The research community is still exploring AI’s potential, searching for applications that transform the way we live and work. This creates a moving target for computer engineers who are working to optimize hardware and software for energy efficiency.

We’ve moved beyond an era where users explicitly request AI services. Today, AI is increasingly embedded behind the scenes — powering recommendations, improving search results, enhancing applications — and this deployment is driving up inference costs.

I recently joined @AliVelshi on MSNBC to discuss how AI is reshaping electricity costs for American households. Most people are already using AI whether they realize it or not. https://t.co/UNGihiXl66

The symposium was a fantastic reminder of the interdisciplinary nature of this field. It was inspiring to share the stage with Axel Kaiser, @MathiasUnberath, and @johntasioulas. Thank you to the JHU DSAI team for organizing an inspiring symposium!

I shared our perspectives on renewable energy, utility-scale batteries, and demand response. I also made the case for specialized AI models that compute the same responses with far fewer calculations.

It was a privilege to speak at the @JohnsHopkins Data Science and AI Institute (DSAI) Fall 2025 Symposium: “AI in Daily Life.” We must rethink the infrastructure powering the tech that promises to change the way we live and work.

I close by highlighting opportunities to collaborate between computer systems and operations researchers. Thanks to Grant Ruan for organizing the session and sharing perspectives on power systems! Enjoyed connecting with Mohammad Hajiesmaili and Prateek Sharma too!

We also estimate how energy is used by individual AI tasks and incentivize users to modulate power in response to grid conditions. Specialized AI models could compute the same response with fewer calculations.

At the INFORMS Annual Meeting earlier this semester, I spoke on sustainable datacenters for AI. We take a holistic approach to infrastructure for datacenters, accounting for energy generation and storage.

With Smithsonian Magazine, I discuss how AI training and inference drive energy use. Sustainable computing requires carbon-efficient energy as well as models that deliver responses with fewer calculations. https://t.co/p2KZqpfT1S

Spoke with @MarketWatch on AI and nuclear power. AI will consume hundreds of TWh annually. From my work on efficient computing, intermittent renewables alone are insufficient for gigascale datacenters. Nuclear is essential part of solution space. https://t.co/Hh3r27ZYQp

The University of Pennsylvania's Department of Electrical and Systems Engineering invites applications for tenure-track faculty positions at all ranks in computer engineering and photonics/electromagnetics. Review begins December 1, 2025. Apply:

— Industry may shift toward more efficient, specialized models and away from the diminishing returns from ever larger models. — Datacenter compute will eventually be useful, but when?

— Today’s investments are, by definition, risky because we see massive capital flow in datacenter capacity even as demand for AI is uncertain. — Demand for AI hinges on accelerating technology adoption, emergence of a “killer app”.

I spoke with @MarketWatch on the massive @Oracle -@OpenAI deal and potential for investment bubble in AI datacenters, a critical issue as industry pours capital in infrastructure. Datacenters are built for uncertain demand and is, by definition, risky. https://t.co/bllLqeCdDg

I shared perspectives with @WIRED's @sophiecharara on relationship between @Apple's AI and sustainability. AI may increase refresh rates for consumer electronics, which are poorly utilized and emit carbon during manufacturing. Hurdles for sustainability. https://t.co/AU2LsOEVID

Excited about (i) building datacenters alongside carbon-efficient energy sources, (ii) developing specialized AI models that deliver capability with fewer calculations, (iii) managing computational jobs to match energy cost. #EnergyEfficiency #AI #DataCenters #Sustainability

Thrilled to co-host a recent workshop with Arthur van Benthem at @Wharton. We convened an exceptional, multi-disciplinary group to tackle these issues. While challenges are significant, opportunities for innovation are immense. https://t.co/biZryvPsyr

Excited to contribute to @engadget's coverage of @OpenAI's open-weight LLMs, their first in six years. Sharing learned weights enables developers to use and fine-tune models without needing hyperscale infrastructure, lowering barriers to innovation. https://t.co/G1EcZl4Rlr

Great to see David Irwin's insights too! We agree on opportunities and challenges. #AI #Datacenters #CarbonAwareComputing #EnergyEfficiency #GreenAI #ComputerScience