Smart, Not Supersized: Rethinking AI for Sustainable Government Services
By Dr. Darren Pulsipher, Chief Solutions Architect, Intel
As government agencies steadily improve their digital services, many see artificial intelligence as a way to accelerate their efforts. But there’s a catch: AI systems consume tremendous power. Tech giants like Microsoft, Amazon, and Google recently invested in nuclear power to meet their skyrocketing needs. While that works for big tech, it’s not a realistic solution for government agencies trying to balance citizen services with sustainability goals. The good news? Most agencies don’t need that level of computing power in the first place.
Also Read: Taking Advantage of Gen AI With Next-level Automation
The Power Problem
Data centers currently consume 1 to 2% of the global power supply, but Goldman Sachs Research estimates that the demand for AI will push that up to 3 to 4% by the end of the decade. Along with more power, the carbon dioxide emissions of data centers may more than double between 2022 and 2030.
Governments at every level try to address environmental concerns by changing their operations. At a federal level, the Office of Management and Budget directs agencies buying AI services to consider the environmental impact — including carbon emissions and data center resource consumption — and see how vendors can improve the efficiency of their services. The recent Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence ordered millions of investments into developing more efficient AI models, hardware, and data centers.
Many AI tools that rely on large language models — OpenAI’s ChatGPT, Google’s Gemini, Anthropic’s Claude, and Meta’s LLaMA — are now common household names. How much energy they consume depends on several factors, such as how long it takes to train them, what kind of hardware they run on, and how efficiently the data centers operate. However, the most significant factor is the size of the model. The bigger the model, the more power it needs to fuel performance. These famous LLMs run trillions of parameters, but many tasks agencies need to perform could easily run on smaller, more cost-effective models.
Think of it like choosing between a Ferrari and a golf cart: It really depends on your budget and where you need to drive. Sure, the Ferrari is powerful and impressive. Still, the golf cart is the smarter choice if you’re moving between a couple of buildings on campus. The same principle applies to AI and digital services.
Making Smart Choices
So, how can agencies deliver better digital services while staying green? Start with the problem, not the technology. Before jumping into any solution, ask, “What are we trying to accomplish?” and make sure to stay focused on that goal. Agencies must closely examine their specific workflows and requirements. Does your agency need creative text generation or just accurate data extraction? Do you need to process multiple languages simultaneously or focus on one specific translation pair? By answering these questions first, agencies can choose right-sized solutions that balance capability with sustainability. This approach helps avoid overprovisioning while ensuring the flexibility to scale specific functions as needed.
In the U.S. Census Bureau, analysts spent weeks reading 30- and 40-page Securities and Exchange Commission filings like 10-Ks and 10-Qs, manually extracting data into databases. By focusing on this specific use case — document analysis and data extraction — they could implement a targeted AI solution that dramatically reduced processing time without requiring massive computing resources.
While going for the biggest, most powerful AI solutions is tempting, specialized tools often work better. Take translation services, for example. As an immigration hub, Los Angeles County serves over 75 language communities and needs a model that can provide accurate translations. While a general-purpose LLM is an option, a specialized AI model running on a laptop can handle this perfectly, using about as much power as a couple of light bulbs. Compare that to the massive energy needed to run large AI models in cloud environments — and accompanying large bills — and the choice becomes clear.
Also Read: What is a CAO and are they needed?
A Better Way Forward
By making smarter technology choices, agencies can deliver excellent digital services while meeting environmental goals. Just as we’ve learned to think twice before printing emails, it’s time to think carefully about our computing choices.
Look at the big picture. Sustainability isn’t just about power consumption. When agencies move services online, they’re not just saving energy in their operations — they’re helping citizens save time and reduce car trips. These ripple effects often have an even bigger environmental impact than direct energy savings.
The technology is ready. The solutions exist. Now it’s just a matter of making smarter choices. And sometimes, that means choosing the golf cart over the Ferrari.
Comments are closed.