NVIDIA explores non-stick cookware material to boost AI chip speeds

The transmission bottleneck

Modern AI chips are currently pushing the boundaries of how fast data can move between components. As these systems scale, they are reaching a transmission bottleneck where they are pushing 200Gbps per lane. At these extreme speeds, even the slightest amount of interference can significantly degrade the performance of the hardware.

To maintain optimal server throughput, engineers need to find ways to keep signals clear over long distances. Because co-packaged optics—a technology that integrates optical fibers directly into the chip package—is not yet ready for widespread use, the industry must find a way to improve the circuit boards themselves. This makes the physical material of the board a critical factor in the next generation of AI infrastructure.

NVIDIA is reportedly looking for a solution to these headaches by exploring an unlikely material: PTFE. This is the chemical compound better known by the brand name Teflon, which is famously used as a non-stick coating on cookware. By repurposing a household material for high-end computing, the company hopes to overcome the limitations of current electronics manufacturing.

Why Teflon works for data

The primary reason for considering PTFE is its ability to reduce signal loss. In the world of high-speed electronics, signal energy is often wasted as it travels across a board. PTFE wastes far less signal energy than the resins that are currently used in standard circuit boards, making it a strong candidate for high-performance applications.

Industry sources suggest that NVIDIA plans to use this material in a Rubin Ultra switch board or a next-generation orthogonal mid-plane. These components allow high-speed switching to happen on the printed circuit board (PCB) rather than relying on complex copper cables. Moving the switching onto the board itself can simplify the architecture while maintaining the necessary speed.

Analysts have noted that current boards are reaching 70-plus layers in complexity. At this scale, even top-grade M9 materials are falling short of the required specifications. This leaves PTFE as a logical choice for engineers who need to maintain signal integrity while building increasingly dense and complex hardware structures.

Manufacturing hurdles

Despite its electrical properties, PTFE presents significant challenges for mass production. One of its defining characteristics is its non-stick nature, which means it bonds poorly with other materials. This makes it difficult to create a board with dozens of layers, as the layers cannot simply stick together like traditional epoxy or hydrocarbon resins.

The material is also described as thin and soft, which complicates the manufacturing process. Unlike standard resins that bond well under heat, drill cleanly, and plate easily, PTFE is harder and costlier to process. These physical limitations mean that while the material is technically superior for signals, it is much more difficult to work with in a factory setting.

Because of these difficulties, current production yields are reportedly below 5 percent. The material is still in the sample stage, meaning it has not yet been proven for large-scale commercial use. These manufacturing complexities have led to skepticism among some industry observers who question why the material hasn't been used in circuit boards for the last 30 years.

The path to 2030

The move toward PTFE is not without its critics. Some industry chairmen have suggested that NVIDIA may only be asking about the material because of shortages in high-end laminates. They argue that the material's history is limited mostly to antenna-type applications in military radars, satellites, and automotive radar systems, rather than complex multi-layer boards.

However, NVIDIA's partners across China, Taiwan, and the United States have already begun securing the equipment necessary to handle PTFE. This suggests that the company is serious about moving past the sample stage and into actual production. The goal is to create a reliable supply chain that can eventually overcome the current processing bottlenecks.

Analysts predict that the processing hurdles for PTFE could be solved within the next two years. While the technology is still in its early stages, a clear outcome for its use in mainstream AI hardware is not expected until around 2028 to 2030. Until then, the industry will watch closely to see if this non-stick material can truly solve the AI hardware bottleneck.