There is a battle brewing of AI chips from major cloud vendors, as Google has entered preview with its custom chip, Trillium, for the training and running of its AI models, with Microsoft's Maia potentially not far behind.
Not to be outdone, Amazon Web Services has AI chips, too: Trainium, Inferentia, and Graviton. The company is promoting the former by launching a new grant program for AI research.
The new program will be dubbed Build on Trainium, which will give out altogether $110 million in funds to institutions, scientists, and students studying AI. AWS will provide up to $11 million in Trainium credits to universities with which it has strategic partnerships. It will also give individual grants of up to $500,000 to the broader AI research community.
More, AWS claims it is building a "research cluster" comprised of as many as 40,000 Trainium chips, available to research teams and students using self-service reservations.
Gadi Hutt, Annapurna Labs' senior director at AWS, acquired the chipmaking firm AWS acquired in 2015 said that Build on Trainium is aimed at providing researchers with hardware support they need for their work. Grant recipients will also be linked to Trainium education and enablement programs, Hutt furthered.
"AI academic research today is highly bottlenecked by a lack of resources, and as such, the academic sector is falling behind really quickly," added Hutt. "By building on Trainium, AWS is investing in a new wave of AI research guided by leading AI research in universities that will advance the state of generative AI applications, libraries, and optimizations."
Indeed, AI researchers do not have such an enormous infrastructure which giants like tech companies have at their disposal. Meta bought more than 100,000 AI chips for training its main models. Meanwhile, Stanford's Natural Language Processing Group can manage to maintain only 68 GPUs for all of its research.
But not everyone is buying into Amazon's good nature.
"It feels like this is trying to make a generalization on corrupt funding of academic research, "Os Keyes, a PhD candidate at the University of Washington who studies the ethical impact of emerging technologies, told TechCrunch.
With Build on Trainium, AWS would maintain complete decision-making authority over which projects to fund. The process of selecting winners is murky; Hutt would only say that AWS would distribute funding "based on research merit and needs" and "evaluate program success and outcomes."
An AWS spokesperson later clarified that a committee of "AI and application practitioners" will vet the proposals and select those projects most promising and impactful in advancing machine learning science forward.
There is evidence that corporate-funded AI research skews toward more applied work that may potentially be commercially viable than more theoretical topics. The authors found that, contrasted with the normal research, top AI organizations produce much less work that is critical of the values and principles of AI. The "responsible AI" research that big firms do is also more constricted in scope, say the co-authors, and lacks variety in the issues covered.
Researchers insist on judicial and technical protection to analyze AI without fear of account cancellation or threats of lawsuits by the vendors.
Build with Trainium is seeking, obviously, Trainium. Is AWS' other edge an attempt to get researchers onto its platform? I asked if accepting an award would "lock in" a grant recipient to the AWS ecosystem or to Trainium. Hutt said it won't. "Really our only obligations they'll have are to publish a paper and open source their results on GitHub under a permissive license.".
"There is no contractual lock that makes universities exclusive technology partners," he said. "What we ask in return is that the outcomes of the research will be open sourced for the benefit of the community."
In any case, it's not clear Build with Trainium will do much to bridge the gap between AI academia and industry.
In 2021, US government agencies, excepting the department of Defense, budgeted 1.5 billion dollars for academic funding for AI research. During the same year, AI industry worldwide spent more than $340 billion overall, not just research money.
Nearly 70% of graduates with a PhD in AI find employment with industry, not just because of the salariesthe most competitive ones-but also because of access to compute and data-and the means by which to process it. Indeed, over the past few years companies have really stepped up poaching faculty AI researchers, setting aside larger grants for PhD students doing their research.
The end result? The largest AI models developed in any given year now come from industry more than 90 percent of the time, and the number of AI papers published with industry coauthors nearly doubles each decade since 2000.
That narrowed the funding gap between academia and industry somewhat. The National Science Foundation recently agreed to invest $140 million in support of seven university-led National AI Research Institutes dedicated to how AI can be used to mitigate climate change impacts and enrich learning. Other initiatives include the U.S. National AI Research Resource, a $2.6 billion effort that'd grant access to computational resources and datasets for AI researchers and students.
These are still minuscule compared to the corporate programs, and little reason exists to believe that the status quo shall ever change.