.jpg)
In our latest MARA Fireside, CEO Fred Thiel sits down with Janet George — Executive Vice President of AI at Mastercard and MARA board member — to explore how Bitcoin mining is unlocking a new model for energy-efficient AI infrastructure.
In this MARA Fireside chat, Janet George explores how energy efficiency will shape the future of AI and how Bitcoin mining could play a surprising role in solving AI’s power problems. Large language models (LLMs), such as those used in generative AI, consume energy in unpredictable, spiky patterns. This creates inefficiencies and cost burdens. Bitcoin mining, which can be easily ramped up or down, offers a unique solution: it can absorb unused energy during AI’s idle times, effectively acting as a power balancer.
George emphasizes that MARA is positioned to become an energy intermediary that decouples energy supply from demand. Just as cloud computing decoupled applications from hardware, MARA aims to abstract energy usage from rigid infrastructure. This could enable a more balanced, efficient system where energy is directed to wherever it’s most needed, whether for AI workloads, data centers, or even factories.
One of the biggest opportunities lies in transforming how data centers operate. Traditional centers are inefficient, both in energy use and cooling. New two-phase immersion cooling technologies not only reduce energy and water waste but also allow the reuse of heat, for example, to warm nearby buildings. This creates potential for decentralized, modular data centers that are both more sustainable and cost-effective.
George also introduces the idea of using MARA’s systems to optimize LLMs by understanding their energy needs in greater detail. Encoding stages of AI models require intense compute power, while decoding relies more on memory. By analyzing these patterns, MARA can fine-tune energy usage and drastically reduce waste. The energy that would otherwise be unused could then be monetized through Bitcoin mining, creating a new value stream.
As AI applications become more personalized the need for decentralized, edge computing will grow. These edge systems can’t rely on traditional, centralized infrastructure. MARA's ability to orchestrate energy and workloads dynamically could support this future, making AI more scalable and available to industries like healthcare, education, and transportation.
Ultimately, George sees energy as the bottleneck to widespread AI adoption and the biggest lever for unlocking its potential. By reducing waste, improving infrastructure, and enabling new models of load balancing, MARA aims to accelerate AI deployment while contributing to a more sustainable and equitable energy system. The intersection of AI, Bitcoin, and smart energy orchestration, she argues, is not just a technical opportunity. It’s a generational one.
