Corintis and the Thermal Frontier: Reinventing GPU Cooling for the AI Era

As AI accelerators push toward 1000W power envelopes and 3D chip architectures, the industry faces a bottleneck that’s not computational—it’s thermal. Enter Corintis, a Swiss startup emerging from stealth with a radical solution: in-chip microfluidic cooling. In collaboration with Microsoft, Corintis has demonstrated a system that cools GPUs up to 3× more effectively than traditional cold plates, reducing silicon temperature rise by 65%.

This article explores Corintis’s technology, its partnership with Microsoft, the physics of microfluidics, and what this breakthrough means for the future of AI infrastructure, chip design, and data center economics.

The Problem: Heat Is the New Bottleneck

AI chips are getting hotter. The early versions of ChatGPT trained on 400W Nvidia GPUs. Today’s top-tier accelerators—like Nvidia’s Blackwell and AMD’s MI300X—are pushing 800–1000W, with plans for stacked 3D chips that will generate even more heat.

Traditional cooling methods—cold plates, air flow, and immersion—are hitting their limits:

• Cold plates sit atop the chip, separated by thermal interface layers.
• Air cooling is inefficient at high densities.
• Immersion cooling is expensive and hard to retrofit.

Without a breakthrough, AI compute will be throttled by heat—not silicon.

The Solution: Microfluidics Inside the Chip

Corintis’s approach is bio-inspired. Instead of cooling the chip from the outside, it etches hair-sized channels directly into the silicon, allowing coolant to flow through the chip itself.

Key features:

• Direct-to-silicon cooling: No thermal interface layers.
• AI-optimized routing: Coolant flows to hotspots based on chip layout.
• High-grade waste heat: Output temperatures reach 70°C, enabling reuse.

Microsoft’s tests showed:

• 3× better heat removal than cold plates.
• 65% reduction in maximum temperature rise.
• Potential for overclocking during peak demand (e.g., Teams meeting spikes).

Sponsored Banner

Corintis: From EPFL to Microsoft

Founded in 2022 out of research at the Swiss Federal Institute of Technology in Lausanne (EPFL), Corintis combines:

• Simulation software to model chip heat maps.
• Optimization algorithms to design coolant paths.
• Advanced manufacturing to etch microfluidic channels.

The company has shipped over 10,000 cooling systems for AI deployments and raised $33.4M in funding, including a $24M Series A led by BlueYard Capital. Notable board members include:

• Lip-Bu Tan, Intel CEO and Walden International chairman.
• Geoff Lyon, founder of CoolIT.

Microsoft Partnership: A Strategic Validation

Microsoft partnered with Corintis to test microfluidic cooling in its Azure data centers. The results were published in September 2025 and include:

• Four design iterations over 12 months.
• AI-assisted routing of coolant channels.
• Bio-inspired layouts, mimicking butterfly wings and leaf veins.

Microsoft’s goal is to enable:

• Power-dense chip designs with more features in less space.
• Dynamic overclocking during peak usage.
• Reduced cooling energy and improved sustainability.

The company believes microfluidics could unlock 3D chip architectures, where stacked layers require internal cooling.

Technical Challenges and Design Tradeoffs

Microfluidics isn’t trivial. Key challenges include:

• Channel depth: Too deep weakens silicon; too shallow clogs.
• Reliability: Liquid inside chips raises concerns about leaks and corrosion.
• Manufacturing: Requires precision etching and co-packaged cooling.

Corintis addresses these with:

• Co-design workflows between chip and cooling teams.
• Modular integration with existing data center infrastructure.
• Simulation-driven iteration to meet short cycle times.

CEO Remco van Erp describes chips as “cityscapes with billions of transistors.” Cooling them requires custom coolant networks, not generic copper fins.

Industry Implications: Infrastructure, Efficiency, and Sustainability

Microfluidic cooling could reshape:

• Data center design: Smaller footprints, higher density.
• Energy efficiency: Less power spent chilling coolant.
• Waste heat reuse: High-grade heat can be repurposed for buildings or industrial processes.

It also enables:

• Longer chip lifespans by reducing thermal stress.
• Higher performance through sustained overclocking.
• Lower operational costs via improved power usage effectiveness (PUE).

Microsoft says the technology could allow chips to “flow liquid through themselves,” a concept once considered science fiction.

Sponsored Banner

Competitive Landscape: Nvidia, AMD, Intel

Nvidia has adopted liquid cooling for its latest GPUs, but not yet microfluidics. AMD’s MI300X uses cold plates. Intel, under Lip-Bu Tan, is expected to explore co-packaged cooling for future accelerators.

Corintis’s edge lies in:

• In-chip integration
• AI-assisted design
• Bio-inspired layouts

If adopted at scale, microfluidics could become a standard feature in next-gen AI chips.

Future Outlook: From Cooling to Compute

Corintis envisions a future where:

• Cooling is part of chip design, not an afterthought.
• AI helps cool AI, using models to optimize thermal flows.
• 3D chips become viable through internal cooling.

The company is opening offices in the US and Germany to support global deployments. It’s also working with “many of the world’s largest tech companies,” though names remain undisclosed.

Cooling as a Catalyst

Corintis isn’t just solving a problem—it’s enabling a future. By embedding cooling into the chip itself, it removes one of the last physical barriers to AI scale. Its partnership with Microsoft validates the approach, and its funding and board signal serious momentum.

As AI chips grow hotter and more complex, Corintis’s microfluidic cooling may be the key to unlocking the next generation of compute—where performance is no longer throttled by heat, and cooling becomes a catalyst, not a constraint.

Read more in our Breakthroughs section.