Universities have largely stepped aside in the race to build frontier artificial intelligence systems, leaving the field to industry while struggling to prepare students for an AI-transformed world. More than 90 percent of notable AI models released in 2025 came from industry rather than academia, including GPT-5, Gemini 3, Claude Opus 4.5, Grok 4, Llama 4, DeepSeek-V3.2, Qwen 3, and Kimi K2. Meanwhile, the computing infrastructure supporting cutting-edge AI development has become almost entirely dominated by private companies and hyperscalers, creating a widening gap between what universities can offer and what the AI frontier demands.

Why Is the Compute Gap So Dramatic?

The disparity in computational resources tells the story most clearly. Global AI compute capacity has been growing at 3.3 times annually since 2022, doubling every seven months, driven by hyperscalers and enormous data-center investments. To put this in perspective, ten New York universities spent two years and $340 million to build approximately 400 graphics processing units (GPUs), which are specialized chips essential for training AI models. In contrast, xAI in Memphis stood up 100,000 GPUs in just 122 days during 2024, and a site in Abilene, Texas reached 500,000 GPUs in the past two years. This gap reflects a fundamental mismatch between university budgets and the scale required for frontier AI research.

Even in China, where universities have historically played a larger role in national technology development, the pattern is similar. DeepSeek, Alibaba, ByteDance, Moonshot, and other major model developers are working outside of universities. However, Chinese universities appear to be more strategically aligned with national AI goals than their American counterparts. A Hoover Institution analysis of 356 researchers appearing on DeepSeek's foundational papers found that 53.5 percent of those with known affiliations had spent their entire recorded careers at Chinese institutions. Of DeepSeek researchers who spent part of their careers abroad, over 70 percent ultimately returned to China, suggesting that universities are part of how China built a domestic frontier-research workforce.

What Role Remains for Academic AI Research?

This does not mean universities have become irrelevant to AI progress. Computer science and engineering departments continue to supply enabling technology, evaluation systems, and trained researchers. Important breakthroughs have emerged from academia, including Berkeley's vLLM PagedAttention, Stanford's FlashAttention (now a foundational technique for efficient transformer computation), and Berkeley's LMSYS-created Chatbot Arena, a major open evaluation platform. These contributions matter significantly for the field's infrastructure.

The number of new AI PhDs awarded in the United States and Canada rose 22 percent between 2022 and 2024. However, the distribution of these graduates has shifted. Industry's share of new AI PhDs fell to 62.75 percent from a 77 percent peak in 2022, while academia's share rose to 31.59 percent. This suggests that universities are retaining more PhD talent in academic positions, even as corporate labs dominate the production of frontier models. Without university research, AI progress would undoubtedly slow, but the highest-value work has moved down the technical stack, away from the flashy model-building that captures public attention.

How Universities Can Prepare Students for an AI-Transformed Future

• Broaden AI literacy across disciplines: The AI frontier extends far beyond computer science and philosophy. Historians, linguists, biologists, artists, and every person entering business, culture, and government need to understand the capabilities and limitations of AI systems as much as computer scientists do.
• Invest in open-source model infrastructure: More firms are using open-source AI models, with 63 percent of organizations running an open model in production and 72 percent of technology companies doing so, according to a McKinsey survey of over 700 global technology leaders. Universities can leverage cheaper, customizable, locally deployable models to teach students practical skills.
• Create institutional alignment with AI advancement: Individual voices like Ethan Mollick at Penn, Tyler Cowen at George Mason University, Scott Latham at UMass Lowell, and Michael Madison at Pitt have begun shifting their institutions' approach to AI education and research. Institutional leadership matters for scaling these efforts.

The challenge facing higher education is not a lack of awareness but a lack of institutional urgency. Many university leaders remain focused on traditional battles over curriculum, viewpoint diversity, and federal funding for science, leaving little bandwidth for the transformational changes AI demands. One AI-savvy provost explained to the author that battles to keep the lights on amid federal funding changes for science are preventing her from making the institutional changes she wants. This is a real constraint, but it also reflects a priority mismatch in a moment when AI is reshaping how knowledge is accessed and transmitted.

Universities have no direct influence over the pace of frontier AI model development, which is controlled by industry. However, they retain a critical role as institutions responsible for forming the next generation of young people and preparing them for a world transformed by AI. The question is whether they will embrace that role or continue treating AI as a peripheral concern, leaving students to learn the realities of AI capabilities and limitations through trial and error in the real world.