Amazon’s Build on Trainium Program Puts $110M Into AI Research

Amazon has launched a new initiative called “Build on Trainium,” which promises to power a new wave of AI research with grants worth $110 million.

The battle for AI – round next – is clearly on the horizon, as the tech giants take control of hardware that will fuel next-generation AI models, racing against Google, Microsoft, and others in a chase to keep up with the rising tide of AI chips. Amazon offers up to $11 million in credits for universities and individual grants up to $500,000 for researches and students. But this program fuels criticism that corporate money is hijacking the very nature of academic research.

Core to the program, in fact are the Trainium silicon chips designed specifically by AWS for training and running large-scale AI models. In addition to the grant money, AWS has reserved a gargantuan “research cluster” of up to 40,000 Trainium chips for self-managed reservations. This puts researchers firmly into the camp of those with access to high-performance computing capabilities usually beyond the scope of any academic institution.

Gadi Hutt, a senior director at AWS’s Annapurna Labs – the company’s chip-making arm – said the initiative was an effort to overcome the ‘bottleneck’ that had been hamstringing AI research these days by means of a resource severely lacking: computational power. “Academic research in AI is terribly bottlenecked by a lack of resources,” said Hutt. “With Build on Trainium, we are trying to arm these researchers with the hardware they need to innovate.”

This program might indeed partially relieve some of the pressure on academic AI research that most institutions hardly cope with due to outdated or limited resources. For example, whereas Meta is in a position to roll out more than 100,000 AI chips to train its models, Stanford’s Natural Language Processing group is rather reliant on 68 GPUs for its work.

However, the move has critics frowning and worried that it might upset the academic balance concerning the matter of freedom. Os Keyes is a PhD student from the University of Washington-hence engaged in researching ethics associated with the emerging technologies-stated, “This is an effort to commercialize and co-opt academic research funding,” according to TechCrunch.

The program has created uncertainty because AWS holds final say on which research projects will receive funding. While AWS claims that selections will be based on “research merit and needs,” the process is opaque. An AWS spokesperson later clarified that a panel of AI and application experts would evaluate proposals to select the “most impactful and promising projects.”

This gives concern over industry funded AI research, which typically only looks at applications in which there is a higher commercial potential rather than maybe greater in ethical or social need. In fact, a recent report showed that AI companies have published appreciably fewer papers describing the ethical and social dimensions of AI than those done by the academic community and that their research often tends to concentrate on narrowly defined topics from which commercial exploitation is more likely to arise. This is a bad situation for many who still maintain that the scholarly world must remain an independent voice, untouched by the interests of industry.

Another question around AWS’s program is whether accepting a Trainium grant would lock recipients into the AWS ecosystem. Gadi Hutt assured TechCrunch that there would be no such obligation. “There is no contractual lock that makes universities exclusive technology partners,” he said. The only requirements for grant recipients are that they publish their research findings and open-source their work on GitHub under a permissive license, meaning the broader AI community can benefit from the results.

Despite the promises, others worry that the main mission of the program could be to nudge researchers onto its cloud platform rather than on the rival cloud services. AWS has garnered notoriety for marketing its cloud infrastructure to developers and academic institutions, and critics say programs like Build on Trainium can only solidify the giant’s stranglehold on the research ecosystem.

These concerns are further compounded by a more general imbalance between industry and academia in AI. In 2021, the U.S. government invested $1.5 billion into financing AI research through academia. In contrast, more than $340 billion was invested by the global AI industry. This imbalance is also reflected in the numbers: nearly 70% of AI holders with a PhD are employed by the private sector. To that extent, over 90% of the largest AI models are now developed by industry, and the number of academic papers co-authored with private companies has nearly doubled since two decades ago.

Attempts are underway to fill the gap. For example, the National Science Foundation recently committed $140 million to establish seven university-led institutes for AI research. The US government is also constructing a National AI Research Resource as part of a $2.6 billion initiative that will give researchers access to supercomputing power and data. However, these remain small compared to how much corporations are spending on AI research.

Influencers like AWS’s “Build on Trainium” will catalyze this move; corporate-funded research will further drive institutions of learning into the orbit of big tech. On one hand, it may fuel AI innovation in some sense, but critics feel that this cocktail of issues created is more a fueling of the chasm between the commercial nature of the interests of big tech and the mission-driven nature of academic research.

source techcrunch