Nvidia is replacing the gaskets on its chips: the next battleground in thermal design is the tiny gap of millimeters between the chip and the cold plate.

Starting with the Rubin's switch to graphite thermal pads, let's discuss why thermal interface materials have suddenly become the focus.

In 2026, AI chip cooling revolutionized our understanding. First, the traditionally "conservative" NVIDIA announced that its Vera Rubin platform would use 45°C warm water liquid cooling-data centers no longer needed to be iceboxes. Then, supply chain sources confirmed that the standard Rubin version (TDP 1800–2000W) abandoned liquid metal at the last minute before mass production, switching entirely to high thermal conductivity graphite pads. Only the Ultra top-of-the-line version (2500–2850W) retained the extreme solution of liquid metal plus a microchannel cold plate.

Looking at these two events together reveals a more important signal: as single-card power consumption surged from 700W for the H100 to the 2000W level, the bottleneck in heat dissipation was no longer "whether the server rack could keep cool," but rather in a more microscopic area-the interface between the chip and the cold plate, a few tenths of a millimeter. Liquid cooling removes heat from the cold plate, but the heat must first be able to "enter" the cold plate. No matter how smooth the chip and cold plate surfaces are, under magnification, they are all microscopic gaps, and the air within these gaps is one of the worst thermal conductors.

The Vera Rubin (standard version) has a TDP of 1800–2000W, and before mass production, it abandoned liquid metal and replaced it with graphite-based thermal pads;

The Ultra top-of-the-line version has a TDP of 2500–2850W, retaining liquid metal + microchannel cold plate, and entered mass production in Q3.

-Quoted from public reports

This 1-millimeter battlefield is the thermal interface material (TIM). The requirements for the pads are actually very stringent: fast conductivity-the thermal conductivity must keep up with the chip's power density; compactness-soft enough and low-stress to fill uneven surfaces without damaging the chip; and stability-in high-power power supplies, new energy battery packs, and 5G base stations, insulation and flame retardancy are the bare minimum, not bonuses. This is precisely why we created the DP series. Shengtong Qiyuan doesn't offer a single, universally applicable thermal pad; instead, we've developed a complete product line of thermally conductive silicone pads covering different power levels, allowing engineers to choose the right model for their needs, rather than just using whatever's available.

Shengtong Qiyuan High Thermal Conductivity Series Silicone Pads

Three Differences Between Us and Ordinary Thermally Conductive Silicone Pads
Difference 1: Thermal conductivity ranges from 1.5 to 25 W/m·k, with 17 models across a single series.

From DP150 to DP2500: Standard range 1.5–5.0, high conductivity 6.0–10.0, ultra-high conductivity 12/15/18, and the flagship DP2500 achieving 25±1 W/m·k. All models are unified under the DP-Pad series, with thermal conductivity cross-validated according to both ASTM D5470 and ISO 22007-2 standards. The project upgrades from conventional heat dissipation to ultra-high power without needing to change suppliers or re-test materials. Thickness can be ordered up to 20mm, and cutting and die-cutting are supported. Composite fiberglass substrate, single/double-sided silicone backing, or adhesive backing can be customized as needed.

The second difference is that while achieving high thermal conductivity, insulation and flame retardancy are not compromised. A common hurdle in the industry is that "high thermal conductivity comes at the cost of insulation." The DP ultra-high thermal conductivity model maintains a withstand voltage ≥6 KV/mm and a volume resistivity ≥10¹³ Ω·cm at 12–18 W/m·k; the DP standard model has an even higher withstand voltage ≥10 KV/mm. All models meet UL94 V-0 flame retardancy rating and RoHS/REACH environmental requirements. With the new national standard for battery safety setting "no fire or explosion after thermal runaway" as a mandatory red line, insulation and flame retardancy are no longer optional for automotive and power supply customers.

Difference 3: Ultra-soft and low-stress, effectively protecting and supporting the chip.

With a hardness Shore 00 as low as 20-30, the surface is slightly adhesive, exhibiting excellent compression resilience and a maximum elongation at break of 105%. It adheres perfectly to heat-generating modules with uneven thickness and surfaces, leaving no air gaps. Low deformation stress prevents damage to the chip and solder joints, and it is less prone to cracking and failure even under long-term high and low temperature operation and repeated disassembly and reassembly.

Next step in heat dissipation: interface materials.

High-performance graphics cards and AI accelerator cards, high-power industrial power supplies, 5G base stations and optical modules, OBCs and new energy battery packs-in these scenarios, the DP series does the same thing: protect the last few tenths of a millimeter before the cold plate. NVIDIA's replacement of the Rubin gasket is not a mystery, but a clear engineering choice: when power density has exhausted the potential of air cooling and liquid cooling, what truly determines the upper limit of temperature difference is the first step in removing heat from the chip.

If you are also evaluating heat dissipation solutions for high-power modules, feel free to contact us with your structural diagram. Tell us your power, thickness, and insulation requirements, and we'll directly help you find the most suitable option in the DP series-whatever you need, we'll have it.