NVIDIA Engineer Sakeena Fiza Hunts Hardware Failures Before AI Customers See Them
Sakeena Fiza’s job at NVIDIA starts with a deceptively simple question: How could this system fail?
As a validation engineer in NVIDIA’s data center systems engineering lab, Fiza works on the hardware and systems that underpin modern AI infrastructure. Her team tests new systems long before customers depend on them, searching for failures that could emerge when components are pushed to their limits and deployed at scale. NVIDIA profiled Fiza’s work on September 23, 2026.
“Validation engineers look in the shadows and shine a light into every corner,” Fiza said in the NVIDIA profile. “Every time we get a system, our first thought is: how can it break?”
The work illustrates a less visible part of the AI hardware race. Building increasingly powerful GPUs is only part of the challenge. Those components also have to function reliably as complete systems, from initial laboratory bring-up to racks, clusters, production lines and customer AI factories.
The First Test Begins When the System Gets Power
Fiza’s work begins during the earliest stages of a new system.
Engineers bring components online one at a time, integrate boards and work alongside firmware and software teams to determine whether the system is behaving as expected. A successful first boot can become a major milestone because it demonstrates that the individual pieces are beginning to function together.
One of Fiza’s most memorable moments came when an NVIDIA Rubin GPU first enumerated at the system level.
The display simply identified the hardware as “NVIDIA Corporation Device,” she recalled. But inside the lab, the moment prompted celebration because it represented the first time the Rubin GPU had been recognized at the complete-system level, according to NVIDIA.
That first sign of life, however, is only the beginning.
NVIDIA’s current Rubin infrastructure includes systems designed to operate at large AI-infrastructure scales. For example, the company says its DGX Rubin NVL8 uses eight Rubin GPUs, while the broader Vera Rubin platform is designed around multiple rack-scale systems operating together.
That makes system-level validation increasingly important.
Validation Engineers Become the “First Customers”
Fiza describes validation engineers as the first customers for a product.
Their job is to exercise hardware under different conditions before customers have to rely on it. The objective is not simply to demonstrate that a system works once. Engineers have to determine whether it continues working when subjected to stress and real-world variables.
“The goal is to always catch issues before customers catch it,” Fiza said.
The potential problems can exist at very different levels.
A rack-scale failure could involve high-speed signaling, thermal margins or power integrity. A much smaller physical issue, such as a screw being tightened too far or dust inside a customer environment, can also create unexpected behavior.
When a system fails, engineers attempt to reproduce the problem and progressively narrow down its cause. That can involve examining logs, changing test conditions, investigating firmware, eliminating mechanical variables, probing signals and studying oscilloscope measurements.
Why AI Hardware Is So Difficult to Validate
Modern AI infrastructure is not a single component.
A board can contain tens of thousands of components, while an entire rack can contain hundreds of thousands of parts. Those components must operate together under demanding conditions rather than simply functioning independently.
That complexity is one reason Fiza sees validation as a cross-disciplinary role.
Her work can involve hardware, firmware, software, mechanical design, thermal behavior, manufacturing and the eventual customer experience. She described the role as giving her the opportunity to work across several engineering disciplines rather than remaining confined to one area.
“I wish people understood how complex the hardware is that AI needs to run on,” Fiza said.
As AI infrastructure expands from individual accelerators to large-scale systems, problems can also become more difficult to isolate. A failure may involve interactions between multiple layers of the system rather than one defective component.
That is where the investigative approach becomes important.
From Dubai to NVIDIA’s AI Hardware Labs
Fiza’s interest in systems began well before she joined NVIDIA.
Growing up in Dubai, she was introduced to programming through Logo. Her interest in building systems later included working on Mars rovers at a high school robotics camp and unmanned aerial vehicles during college.
She eventually earned a bachelor’s degree in computer science and engineering from the University of California, Irvine.
At NVIDIA, she found a role that allowed her to work with an entire machine rather than focusing exclusively on one component.
The collaborative nature of the work is another defining feature.
Fiza described system bring-up as being similar to “the Avengers assembling,” with architects, designers, software engineers, firmware engineers and validation engineers working together toward a functioning system.
The Hidden Work Behind AI Infrastructure
The broader significance of Fiza’s role is that much of the reliability of AI infrastructure is established before the hardware reaches a customer.
NVIDIA’s Rubin systems are being developed for increasingly large AI workloads and infrastructure deployments. The company says Vera Rubin is designed as a multi-rack system for large-scale AI workloads, while DGX Rubin NVL8 is built around eight Rubin GPUs.
Before those systems can operate reliably in production environments, engineers have to identify problems that may not appear during straightforward testing.
That means deliberately looking for failure.
For Fiza, that mindset turns hardware debugging into an ongoing investigation. Each unexplained failure becomes another puzzle: reproduce it, eliminate possibilities, find the underlying cause and determine whether the fix holds under other conditions.
The work may happen largely behind the scenes, but it is an important part of turning advanced AI hardware into infrastructure that can operate at scale.
“With the products we have in the pipeline, I’m so excited,” Fiza said. “They’re going to change the world.”
