The company’s so-called “cold revolutions” started in 2023 with basic cooling upgrades, followed in 2024 by water cooling plates for AI kit. Now we’re apparently heading into “cold revolution 3.0”, where water cooling plates and vapor chambers are elevated to the status of strategic materials.

The shift is being driven by Nvidia’s upcoming Rubin and Feynman AI platforms. These double transistor counts and computing grunt, while guzzling power at 2,000 to 3,000 watts, which would cook any datacentre without serious cooling upgrades.

Traditional water cooling plates use waterways between 1 and 3 mm wide and usually rely on thermal interface material between the chip and cooling solution. Nvidia’s MLCP design shrinks the waterways down to micron scale and integrates the vapor chamber, water cooling plate, packaging cover and chip. By cutting out a layer of thermal interface material, the coolant is brought closer to the die, improving efficiency.

AI datacentres with hundreds of servers risk meltdown if cooling falls short, so the stakes are high.

Taiwanese cooling specialists Shuanghong, Qihong and Jiance are all working on the new MLCP gear, which promises not only better heat dissipation but fatter profit margins for suppliers.

The dark satanic rumour mill has manufactured a hell on earth yarn claiming Shuanghong has already shipped samples for testing, while Qihong and Jiance have shown Nvidia can design and produce the tech.