First of all, the ~70% figure is for how many people that Bloomberg estimated cast votes. The 143 million figure is the number of votes that had already been counted as of the time of this question (it's up to 148 million as of the time of this answer.) From the Bloomberg article:
Bloomberg’s model anticipates the total number of votes cast for president to range from roughly 157.1 million to 165.0 million (68.6%–72.1% of the citizen voting-age population).
So, the 70% number is referring to approximately 160 million voters, not 143 million. This would give a total electorate size of roughly 229 million. Of course, they also state in this same quote exactly how they're getting the total electorate size - it is indeed total voting-age population.
Yet it doesn't seem right that nearly a third of the population of the united states is either not a citizen, under 18, or currently in prison!
Indeed, it is true (just from those first two alone.) Children alone account for the vast majority of this. According to the U.S. Census Bureau, 22.3% of the U.S. population is under the age of 18. They estimate the U.S. population as of July 2019 as 328,239,523, so approximately 73.2 million are ineligible to vote because they are too young.
The Census also indicates (in the same source as above) that 13.5% of U.S. residents were foreign-born (which they define as anyone who is not a U.S. citizen at birth) between 2014 and 2018. More usefully for purposes of this question, according to this table - available for download from the Census' website, the Census estimated that, as of 2019, there were approximately 21,449,000 non-citizens over 18 years of age residing in the U.S. This represents another 6.5% of the total U.S. population.
So, the observed discrepancy between total population and electorate is explained simply due to them being under 18 (22.3%) or not citizens of the U.S. (6.5%,) totaling 28.8% of the overall population just from these two factors. According to the table linked above, the Census estimated that there were 229.1 million American citizens over age 18 as of 2019, which indeed does match Bloomberg's estimate of the total electorate size.
While this is not affecting the electorate size numbers in the article being asked about (since it explicitly states that it's specifically using voting-age citizen population,) it's true that disenfranchisement due to felony convictions is significant, though this will be manifested in statistics as a reduction in voter turnout, not as a reduction in the size of the electorate.
The advocacy group Sentencing Project (which advocates for restoration of voting rights for convicted felons) estimated that, for the 2020 elections, 5.2 million people were ineligible to vote due to felony convictions. This is approximately 1.6% of the current U.S. population. However, it does not seem clear from their methodology section whether or not they accounted for the portion of convicted felons who wouldn't be eligible to vote anyway due to either not being a citizen or being under age 18. I don't see any discussion of age or citizenship status in their methodology section, though I haven't examined their spreadsheets. Since not all felons would vote even if they could and it's not clear whether children and non-citizens are already being excluded from these numbers, it seems likely that the actual effect this has is probably reducing turnout by somewhere in the ballpark of 1% to 1.5%.