Investors pressure specialized language AI startups to shut down when Big Tech announces models covering their target languages, according to Timnit Gebru, AI ethics researcher at the AI Now Institute. "OpenAI or Meta or something comes with an announcement of a big model, a number of potential investors in these smaller organizations literally told them to close up shop," Gebru said.
The dynamic threatens organizations building domain-specific language models for particular regions. Big Tech's compute-intensive approach concentrates AI development in areas with cheap energy and capital access, leaving startups focused on minority language populations vulnerable to investor flight.
DeepSeek's frugal approach demonstrates alternatives to Big Tech's data-intensive scaling. The Chinese company achieved competitive performance using significantly less computing power than Western counterparts, challenging assumptions that bigger models always perform better.
Specialized AI applications continue delivering value without massive infrastructure across sectors. Ameriabank automated 96% of loan underwriting decisions. Pelican Canada has processed over one billion transactions across 55 countries using AI-driven payment systems developed over 25 years, handling various payment types and global banking standards.
The startup shutdowns reveal that investor perception of AI value remains tied to scale rather than specialization. Organizations building models for specific languages, industries, or applications struggle for funding against the narrative favoring frontier models, despite evidence from banking and healthcare showing targeted systems achieve high accuracy on defined tasks.
Gebru argues the dominant paradigm involves "stealing data, killing the environment, exploiting labor" in pursuit of building what she calls a "machine god." The resource divide shapes AI's geographic distribution, determining which languages and communities receive effective AI tools.