John Eichberger |
May 2024
Forty years ago, in the year 1984, Kyle Reese arrived in Los Angeles to prevent Cyberdyne Systems Model 101 from killing Sarah Connor. He was sent from the year 2029, just five years from now. It was said in the movie that the machines “got smart.” While purely a work of science fiction, there are some similarities from the world of “The Terminator” to what is happening with technology today. The rapid expansion of artificial intelligence has fueled a historic rise in the stock market, created a company valued at nearly $3 trillion and prompted a surge in data center construction in the U.S. AI is taking over responsibilities previously completed by humans and, while not “self-aware” as SkyNet became, AI is learning to “think.” While this is very interesting in general, and ripe fodder for conspiracy theorists, the implications for transportation and energy in general are real and significant. The question for us is how might the rise of AI reduce emissions of on-the-road transportation while preserving and improving affordable and reliable energy for all communities?
This is a topic the Transportation Energy Institute’s Board of Advisors has directed staff to examine and about which to foster discussions. In this blog, we will only scratch the surface on this dynamic and rapidly evolving topic, but we will provide more opportunities to engage and learn in the near future. Let’s dip our toes in the water first.
Energy Demand
I decided to ask AI to tell us about itself and asked ChatGPT, “write an article evaluating the energy demand associated with the rise of artificial intelligence.” Here is what came back regarding energy demand for AI:
The energy demand of AI stems from various components of its infrastructure and operations:
- **Training of AI Models**: Developing AI models requires enormous computational power, often provided by data centers equipped with high-performance computing hardware. These data centers consume vast amounts of electricity to train complex algorithms on massive datasets.
- **Inference and Deployment**: Once trained, AI models are deployed for inference tasks, where they make predictions or decisions based on new data inputs. While inference typically requires less energy compared to training, the deployment of AI at scale across devices and systems still contributes to overall energy consumption.
- **Hardware Requirements**: High-performance hardware, including GPUs (Graphics Processing Units) and TPUs (Tensor Processing Units), are integral to AI operations. These components are power-hungry and require substantial cooling infrastructure, adding to energy demands.
- **Data Storage and Transmission**: AI applications rely on vast amounts of data stored in servers and transmitted between devices and data centers. Data storage facilities and network infrastructure consume significant energy, especially as data volumes continue to grow exponentially.
ChatGPT continued to present the environmental consequences of such increased energy demand as well as opportunities to mitigate these consequences. While interesting observations, I prefer to look at the data behind the observations. If we look at the rise of data center construction and related energy demand in the U.S., we see a six-fold increase since 2020. This is a significant increase in energy demand from just one sector.
Energy Supply and Demand
While not all of the data center construction is exclusively for the deployment of AI, I have heard anecdotal reports and forecasts that energy demand for AI will require a near-tripling of U.S. electricity generation within the next ten years. Alex DeVries, a researcher at the Vrije Universiteit Amsterdam, authored “The Growing Energy Footprint of Artificial Intelligence,” which examines the environmental impact of AI systems. In this paper, he calculated the energy demand of using 1.5 million AI servers expected to be delivered by Nvidia through 2027. If these servers operated at full capacity, they could consume 85.4 – 134.0 terawatt-hours of electricity each year, globally. This amount has been compared to the energy consumed by Argentina, Netherlands or Sweden. (An article on the report can be found here without a subscription.)
This amount is also equal to 3% of the net electricity generation of the U.S. in 2023. While this may seem insignificant, it is important to note that net generation in the U.S. has increased by only 3% since 2005 and in recent years the domestic system has struggled to satisfy demand in some areas of the country.
Can the electricity generating sector accommodate such a rapid rise in energy demand from this one source, not even considering other factors influencing demand levels? According to the U.S. Energy Information Administration, the production of electricity in the U.S. is projected to nearly double by 2050. This represents a significant shift from historic changes in domestic electricity generation capacity and comes at a time when available resources for generation are being limited to those that represent a lower impact on the environment. Globally, EIA projects a 50% increase in generation led by a 150% increase in the contribution of renewable energy resources. (Coal remains relatively unchanged while nuclear and natural gas are projected in increase around 23% each.) Despite such forecasts, achieving such expansion will be challenging. This is exacerbated when one considers that actual future demand could exceed expected production capacity, especially when we consider the uncertainty of demand from AI combined with the global effort to electrify most energy consumption sectors, including transportation.
Artificial Intelligence and Transportation
The overall energy demand of this technology is a fascinating topic, but how the rise of AI affects transportation is the specific interest of TEI. So again, I turned to ChatGPT and asked for insights. It returned the following (I have summarized some of the content for brevity):
Artificial Intelligence (AI) has had a profound impact on transportation, revolutionizing various aspects of the industry, from vehicle autonomy to traffic management and logistics. Here’s how AI influences transportation:
- Autonomous Vehicles [ChatGPT provides commentary on development and deployments, safety improvements and mobility solutions.]
- Traffic Management [Again, commentary provided relative to predictive analytics, dynamic routing and smart traffic lights.]
- Logistics and Supply Chain [Commentary on route optimization, predictive maintenance, and warehouse automation.]
- Public Transportation [Demand forecasting, fare collection systems and safety and security.]
Conclusion: Artificial Intelligence is transforming transportation by making it safer, more efficient, and more accessible. From autonomous vehicles and smart traffic management to logistics optimization and public transit enhancements, AI-driven innovations are reshaping the way people and goods move around the world. As AI technologies continue to evolve, they will play an increasingly vital role in shaping the future of transportation towards a more sustainable and interconnected ecosystem.
What’s Next?
I do not disagree with the ChatGPT summary of AI’s impact on transportation – there is tremendous potential for beneficial development. And despite my introduction to this article, I do not believe SkyNet is coming for us…yet. But the energy demand profile of AI does concern me, especially when framed against a backdrop of increased demand for electricity in general and the historic slow pace of adding generating capacity.
As we endeavor to reduce emissions from the transportation sector, preserve affordable and reliable access to energy and enhance safety, we must continually think about external factors that could affect any transition that might be required. Artificial intelligence is a factor that can and most likely will disrupt industries throughout the world, including transportation. It is incumbent upon transportation energy leaders to consider what role AI will play and how it may affect efforts to achieve the industry’s objectives and to prepare to accommodate its involvement the system.
More on this topic to come.
