They're citing a machine learning model that tries to correlate various other measures (like GDP/capita, unemployment, etc.) with mortality in some countries, and then extrapolates to every country. That model spits out ridiculous numbers for China. A normal statistician would look at that and say, "This model doesn't generalize well," not, "1.7 million people died without leaving a trace."
The best actual measurement of excess mortality in China was published in the BMJ (formerly known as the British Medical Journal).[0] It found that in the initial outbreak in early 2020, there were just under 5,000 excess pneumonia deaths. Almost all of them were in Wuhan itself. In most of China, there were actually negative excess deaths, because the lockdowns reduced flu and car crash deaths.