What resources are AI2050 fellows getting to address the funding gap?

The Schmidt Sciences AI2050 program offers fellows funding they can use to buy GPUs, which researchers said was a major benefit of participating. However, money remains a pressing concern overall, especially given the reduction of federal scientific funding in the United States.

What kind of research are academics now focusing on instead of frontier model development?

Many are targeting questions unlikely to be addressed by tech companies, such as Anjalie Field's research at Johns Hopkins showing that language models give less sophisticated responses to prompts phrased in ways more commonly used by women than men. Others work on specialized AI models for applications like climate change, or explore how to make models smaller and more efficient.

Have any prominent academics left universities for frontier labs?

Yes, several prominent academics have recently taken leave from their universities to join frontier labs, and many AI2050 fellows hold industry positions alongside their academic jobs.

Large Language ModelsAI Safety & AlignmentAI Business & IndustryMIT Technology Review AIAug 10, 2026Aug 11, 2026, 06:01 JST

Academic AI researchers face funding crunch as frontier labs dominate

Send on LINE

3 Key Points

What happened
University AI researchers gathered at the Schmidt Sciences AI2050 program convening near San Francisco found themselves unable to access the computational resources and proprietary details of frontier AI models developed by companies like OpenAI and Anthropic, forcing a shift in research priorities away from cutting-edge capability work.
Why it matters
As private companies have consolidated control over large language model development, academic institutions—which historically drove AI breakthroughs—lack the GPUs and funding to compete, and face restrictions on studying the design of tools like ChatGPT and Claude. This mirrors what UC Berkeley professor Nika Haghtalab described as being like biologists in a world where companies had exclusive control over gene-editing tools. Many researchers are pivoting to questions companies won't address, such as bias in language models across genders, or to specialized AI models outside the LLM space—but even those scholars struggle when public understanding of 'AI' centers on energy-intensive large language models.
What to watch
Some academics worry that AI systems have begun solving pure mathematics problems, raising concerns about whether mathematicians will have a future in the field. However, researchers like Tim Dettmers at Carnegie Mellon argue AI scientists could instead make human researchers more efficient rather than replace them, and resource constraints may drive academics to discover more efficient model architectures—meaning the next major AI breakthrough could come from a university lab rather than a major company.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Get news like this every morning, free

Context & Analysis

The convening near San Francisco crystallized a structural shift in AI research that has been building for years. Four years of reorientation around large language models has concentrated the cutting edge in private companies—Anthropic, OpenAI, Google—that possess the computational resources and proprietary models that universities cannot match. The comparison Haghtalab drew to exclusive control over CRISPR gene-editing is apt: researchers can observe the outputs of ChatGPT and Claude, but cannot examine their design, training, or internals, nor steer their development. This creates a two-tier system. Academics with funding can buy GPUs through programs like AI2050, but even so, the cost of querying commercial APIs for rigorous study remains prohibitive, and federal scientific funding has contracted, leaving universities under sustained financial pressure.

The response from the research community has been pragmatic and bifurcated. Some, like Field, have explicitly chosen to work on problems that industry will not touch—research that might damage company reputation or lacks profit potential. Others have stepped away from LLMs entirely, focusing on specialized models for climate science, drug discovery, and protein structure prediction (exemplified by DeepMind's disbanded AlphaFold team). A third group has hedged their bets, holding both academic and industry roles. What all three share is an implicit acknowledgment that the frontier of AI capability now lies outside academia. The downstream effect is a quiet crisis of confidence: mathematicians worry that AI systems are beginning to solve problems in their domain, and researchers in other fields struggle to justify their work in an era when 'AI' has become synonymous with energy-intensive large language models.

FAQ

What resources are AI2050 fellows getting to address the funding gap?: The Schmidt Sciences AI2050 program offers fellows funding they can use to buy GPUs, which researchers said was a major benefit of participating. However, money remains a pressing concern overall, especially given the reduction of federal scientific funding in the United States.
What kind of research are academics now focusing on instead of frontier model development?: Many are targeting questions unlikely to be addressed by tech companies, such as Anjalie Field's research at Johns Hopkins showing that language models give less sophisticated responses to prompts phrased in ways more commonly used by women than men. Others work on specialized AI models for applications like climate change, or explore how to make models smaller and more efficient.
Have any prominent academics left universities for frontier labs?: Yes, several prominent academics have recently taken leave from their universities to join frontier labs, and many AI2050 fellows hold industry positions alongside their academic jobs.

MIT Technology Review AIRead Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.