FinTech moves fast. News is everywhere, clarity isn’t.
FinTech Weekly delivers the key stories and events in one place.
Click Here to Subscribe to FinTech Weekly's Newsletter
Read by executives at JP Morgan, Coinbase, BlackRock, Klarna and more.
Nvidia answered one of the two hard questions about AI agents at GTC this week. The other one — the financial one — it did not touch.
On March 16, at the SAP Center in San Jose, Jensen Huang announced NemoClaw, an open-source software stack that installs on top of OpenClaw in a single command and adds the security and privacy controls that have kept enterprises on the sidelines.
OpenClaw, which Huang described as the operating system for personal AI and the fastest-growing open-source project in history, was built for individuals. It had no controls over what data an agent could access, where it could send information, or how its actions could be audited.
NemoClaw addresses all three. It installs OpenShell, a runtime that isolates each agent in its own sandboxed environment, enforces company-defined access policies, and supports both local models running on-device and cloud models accessed through a privacy router that prevents internal data from being exposed externally. The platform is hardware-agnostic — it does not require Nvidia chips to run.
Huang framed the enterprise deployment question in explicit terms. Every company, he said, now needs an OpenClaw strategy in the same way every company once needed a Linux strategy and an HTTP strategy. NemoClaw is the infrastructure layer that makes that viable for regulated industries. What had held procurement, legal and compliance teams back was not capability. It was the absence of a way to define and enforce limits on what agents could access, execute and report. NemoClaw removes that objection.
What it does not address is what happens when an agent reaches the boundary of a company's internal systems and needs to transact with the outside world.
Traditional financial infrastructure was built around a single assumption: the entity on the other side of a transaction is a human being with a verified legal identity. KYC rules, AML requirements, and payment network participation agreements all reflect that assumption.
An agent that autonomously books cloud compute, purchases datasets, or pays for external services does not fit that model. It cannot open a bank account. It cannot hold payment credentials tied to a verified human account. The rails that carry most business payments were not designed for it.
As FinTech Weekly reported in March, crypto infrastructure offers a parallel path. Crypto wallets require no identity verification. An agent that holds a wallet can send and receive value and pay for services autonomously.
Coinbase launched Agentic Wallets on February 11 on its x402 protocol, a payments standard built specifically for machine-to-machine transactions. That infrastructure is live. The problem is that most business payments still run on traditional rails, not crypto rails. The gap between where agent payment infrastructure is today and where most commercial transactions actually happen is the question the financial system has not yet answered.
Nvidia built the governance layer for AI agents this week. The question of how those agents participate in the traditional financial system — not the crypto alternative, but the infrastructure that processes most of the world's business payments — remains open.
Editor's note: We are committed to accuracy. If you spot an error, a missing detail, or have additional information about any of the companies or filings mentioned in this article, please email us at [email protected]. We will review and update promptly.
