How academic science is evolving in the era of AI giants' dominance

Academic circles in artificial intelligence are experiencing an identity crisis. Over the past four years, leadership in developing cutting-edge models has shifted from universities to private companies like OpenAI, Anthropic, and Google. The reason is simple: training and operating large language models require massive computational power that even top-tier universities cannot afford. Moreover, corporations do not disclose the architectural details of their systems, leaving academics with no choice but to study model behavior from the outside.

Financial constraints exacerbate the situation. Even programs like AI2050, which provide grants for GPU purchases, fail to fully address the issue. Many researchers must pay for access to commercial model APIs, making large-scale experiments unaffordable for university budgets. As a result, scientists increasingly focus on topics that hold little commercial interest—such as examining gender or social biases in model responses, as seen in Professor Anjalie Field's research at Johns Hopkins University.

Not all academics work with language models. Many develop specialized AI tools for science, climate, or medicine. However, their work remains overshadowed by the popularity of LLMs, making it difficult to attract attention and funding. The situation is further complicated by cases where successful projects, like Google DeepMind's AlphaFold, may be shut down for corporate reasons. These challenges are pushing scientists toward industry careers or combining academic and commercial work.

Despite these obstacles, experts see positive scenarios emerging. For instance, AI could accelerate scientific discoveries by freeing researchers from routine tasks. Additionally, resource limitations are driving innovation—from model optimization to the development of entirely new architectures. Perhaps the next breakthrough in AI will come not from a corporate lab, but from a university classroom.