TLDR: Sid Ghatak, a former tech advisor to the Biden administration, has issued a warning to the incoming Trump administration regarding the overlooked ‘fully loaded costs’ of seemingly inexpensive AI deals with companies like OpenAI and Anthropic. He highlights that beyond the initial $1 per agency fee for a year’s access, significant expenses will arise from training, data consolidation, and ongoing inference, urging a comprehensive financial assessment.
An Indian-origin former US government tech advisor, Sid Ghatak, has raised concerns about the true financial implications of low-cost artificial intelligence (AI) deals being offered by leading AI firms, specifically OpenAI and Anthropic. Ghatak, who previously served as a tech advisor during the Biden administration, is cautioning the Trump administration to look beyond the initial attractive pricing, such as the ‘$1-a-year deals’ for federal workers, and consider the substantial hidden costs involved in implementing these advanced AI solutions.
OpenAI and Anthropic recently announced plans to provide federal agencies with a year’s access to their AI models for a nominal fee of $1 per agency. However, Ghatak emphasizes that this token fee does not represent the total expenditure the government will incur. In an interview with Business Insider, he pointed out critical areas of additional investment. “What does $1 really mean? Does that mean access to the open model and that compute will be charged incrementally?” Ghatak questioned, highlighting the potential for escalating compute costs.
He further elaborated on the expenses associated with tailoring and operating these AI models. “So, there are all of those costs in terms of training, and then once you have built the model, does the government have to pay for inference whenever federal workers and contractors use it? Is there another incremental charge?” These questions underscore the need for the government to budget for model training, which involves adapting the AI to specific governmental requirements, and subsequent inference costs, which are incurred each time the AI models are utilized by federal personnel.
Beyond direct payments to AI providers, Ghatak stressed the necessity for significant internal investment in data infrastructure. He noted that the government would need to “invest heavily in consolidating its data for AI.” This crucial step ensures that the AI models can operate effectively and produce reliable outputs. Ghatak concluded with a strong recommendation: “The government has to be really careful and understand what the fully loaded cost is of these solutions beyond the very attractive zero to $1 charge. These AI models are powerful engines, but they require really fantastic data to run cleanly and produce reliable output. The investment in that is something that needs to be understood.”
Also Read:
- Microsoft and OpenAI Ink Memorandum of Understanding for Enhanced AI Collaboration and Corporate Restructuring
- FTC Launches Broad Inquiry into AI Chatbot Risks for Minors, Targeting OpenAI, Meta, Snap, and Others
This warning serves as a critical reminder for policymakers to conduct thorough cost-benefit analyses that encompass all phases of AI integration, from initial setup and customization to long-term operational and data management expenses.
