Energy-Aware AI: Rethinking AI Platforms for a Sustainable Future
About this session
Energy-Aware AI: Rethinking AI Platforms for a Sustainable Future
The rapid adoption of AI is driving unprecedented energy consumption. Data centers are consuming record levels of electricity, with global demand projected to reach 133 gigawatts by 2026, driven largely by AI workloads. At the same time, major technology companies continue to invest hundreds of billions of dollars in expanding AI infrastructure and data center capacity.
While AI has dramatically lowered the cost of creating digital assets, it has also led to significant overproduction and duplication of content. The ease of generating text, images, code, and other artifacts encourages repeated computation with little visibility into its energy cost. Today, AI users understand token consumption, but have virtually no awareness of the energy required to produce those tokens.
I believe this needs to change. Energy awareness should become a first-class concern of AI platforms, where every AI operation has a measurable energy footprint. Just as cloud platforms made compute, storage, and network usage visible, future AI platforms should expose energy consumption as a fundamental metric alongside latency, throughput, and token usage.
Achieving this vision requires more than operational monitoring—it demands a fundamental architectural rethink. Most AI platform components today, including retrieval-augmented generation (RAG) pipelines, vector databases, transactional databases, and NoSQL stores, were built on architectures designed for traditional applications. AI transformations are stored, retrieved, and processed much like conventional data, without optimizing for the unique compute, network, and storage characteristics of AI workloads.
The next generation of AI platforms must embed energy optimization into the lowest levels of processing. This may require rearchitecting core platform components—including databases, storage engines, retrieval systems, and inference pipelines—to minimize compute, data movement, and storage overhead while maintaining accuracy and performance.
This talk explores two key themes:
Energy-aware AI — moving beyond token-based metrics to quantify AI processing in terms of energy consumption and making energy visibility a core platform capability. Rearchitecting AI platforms for sustainability — examining how AI infrastructure, data storage, retrieval engines, and processing pipelines can be redesigned with energy efficiency as a primary architectural principle rather than an afterthought.
As AI becomes foundational to nearly every industry, improving model efficiency alone will not be sufficient. Sustainable AI requires rethinking the architecture of the platforms that power it, making energy optimization a fundamental design objective from the ground up.
Speaker
Key takeaways
- Learn how to define and measure energy consumptions at the foundations of processing
- Learn how to think about completely re-architecting existing systems
- Learn how to define and measure sustainability