it can hold vastly more parameters while activating a fraction per query — capacity goes up, per-inference cost does not rise proportionally.
“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. 11,790,214 source
That decoupling is the financial heart of the technique. It is how labs claim larger, more capable models without a linear blow-up in serving cost, and it directly shapes the inference economics every AI vendor argues about. Google holding a foundational grant on it, authored by the technique's originators, is a meaningful ownership fact.
Google's filings discuss AI investment in aggregate and do not credit MoE specifically; we don't invent a number, and a grant proves invention and ownership, not revenue. We also make no infringement claim about any competitor's MoE use — enforceability is a separate, claim-by-claim question.
For the markets reader, the durable point is that the technique behind “bigger models that don't cost proportionally more to run” has dated, named, foundational IP at Google. When the industry debates inference economics, this grant is one of the primary documents the debate is standing on.