The growing cost of artificial intelligence research means that fewer people can easily access computing power to advance the technology, which scientists warn could shortchange university laboratories and make behemoths like Google, Microsoft, and Facebook dominant. The Allen Institute for Artificial Intelligence estimated that the calculations required to perform AI tasks have ballooned approximately 300,000 times in the last six years. University of Massachusetts, Amherst researchers also found AI software can consume an enormous amount of power. Allen Institute scientists suggest changing the AI success quantification model could help resolve both cost and power issues -- with efficiency as well as accuracy factored in. Large tech firms are focusing on efficiency upgrades, and they argue this will make computing power more available to outside developers and academics.