Companies are grappling with whether to implement powerful artificial intelligence (AI) models that promise enhanced reasoning capabilities but come with substantial computing and integration costs.
The surge of new AI releases from industry leaders like OpenAI and Google has sparked intense debate among business leaders about the real-world value proposition of these systems. While proponents argue the latest models could revolutionize strategic planning through improved analysis and forecasting, many companies are proceeding cautiously as they evaluate infrastructure requirements, staff training needs, and potential return on investment. The key tension centers on whether the gains in areas like data analysis and automated reasoning can justify implementation costs.
“We see that organizations are cost-optimizing their use of AI to ensure that only the required capabilities of AI models are used for specific employee or customer needs,” Omer Rosenbaum, co-founder and CTO of Swimm, an AI documentation platform, told PYMNTS.
“Multiple models may be used in different components of a solution. For example, the same solution could use [GPT] 4o-mini for common requests with low cognitive requirements and o1 or even o1-pro for specific, more complex requests that have high value to users.”
The last few weeks saw a wave of new AI model releases from major tech companies. OpenAI introduced o3, an advanced reasoning model that surpassed its predecessor, o1, in tasks like coding and mathematics. Currently, in safety testing, o3 is set for public release in January 2025.
Google launched Gemini 2.0, a model capable of multistep problem solving with minimal human input, integrated across its AI services, including Search. It also released Whisk, an AI tool that generates and remixes images for creative workflows.
Meanwhile, Meta unveiled Meta Motivo, enhancing digital avatars’ realism in the metaverse to improve virtual experiences. These advancements underscore the rapid evolution of AI.
Corporations are evaluating the adoption of advanced AI models like OpenAI’s o1 model, which offers enhanced reasoning capabilities. Introduced in September 2024, it is designed to handle complex multistep tasks with advanced accuracy, making it particularly effective in fields such as math, science and coding.
However, the financial implications are significant. Training state-of-the-art AI models can cost up to $500 million per session. Beyond training expenses, companies must consider integration costs depending on complexity and customization.
Despite these challenges, several corporations are pressing forward. For instance, Microsoft recently purchased 485,000 Nvidia Hopper AI chips, double the amount bought by Meta, as part of its $13 billion investment in OpenAI.
“The steep costs for cutting-edge AI are a temporary hurdle,” Mike Conover, CEO and co-founder of Brightwave, an AI platform for financial services, told PYMNTS.
“If history has taught us anything, it’s that technology democratizes over time. Today, businesses focus on tools that integrate seamlessly into existing workflows rather than requiring bespoke build-outs. The real trade-off is between upfront implementation costs and the opportunity cost of falling behind in adopting systems that can uncover insights at a scale and precision that humans cannot match. As AI capabilities evolve, the value proposition for such tools becomes increasingly clear.”
According to industry experts, that rapid evolution is already visible in the market. A data scientist, Rebecca Cavallo, told PYMNTS that the trade-off between implementation costs and AI capabilities is continuous.
“The capabilities of large language models are doubling every 5-14 months,” Cavallo said. She noted that “GPT-3.5 (released March 2022) vs. GPT-4o-mini (more advanced reasoning model released July 2024) is around 3.3 times cheaper.”
Cavallo emphasized that businesses need to remain agile. “You need to evaluate — do you have the right infrastructure? Can your developers handle any changes to processes if you adopt a new model? Does it make sense to pay for the extra work to get a new model in production?” she said.
The post AI Race Pushes Companies to Weigh Costs Against Promised Gains appeared first on PYMNTS.com.
