Nvidia's Jetson Orin Nano Gets Better With Age

Here's a question the tech industry rarely asks: What happens to hardware after the press release? Most devices follow a predictable arc—launch, brief moment of relevance, gradual abandonment. The Nvidia Jetson Orin Nano is doing something different, and it's worth paying attention to why.

Gary Sims, who runs the Gary Explains YouTube channel, has been tracking this $249 AI development board since 2023. His latest video—his third on the platform—makes an argument that runs counter to our usual hardware narratives: the Jetson Orin Nano is actually improving with age.

That's not how this usually works.

The Price Cut That Stuck

The "Super" version of the Jetson Orin Nano launched in 2025 at $249, down from the original $599. In consumer electronics, dramatic price cuts often signal either desperation or artificial scarcity setup—drop the price to move inventory, then watch it creep back up or vanish entirely.

Sims checked. A year later, the board is still available at roughly that price across multiple retailers. No scalpers. No artificial scarcity. Just consistent availability at the stated price point. For anyone familiar with GPU markets over the past few years, this feels almost deliberately un-Nvidia.

The hardware specs remain what they were: six 64-bit ARM Cortex-A78 cores, a 1,024-core Nvidia Ampere GPU, 8GB of RAM, and an M.2 slot for storage. The board can scale between power modes—critical for edge deployment where you might be running off batteries or need to prioritize performance over efficiency.

Software Updates as Product Strategy

What's genuinely unusual is what's happened on the software side. The original 2023 boards received a 1.7x performance gain in generative AI workloads through software updates alone. Memory bandwidth jumped from 68 GB/s to 102 GB/s—again, not through new hardware, but through optimization.

"Even today, Nvidia worked to make the 2023 model have all of these gains in 2025," Sims notes. "And basically, it hasn't stopped. It's still developing and releasing updates."

The current software stack runs JetPack 6.2.1 (released June 2025) with Ubuntu 22.04 LTS and kernel 5.15. But JetPack 7.2 is coming, bringing Ubuntu 24.04 and kernel 6.8 to the Orin Nano series. That's another five years of Ubuntu support stacked on top of whatever Nvidia continues to ship.

This matters because most development boards exist in a kind of software purgatory—they ship with whatever Linux distribution was current at launch, maybe get one or two updates, then languish. The ecosystem fragments. Libraries break. Eventually you're stuck choosing between security updates and functionality.

The Jetson Orin Nano is getting the opposite treatment: active development, major version upgrades, continued integration with the broader Ubuntu ecosystem. It's the software equivalent of keeping your phone updated instead of buying a new one every year.

Benchmarks Against the Field

Sims ran comparative benchmarks that illuminate where the Jetson Orin Nano actually sits in the market. Running Gemma 3 (4 billion parameters), the board delivers 12 tokens per second. That's compared to:

• Raspberry Pi 5: 3.5 tokens/second
• Fourth-gen Intel i3 desktop: 5 tokens/second
• Celeron J4125: 1.5 tokens/second

The performance delta comes from that integrated Nvidia GPU, which handles inference workloads far more efficiently than CPU-only approaches. For local LLM work, the difference between 12 tokens/second and 3.5 is the difference between usable and frustrating.

Sims also tested Ollama running Llama 3.2 (3 billion parameters). In January 2025, he measured 20.5 tokens/second. By June, that had jumped to 22.5 tokens/second—a 10% improvement from Ollama's own optimization work. No special builds required. Just install and run.

"You just go and install Ollama and it just runs and you get a performance increase," Sims explains. The board remains compatible with mainstream tools as they evolve.

What Longevity Actually Costs

There's a tension here worth examining. Nvidia isn't exactly known for supporting budget hardware into perpetuity. The company's business model has historically centered on planned obsolescence—new GPU architectures every couple of years, software features locked to newer hardware, pressure to upgrade.

So why the continued investment in a $249 development board?

One possibility: the Jetson line serves as a gateway drug for Nvidia's enterprise offerings. Developers who cut their teeth on Orin Nano boards eventually need to scale up, and Nvidia has the entire product stack ready. Supporting the entry-level hardware keeps the ecosystem healthy and the pipeline full.

Another: edge AI is strategically important enough that fragmenting the market with abandoned hardware would be counterproductive. Better to maintain one well-supported platform than chase the latest specs every year.

Or maybe—and this feels almost quaint to suggest—there's value in building things that last. The environmental and economic cost of constant hardware churn is real. A development board that remains relevant through software improvements rather than forced upgrades is a different relationship with technology.

The Unsexy Metric

Sims's entire thesis boils down to a single word: longevity. Not performance. Not features. Not disruption or innovation or any of the other terms we use to juice press releases.

Longevity.

It's possibly the least exciting metric in consumer technology, which might explain why it's so rare. We don't review products three years after launch to see if they're still supported. We don't benchmark software improvements on existing hardware. We focus on what's new because what's new is what generates attention.

But for developers actually building things—robotics projects, vision systems, edge inference deployments—knowing your hardware will remain supported matters more than chasing marginal spec improvements. The Jetson Orin Nano appears to be delivering on that promise in ways its price point wouldn't suggest.

Whether that continues remains an open question. JetPack 7.2 is promised, but promises are cheap. The real test is five years from now, when the board is genuinely old by tech standards. Will it still receive updates? Will the ecosystem still support it?

For now, at least, Nvidia seems committed to making a $249 board from 2023 work better in 2026 than it did at launch. That's unusual enough to be worth documenting.

Marcus Chen-Ramirez is a senior technology correspondent for Buzzrag.