Navan has deployed NVIDIA's Nemotron 3.5 Super VL to power its AI travel agents in production, moving the corporate travel and expense platform's automation effort onto one of NVIDIA's newest large language models.

The announcement, carried by marketscreener, is a deployment claim, not a results claim. Navan hasn't yet published numbers on booking conversion, deflection rates, or cost savings tied to the model swap. That gap matters for travel sellers watching whether AI agents can actually handle corporate bookings end to end — or whether they still push users back to human agents at the moment of payment.

What does the deployment actually change?

Nemotron 3.5 Super VL is a vision-language model, meaning it can process images alongside text. For travel, that capability typically maps to tasks like reading invoices, receipts, and screenshots of itineraries — the unstructured paperwork that corporate travel management still handles with manual entry.

Navan's AI agents sit inside a platform that serves corporate clients for booking, expense management, and card spend. The company has positioned AI as a way to cut service costs and speed up travelProgram workflows. What Navan hasn't said in this announcement is which agent functions run on the new model, how much traffic it carries, or whether NVIDIA's model replaced an incumbent model or runs alongside one.

For competitors — Amex GBT, CWT, SAP Concur, and a wave of AI-first booking startups — the signal is that foundation-model upgrades are now a competitive lever in corporate travel distribution. Cheaper, faster inference changes the economics of automated servicing, and platforms that own their model stack can undercut rivals on per-transaction cost.

Why NVIDIA's name carries weight here

NVIDIA has been marketing its models and infrastructure aggressively to enterprise software buyers, and travel is a visible proving ground. A production deployment at a platform of Navan's scale gives NVIDIA a reference customer in a vertical where AI agent reliability is genuinely hard: fares change mid-search, loyalty rules complicate pricing, and corporate travel policies constrain what an agent can book.

The caveat: a production deployment is not the same as measured performance. Until Navan publishes benchmarks — booking success rates, agent handle times, service cost per trip — the claim rests on architecture, not outcomes. Trade buyers should treat it as a pipeline upgrade announcement, not proof that AI agents are taking share from human travel advisors.

The takeaway in plain words: Navan has moved its AI travel agents onto NVIDIA's newest model in live operations, but the numbers that would prove it matters commercially haven't been published yet.

Expect Navan to follow with usage or efficiency data as it pitches the deployment to corporate clients; whether that data holds up under scrutiny will determine if this is distribution muscle or marketing theater.