Cooling the AI Revolution: Castrol's Fluids Power NVIDIA's AI Factories

PANGBOURNE, England – August 18, 2026 – In a move that underscores a critical inflection point for the AI industry, legacy lubricant giant Castrol announced that two of its specialized cooling fluids have been officially validated by NVIDIA for use in its AI factory and data center infrastructure. While the announcement of a new industrial fluid might seem esoteric, it represents a foundational piece of the puzzle in sustaining the explosive growth of artificial intelligence. The validation of Castrol ON PG25 and PG25T direct-to-chip coolants is more than a simple product endorsement; it’s a signal that the very architecture of computation is being redesigned from the inside out, driven by the immense thermal challenges of modern AI.

For years, data center operators have relied on sophisticated air conditioning to manage heat. But the computational density of today's AI workloads, powered by GPUs that can each draw over a kilowatt of power, has rendered that approach obsolete. The industry is facing a thermal barrier, and overcoming it requires a fundamental shift in systems thinking—moving from cooling the room to cooling the chip itself. This partnership between a fluid engineering expert and a semiconductor titan provides a powerful glimpse into the new industrial ecosystems forming to build the future.

The Heat is On: Why AI Needs a Cooling Revolution

The engine of the AI revolution is generating an unprecedented amount of heat. As models become more complex and data sets grow exponentially, the server racks powering them are becoming incredibly dense and power-hungry. Thermal Design Power (TDP) for a single high-end chip now routinely exceeds 1,000 watts, and a single server rack can consume hundreds of kilowatts—the equivalent of a small neighborhood. At these densities, simply blasting cold air is like trying to cool a blast furnace with a desk fan. The result is thermal throttling, where chips slow down to prevent overheating, which negates the very performance gains they were designed to deliver.

This is where liquid cooling, specifically direct-to-chip (DTC) technology, becomes a mission-critical enabler. The market recognizes this necessity, with projections showing the global data center liquid cooling market rocketing from around $5.6 billion in 2024 to nearly $50 billion by 2034. Castrol's ON PG25 fluids are designed for this new reality. In a DTC system, these propylene glycol-based fluids are circulated through small cold plates mounted directly onto the heat-generating components like CPUs and GPUs. Liquid is up to 1,000 times more effective at transferring heat than air, allowing it to be whisked away from the processor with surgical precision.

Beyond raw performance, this shift has profound implications for sustainability. Castrol's fluids operate in a closed-loop system, meaning the coolant circulates continuously without atmospheric exposure. This design drastically improves Water Usage Effectiveness (WUE) by minimizing the evaporative water loss common in large-scale air-cooling systems. In an era where data centers' water consumption is under intense scrutiny, this is a significant operational and environmental advantage. Furthermore, by reducing the energy needed for fans and chillers, liquid cooling can lower a data center's overall Power Usage Effectiveness (PUE) and cut energy costs by as much as 40%, creating a compelling business case that aligns performance with planetary responsibility. The higher-grade heat captured by liquid is also far easier to reuse for applications like district heating, turning a waste product into a valuable asset.

From Engine Oil to AI's Core: A Strategic Systems Pivot

For a 125-year-old company built on lubricants for engines and industrial machinery, Castrol's foray into data center thermal management is a masterclass in strategic evolution. The company is leveraging its deep, historical expertise in fluid dynamics, material compatibility, and thermal stability to solve one of the 21st century's most pressing technological challenges. This isn't just about creating a new product; it's about repositioning the company as an indispensable partner in the digital infrastructure supply chain.

The NVIDIA validation is the linchpin of this strategy. It provides a powerful stamp of approval that goes far beyond a technical datasheet. According to the company, the validation process was exhaustive, covering not only performance metrics like corrosion resistance and thermal stability but also Castrol's business resilience and global supply-chain capabilities. For hyperscale data center operators and AI factory architects, this dual assurance is non-negotiable. They need to know not only that the fluid will prevent their multi-million dollar hardware from melting, but also that the supplier can deliver it reliably at a global scale.

"The inclusion of Castrol ON PG25 and PG25T within NVIDIA-Validated AI Factory Infrastructure gives customers added confidence in both the fluids' performance and Castrol's ability to provide consistent global support across markets," said Peter Huang, Global President of Thermal Management & Data Centres at Castrol. His colleague, Sung A. Kim, the company's Data Centre Global Technology Manager, reinforced this commitment to quality, stating, "We evaluate every thermal management fluid under rigorous test conditions before it reaches the market. The validation from NVIDIA confirms that PG25 (OAT) and PG25T (IAT) hold up to that standard, even at the scale and intensity of today's AI infrastructure."

This endorsement significantly strengthens Castrol's hand in a competitive field that includes chemical giants like 3M and BASF, as well as specialized cooling hardware firms like CoolIT Systems and Vertiv. By securing a place within NVIDIA's tightly curated ecosystem, Castrol moves from being one of many fluid providers to a validated component of a complete, high-performance solution.

Building the AI Factory: An Ecosystem of Innovation

NVIDIA's concept of the "AI Factory" reframes high-performance computing as an industrial-scale utility. It's an integrated system of hardware, networking, and software designed to produce intelligence. Like any factory, it depends on a complex and reliable supply chain. The validation of third-party components like Castrol's fluids is central to building out this ecosystem, allowing customers to assemble best-in-class infrastructure with confidence.

The rigor of the validation process ensures that every component is fit for purpose. The review of material compatibility, for instance, is critical for long-term reliability. In a liquid-cooled system, an improperly formulated fluid can cause galvanic corrosion between different metals, leading to catastrophic leaks and system failure. By pre-validating fluids like PG25, NVIDIA de-risks the adoption of liquid cooling for its customers and accelerates the deployment of next-generation AI systems.

Castrol's alignment with broader industry movements, such as its recent OCP Inspired™ recognition from the Open Compute Project Foundation, further cements its role as a collaborative player. Such standards are essential for creating interoperable, scalable, and cost-effective solutions that prevent vendor lock-in and foster innovation across the industry. The future of cooling technology is moving toward even tighter integration with hardware, with research at companies like Microsoft exploring microfluidics that embed cooling channels directly into the silicon. These advancements will demand even deeper collaboration between chip designers, system architects, and fluid engineers.

What the Castrol-NVIDIA partnership makes clear is that the next industrial revolution will be built not by lone geniuses, but by complex ecosystems of specialists. The unglamorous but essential work of managing heat has officially moved from a facilities-level concern to a core element of computer architecture, proving that enabling the most advanced artificial intelligence often comes down to mastering the fundamental laws of physics.