Headline should read, DeepMind calculated 2.2 million hypothetically stable crystal structures with the possibility to advance everything from materials science to biology (emphasis mine)

There's a large difference between a hypothetical crystal structure and being able to produce that crystal structure in the real world. And an even larger difference between the hypothetical and the ability to produce the crystal structure in an efficient and cost-effective enough manner to be usable. There are already multiple materials that are known that would make better batteries than lithium-ion batteries, but either the manufacturing method is too hard or expensive, or the materials are extremely rare. So it's nice DeepMind was able to do this, but it's years from being usable still.