AI
Agentic AI is the monetization story of the moment. Two 2026 Salesforce grants on AI agents in a database system show what the revenue is built on.
AI
The attention mechanism powers every modern model. A Microsoft grant on a mask-attention transformer is a reminder that the foundational IP is held by a handful of incumbents.
AI
Vendors compete on how much context a model can hold — and charge by the token to hold it. A 2026 grant on integrated inference-time memory is the IP under that pricing axis.
AI
High-density AI racks run hot, and cooling them is a growing slice of data-center capex. A 2025 application on combined air-and-liquid cooling shows where that spend goes.
AI
Before a hyperscaler commits billions to an accelerator, it models how the chip will actually perform. A 2025 grant on doing exactly that sits under every procurement decision.
AI
Vendors love to promise cheaper inference. Two granted-and-pending quantization patents show the real lever — and the accuracy trade-off the marketing skips.
AI
Intel is fighting to stay relevant in AI silicon. A granted sparse-matmul patent shows the efficiency angle it is patenting to compete on cost-per-compute.
AI
Amazon rarely headlines its silicon. A granted patent on a processing-element array, by a named Amazon chip architect, is a window into the cost play behind AWS AI capacity.
AI
Microsoft keeps telling investors it is capacity-constrained. Its own mixture-of-experts patents are part of the answer to that constraint — do more with the compute you have.
AI
A 2024 application names the engineers behind Google's tensor processing units and describes a multi-modal matrix-multiply array — the math engine the cloud-capex debate runs on.
AI
Every enterprise-AI deck promises answers grounded in your own data. Two early-2026 RAG grants show the mechanism vendors are selling — and patenting.
AI
Inference is the cost that scales with every query. A December 2025 application on optimizing the key-value cache is a claim on shaving that recurring bill.