Introducing DaTEG® 2.0: Active Thermal Management for the Next Generation of AI Chips

DaTEG® 2.0 is here and ready

After two years of development and more than a year of testing, GemaTEG's next-generation active thermal management system is market-ready. It is patented. It is validated. And it was designed to support not only today's AI chips, but the next three to five generations.

Why This Matters Now

Everyone is talking about the agentic revolution. AI agents that manage workflows, optimize operations, make decisions. They are becoming more pervasive every day.

What most people forget is that all of this runs on physical infrastructure. Servers. Chips. Data centers. And that infrastructure is already at capacity.

AI runs on hardware. The more we use agents, the more computing power they require. More computing power means more load on infrastructure that is already running hot. If you think computing power is abundant and available on demand, you are in for a difficult surprise.

The agentic future has a bottleneck. It starts at the chip.

Three Challenges DaTEG® 2.0 Was Built to Solve

Today's AI chipsets face three issues simultaneously.

Power density is unprecedented. A single chiplet can consume the same power as an oven, packed into a space smaller than a smartphone. The cooling technologies designed to handle this have been around for decades. Their ability to keep pace is reaching its limits.

Chiplets are not uniform. High Bandwidth memory (HBM) and the GPU integrated within the same package have different thermal tolerances, with HBM requiring a lower maximum operating temperature than the GPU. This is a defining characteristic of the architectures powering AI today.

Workloads are not constant. Training, inference, mixed operations. Data center loads shift continuously, and each shift changes how the chiplet is exercised thermally. We are no longer dealing with static hotspots. We are dealing with entire regions activated differently depending on what the system is doing at any given moment.

DaTEG® 2.0 was built to address all three.

Cooling vs. Thermal Management

The industry's default response to heat has been straightforward: make things colder. Lower the fluid temperature. Increase the flow. Push harder. The goal is the lowest possible temperature, all the time, everywhere.

This was designed for a simpler era, when chips were more uniform and workloads were more predictable.

Cooling and thermal management are not the same thing.

Cooling focuses on heat removal. Thermal management focuses on temperature control: precisely where and when it matters most. The difference is enormous when your chiplets are non-uniform, your workloads are shifting, and your margins are shrinking.

A uniform cooling approach, no matter how aggressive, cannot address what modern AI chips actually need. What is required is the ability to control temperature locally, precisely, and in real time.

That is the paradigm GemaTEG operates in. That is why we built DaTEG® 2.0.

What Is Inside the Package

From the outside, DaTEG® 2.0 looks and feels like a standard cold plate. It sits on top of the chiplet the same way any liquid cooling solution would.

Inside, it is a fundamentally different system.

DaTEG® 2.0 integrates multiple technologies into a single, compact package: active elements that precisely control temperature, passive elements for heat distribution, single- and dual-phase thermal transfer, and electronics with real-time software control. Each region of the package performs a different function depending on how the chip is operating at that moment.

The result is a system that orchestrates the full state of the art in thermal management within one integrated unit. It responds to changing workloads in real time, faster than any solution currently on the market.

This is not a prototype. It is a product. And because the system is integrated by design, it is also built to last.

Patented. Tested. Future-Proof.

Three things make this release significant.

It is patented. The architecture behind DaTEG® 2.0 is protected by a published worldwide patent covering its integrated approach to differential thermal control.

It has been tested for over a year. We could have announced this sooner. We chose to wait until we had achieved the level of validation and readiness we set for ourselves.

It is future-proof by design. Because DaTEG® 2.0 integrates multiple thermal technologies into a single adaptable architecture, it was not built for a single family of chiplets. It is designed to support three to five future generations without requiring a complete redesign. Every new chip generation brings a significant jump in power. Thanks to our strategic collaborations with the R&D departments of chip manufacturers, we anticipated that trajectory and built ahead of it.

In an industry where every new generation typically forces a full rework of the thermal solution, this changes the equation.

‍

Who DaTEG® 2.0 Is For

We work as closely as possible with chip makers because integration at the source creates the most value. When DaTEG® is part of the package from the start, everyone downstream benefits: server builders, system integrators, and data center operators.

At the same time, we work directly with data centers facing three realities.

• Data center operating costs in 2026 are escalating rapidly, driven primarily by an insatiable demand for AI compute power. DaTEG® 2.0's advanced thermal management capabilities drastically reduce OpEx for cooling, which typically accounts for around 50% of the overall cost of running a data center.
• The second is the cost of constant reinvestment. Every new generation of chiplets typically requires reworking the thermal infrastructure. DaTEG® 2.0 extends the useful life of that investment.
• The third goes beyond hardware. Through the GemaTEG ecosystem, our installations feature AI-driven controls that optimize thermal performance and operational efficiency in real time, tuned to the KPI that matters most.

‍

Integration Is the Point

GemaTEG is not a cooling company. We are a thermal management company. Our mantra is integration.

We are software and hardware. In a future where agentic AI is managing not only your workflows but also your data centers, the infrastructure layer needs to be intelligent, adaptive, and integrated from the chip level up.

The agentic revolution will only happen if the infrastructure beneath it can keep up. DaTEG® 2.0 is how we make sure it does.

Watch the Full Conversation

I recently sat down to walk through DaTEG® 2.0 in detail: what it is, how it works, and why it matters now.

Watch the full interview here: https://www.youtube.com/watch?v=Pl83IpRnFMs

To see DaTEG® 2.0 in action, contact us to request a demo: https://gemateg.com/contact

Maurizio Miozza Co-Founder & CCO, GemaTEG

‍