You’re comparing apples and oranges. They’re both fruit, and both tasty, but they’re not the same.
Your graph is about someone puts together a project, bids at a certain price, they win; that gets a point on the graph, dated on the day of the auction.
IRENA is calculating the unsubsidized price of commissioned plants. Someone builds a power plant, then the day it’s completed IRENA calculated how much it costs, adding in the subsidy.
Some projects will take a few years to get financing, permits, and construction done. They’ll come in your graph in, say, 2013 at $0.05/kWh. Four years later the project is actually commissioned. It comes into IRENA in 2017 at $0.07/kWh (calculating out the 30% ITC).
Another source of disparity: you’re focused on the best winning auction bids in the US. It’s reporting more about the whole range globally. IRENA recommends competitive auctions as a way to get prices to drop faster, based on its data about both auction prices and non-competitive projects (where a utility decides to build a certain plant in a certain place).
IRENA’s report predicts much lower by 2020 because it’s kerping track of the recent auction prices. Presumably, auction prices in 2020 will still be a fair bit below commissioned prices in 2020.