Accelerated AI


Accelerated computing for AI in government

By Anthony Robbins and Edward Van Buren

A joint publication from the Deloitte AI Institute for Government and NVIDIA.

Accelerated computing to the rescue

The government is building momentum in adopting artificial intelligence (AI) and is seeing more use cases go into production. Between 2020 and 2021, federal agency budget requests for AI research and development increased by more than 50%.

Major legislation, including the 2020 National AI Initiative Act, the 2019 AI in Government Act, and the 2018 OPEN Government Data Act, is laying the groundwork for widespread integration of AI across government agency operations. To further coordinate and advise federal agency AI efforts, the National AI Advisory Committee was launched in September 2021 and the White House created a National AI Initiative Office in January 2021.

Despite these advances, many agencies on the AI journey will soon confront substantial challenges and obstacles—if they haven’t already. At scale, AI often requires access to large data sets, demanding compute resources, speed, and complex networking. As AI implementation gathers pace across government, agencies may find that they do not have ready access to such capabilities and assets. The challenges arise because the unique computational needs of AI quickly outpace the capabilities of traditional data center architectures. 

The trade-offs and dilemmas posed by AI needs and computational considerations are rarely obvious and always evolving. Agencies need to design a flexible plan for incorporating accelerated computing in the early stages of AI adoption. Moreover, AI needs and applications will evolve over time.

Accordingly, the following steps are important considerations:

  • Build a clear AI strategy that identifies how an agency’s mission and operations can be reinforced and scaled up through the prediction, simulation, automation, and other core functions of artificial intelligence.
  • Identify within your existing AI strategy those use cases that will be most aided by accelerated computing, specifically ones that involve distributed data collection and flows, provide real-time analysis and speed, and involve complex dynamics, among other factors.
  • Be mindful of user perspectives and interests. Existing workflows and workforce practices will have a major impact on the potential benefits of AI as well as the shape, location, and distribution of accelerated computing architectures. Significant gains could be left unrealized if human design considerations are treated as an afterthought.
  • Identify a multidisciplinary team that can help define an execution plan. Team members need extensive industry knowledge, AI expertise, and a demonstrated understanding of accelerated computing and solution development.
  • Consider future needs. Issues to weigh in selecting the right projects include long-term relevance; relative costs and benefits of an AI solution; scaling up pilot initiatives; and security and confidentiality. 
  • Consider dedicated computing options to realize the inherent benefits of greater computing power and speed as usage increases. This is especially relevant as usage and data processing requirements increase and needs and inference become more nuanced. 

Agency leaders have a great opportunity to apply AI to help their organizations. As usage gets more widespread and complex, they are going to need to utilize accelerated computing architecture solutions, such as GPUs and optimized software, to support their AI journey. Planning for these capabilities early will allow for seamless expansion and sophistication of AI use cases.


1 NITRD, “Artificial intelligence R&D investments,” Office of Science and Technology Policy, The White House, accessed October 11, 2021.

Did you find this useful?