Over 80% of AI Projects Fail Due to High Costs and Data Issues, Driving Shift to Smaller Solutions

The growing investment in artificial intelligence (AI) by Silicon Valley and global tech firms has yet to translate into tangible returns, prompting calls for a shift toward “small AI” solutions tailored to specific business needs. While large-scale, generalized models like GPT-4 or Llama 3 dominate headlines, their exorbitant costs and unclear productivity benefits raise questions about their viability for most enterprises. The industry’s focus on massive datasets, compute power, and broad applicability has led to underwhelming outcomes, with over 80% of AI projects failing due to data quality and operational inefficiencies [1].

Small AI, by contrast, emphasizes targeted models designed to solve singular problems. These systems require fewer parameters, reduced data volumes, and lower computational resources, making them more accessible and cost-effective. Innovators such as Bonsai Robotics have demonstrated success by optimizing AI for niche applications, such as precision agriculture, where specialized algorithms enhance automation in challenging environments [1]. Similarly, Microsoft’s integration of OpenAI’s GPT-based technology into its Copilot tools exemplifies how focused models can deliver practical value through software-specific optimizations [1].

The advantages of small AI extend beyond cost. Smaller models can run faster, consume less energy, and produce more reliable results in controlled scenarios. For instance, retrieval-augmented systems or fine-tuned models can address specific tasks—like summarizing documents or automating coding—without the overhead of generalist frameworks. This approach aligns with the reality that most businesses prioritize incremental, measurable improvements over speculative, large-scale deployments [1].

Critics argue that small AI does not diminish the importance of foundational research but rather realigns priorities to prioritize sustainability and scalability. By narrowing focus, companies can avoid the pitfalls of overengineering and instead allocate resources to high-impact, well-defined use cases. This strategy also mitigates risks associated with data privacy and environmental costs, as smaller models inherently require less energy and generate fewer ethical concerns [1].

Industry experts suggest that the hype surrounding AI should not obscure the need for pragmatic innovation. While OpenAI and others project ambitious futures—such as personalized AI agents operating on individual data—the immediate value lies in refining existing tools to deliver tangible returns [1]. The shift toward small AI mirrors historical patterns in tech, where breakthroughs often stem from incremental advancements rather than singular, monolithic solutions.

As investors seek clarity on AI’s ROI, the industry’s trajectory may hinge on balancing ambition with practicality. Small AI offers a path to achieving both: cutting costs, enhancing efficiency, and delivering measurable outcomes. For now, the message is clear—until large models prove their mettle in real-world productivity, businesses would do well to think smaller [1].

Source: [1] Silicon Valley’s billions of dollars on AI haven’t actually generated a return yet. Here’s why most companies should embrace ‘small AI’ instead (https://fortune.com/2025/07/30/what-is-artificial-intelligence-return-on-investment-small-ai/)