NEAR Intents Tops $37B as AI Infrastructure Expands

NEAR Protocol says its cross-chain execution infrastructure has surpassed $37 billion in cumulative volume as the ecosystem expands its confidential AI and agent tooling. In an official November 3 ecosystem recap, NEAR reported more than $37 billion in all-time NEAR Intents volume across over 37 million swaps, alongside more than 1 million NEAR committed through its AI staking infrastructure. The figures are project-reported cumulative metrics rather than a measure of assets currently locked in the system.

NEAR Intents is designed around outcome-based execution: users or applications specify the result they want while competing solvers determine how to execute it across supported networks. The architecture abstracts bridging, routing and chain-specific execution rather than requiring users or AI agents to manually coordinate every underlying transaction. Confidential Intents adds a privacy layer to that framework, extending an approach previously introduced through private cross-chain execution infrastructure.

What NEAR shipped last month:

✦ NEAR Intents crossed $37B+ in all-time volume and 37M+ total swaps.
✦ Confidential Intents TVL crossed $145+ million.
✦ 1,000,000+ $NEAR is now staked to power confidential inference.
✦ @brave Leo now runs on private AI inference from NEAR AI…

— NEAR Protocol (@NEARProtocol) November 3, 2026

NEAR AI Staking Passes 1 Million NEAR

The AI side of the ecosystem has expanded separately. NEAR’s November recap says more than 1 million NEAR is now staked through NEAR AI’s infrastructure. Under the official staking model, users retain ownership of their principal while protocol staking rewards are redirected toward credits for eligible AI services. The mechanism converts staking economics into compute access rather than paying users the staking rewards as investment yield.

Those credits can support confidential inference and persistent IronClaw agent hosting. The distinction matters because the 1 million NEAR figure does not establish how much AI inference has actually been consumed or how the stake is divided between services. Earlier IronClaw deployment infrastructure already linked NEAR staking to always-on agent hosting, while NEAR AI Cloud’s confidential inference stack uses Intel TDX and NVIDIA confidential-computing hardware for supported workloads. Staked value therefore measures provisioned access capacity, not AI request volume or recurring user demand.

NEAR AI Cloud has also expanded its model catalog. Claude Opus 5.5, GPT-6.1 Sol, Kimi K3.1 Pro and other externally provided models are available through its Incognito routing tier, while models including GLM 5.3 Flash and selected Qwen and DeepSeek variants can run directly inside NEAR-hosted confidential TEEs. Model availability and confidential execution are not synonymous: externally routed models remain subject to their providers’ infrastructure and do not receive NEAR AI’s model-level TEE guarantee.

Brave Moves Confidential Inference Into Production

Brave provides the clearest consumer deployment in the latest update. Since October 15, users selecting GLM 5.3 Flash in Brave Leo have their prompts encrypted in the browser and processed through NEAR AI Cloud’s hardware-isolated environment. Brave strips the originating IP before forwarding the encrypted request, while attestation verifies the code and model running inside the TEE. This is a production integration rather than a test deployment, although it applies specifically to users selecting the supported confidential model.

NEAR AI has simultaneously expanded security-oriented infrastructure through IronClaw, its Confidential GPU Marketplace and NEAR AI RED. The marketplace connects confidential-computing-capable GPU providers with protected workloads, while RED gives security teams access to reduced-refusal models for authorized cybersecurity research. These products extend the stack across compute, inference and agent security, but their availability should not be treated as evidence of equivalent production usage across every component.

Developer acquisition is continuing alongside that infrastructure buildout. NEAR’s current AI agent hackathon offers more than $250,000 across projects using Intents, Chain Signatures, chain abstraction and NEAR AI Cloud. The stronger adoption signal will come from sustained swap activity, paid AI workloads and recurring agent usage after those development incentives end, rather than from infrastructure availability alone.