Microsoft tests in-chip microfluidic cooling for Artificial Intelligence chips

Microsoft demonstrated an in-chip microfluidic cooling system that removes heat up to three times better than today’s cold plates, targeting the growing thermal load of Artificial Intelligence silicon.

Artificial Intelligence chips are running hotter as new architectures drive higher performance, and conventional cooling is reaching its limits. Microsoft says it has successfully tested an in-chip microfluidic system that brings liquid coolant directly into the silicon, removing heat up to three times better than advanced cold plates depending on workload and configuration. Microchannels etched into the back of the chip allow coolant to flow precisely where heat is generated, and Microsoft used Artificial Intelligence to identify each chip’s thermal hotspots so the coolant can be routed more efficiently.

In lab-scale testing, the approach reduced the maximum temperature rise inside a GPU by 65 percent, though the exact benefit varies by chip type. The system has effectively cooled a server running core services for a simulated Microsoft Teams meeting, indicating readiness beyond benchtop experiments. Microsoft expects the technology to improve power usage effectiveness and reduce operational costs by enabling warmer coolant temperatures and better heat removal than traditional methods, which are constrained by multiple layers that sit between a chip and a cold plate.

Building a practical microfluidic solution required system-level engineering across silicon, coolant, servers and datacenters. Microsoft collaborated with Swiss startup Corintis to use Artificial Intelligence for bio-inspired channel designs that cool hotspots more efficiently than straight channels. The channels are hair-width and must be deep enough to circulate coolant without clogging yet not so deep that they weaken the silicon. The team iterated on leak-proof packaging, coolant formulations, etching methods and a manufacturing flow to add etching steps, producing four design iterations in the past year.

The company positions microfluidics as a way to enable more power-dense hardware and to manage spiky workloads more economically. For services like Teams, where activity surges around meeting start times, the improved thermal headroom could support safe overclocking during peaks instead of overprovisioning servers that sit idle most of the time. Liquids conduct heat better than air, and by removing insulating layers and cooling the silicon directly, microfluidics can operate with less-chilled fluid, saving energy and improving opportunities for high-quality waste heat use.

Microsoft is investigating how to integrate microfluidic cooling into future generations of its first-party chips, including its Cobalt and Maia families, and plans to work with fabrication and silicon partners to bring the technology into production across its datacenters. Beyond immediate efficiency gains, the company sees microfluidics as a path to denser racks and new chip architectures such as 3D stacking, where fluid could flow between stacked dies. Microsoft also encourages industry-wide adoption, arguing that broader use will accelerate innovation and sustainability benefits for cloud infrastructure.

72

Impact Score

Will agentic Artificial Intelligence disrupt SaaS?

Bain argues that agentic Artificial Intelligence is set to reshape software as a service by automating tasks and rebundling control across a new stack. The firm outlines four disruption scenarios and a playbook for incumbents on data, standards, pricing, and talent.

Intuit advances GenOS to accelerate agentic artificial intelligence development

Intuit expanded its GenOS platform with custom financial large language models, expert-in-the-loop tooling, and agent evaluation frameworks to speed agentic artificial intelligence across its products. Early results show improved accuracy and significantly lower latency in accounting workflows, with more agents rolling out soon.

Advancing cutting edge artificial intelligence in health

Google details how breakthroughs like Gemini, Gemma variants, AlphaFold, and enterprise tools are being applied across healthcare. The company highlights research, open models, and clinical search products that aim to make care more personalized, accessible, and effective with Artificial Intelligence.

Contact Us

Got questions? Use the form to contact us.

Contact Form

Clicking next sends a verification code to your email. After verifying, you can enter your message.