Artificial intelligence (AI) is revolutionizing various industries, but this technological advancement comes with a heavy cost in terms of power consumption. Training and using AI models require substantial electricity, which is straining national infrastructure and impacting operational costs and the environment. For instance, a single request on ChatGPT uses almost ten times more electricity than a Google search, and the AI service's daily power consumption is comparable to that of 180,000 U.S. households. Additionally, a single ChatGPT conversation consumes nearly 17 ounces of water, highlighting the significant resources required for AI operations.
The rapid expansion of the AI market has led to a surge in data center construction, which necessitates enormous amounts of specialized processors, extensive security infrastructures, and substantial electricity. Over the next decade, electricity demand from these data centers is projected to double, and by 2040, 14% of global emissions will be attributed to the Information and Communications Technology (ICT) industry. Companies like Amazon.com Inc. (NASDAQ: AMZN) are expected to invest heavily in building new data centers, with Amazon alone planning to spend more than $150 billion over the next 15 years.
The U.S. power grid is unlikely to manage the increased load without significant investment. Goldman Sachs estimates that over $50 billion will be required for necessary upgrades. This comes at a time when the nation is already investing heavily to modernize the grid, with $22 billion allocated since 2021 to support growing demands from transitioning away from natural gas appliances, expanding the electric vehicle (EV) market, and safeguarding against disruptions caused by extreme weather events or cyberattacks.
Despite these challenges, solutions to reduce power consumption and increase efficiency are emerging. The transition from incandescent light bulbs to energy-efficient alternatives in 2023 serves as a precedent. Similarly, AI applications are now being developed with a focus on efficiency at scale. Companies like Brand Engagement Network Inc. (BEN) are pioneering this effort by optimizing AI solutions to deliver high performance while being scalable and supportable.
BEN's Efficient Language Models (ELMs) exemplify this approach. ELMs concentrate on sectioning and optimization of language models for specialized tasks, contrasting with traditional Large Language Models (LLMs) like those used in OpenAI’s ChatGPT. This specialization significantly reduces the computational and processing power required, allowing ELMs to operate efficiently with less costly and more readily available central processing units (CPUs) instead of the more energy-intensive graphics processing units (GPUs).
Using CPUs provides numerous deployment options, including SaaS, Private Cloud, Mobile, and On-Prem solutions, which are particularly beneficial for industries like Healthcare and Financial Services that prioritize data security. CPUs are also more affordable and abundant compared to GPUs, which face availability issues. BEN's ELMs, augmented with RAFT (Retrieval Augmented Fine-Tuning) systems, ensure applications are reliable, predictable, and efficient, minimizing the risk of AI 'hallucinations'—misleading or false answers generated by AI.
BEN’s approach has gained traction among its growing customer base across various sectors, including healthcare and financial services. The company’s focus on efficient, scalable, and secure AI solutions addresses the critical need to balance technological advancement with sustainable infrastructure and environmental considerations. As AI continues to evolve, BEN’s innovative models demonstrate that it is possible to harness the power of AI while mitigating its impact on the power grid and the environment.
Featured photo by Zosia Szopka on Unsplash.
