Small AI Models Gain Traction Around the World

Small artificial intelligence models are gaining traction globally, delivering services in regions with limited computing power, electricity, and data, according to a report from IEEE Spectrum. These compact systems can run on low-power devices like phones and microcontrollers, performing tasks from counterfeit drug detection to crop disease identification. The push for small AI was crystallized for Adebayo Alonge, founder of the pharmaceutical verification company RxAll, during a 2019 demonstration in Cape Town. His handheld spectrometer, designed to authenticate medications, connected to a data center 14,000 kilometers away. Bandwidth limitations meant a single scan took over five minutes to return a result [1]. "I was shocked," Alonge said. His engineers then shrunk the AI model to run entirely on an Android phone, producing a working version in two hours [1]. This pivot birthed a device that can authenticate pills without broadband or reliable electricity, and turned Alonge into an advocate for the technology [1]. World Bank President Ajay Banga highlighted the disparity in AI access at the World Economic Forum in Davos. "Most people are discussing AI from the LLM/generative side. But that needs a lot of computing power, electricity, massive data, and skilled people to manage it," Banga said [1]. A November World Bank report noted that only 0.7 percent of internet users in the world's poorest countries have used ChatGPT, compared to a quarter of all internet users in the most developed nations [1]. Small AI models are often created by pruning large models, removing parameters irrelevant to a specific task, or through distillation, where a compact model is trained to mimic a larger one [1]. Marcelo José Rovai, a professor at the Federal University of Itajubá in Brazil, demonstrated a language model running on a US $50 Arduino UNO-Q device that analyzes sensor data to detect mosquito breeding pools, consuming just 3 watts of power [1]. "This is the most important area in AI nowadays. It's growing very fast," Rovai said [1]. These systems stand in contrast to the broader AI landscape, where the proliferation of low-quality generative content has led to the coining of the term "AI slop," selected as the 2025 Word of the Year by Merriam-Webster and the American Dialect Society [2]. The phenomenon describes digital clutter prioritizing speed and quantity over substance, a form of synthetic media linked to monetization in the creator economy [2]. Commentators have connected the rise of such content to aspects of the dead Internet theory, which posits that bot activity and automated content are overwhelming human-authored material online [4]. While small AI sidesteps the infrastructure demands of hyperscale models, Alonge cautioned that it does not eliminate the need for foundational development. "You still want to be able to enable periodic syncing for updates with new signatures for the medications and analytics," he said. "And even when you are using batteries, reliable power is important. That phone battery is not going to last forever" [1]. The technology research firm Counterpoint estimates that 45 percent of smartphones shipped worldwide will be capable of running generative AI by the end of this year [1].