During the Ballmer era, Microsoft wasted billions on mobile via acquisitions of Danger and Nokia and their internal efforts to make Windows Mobile a thing. I’m sure they could have found the money from somewhere.
Bing by itself reportedly doesn't even gross that much, overall $20 billion represents about a quarter of Microsoft's entire annual profit. Microsoft already decided it wasn't viable to spend that much to compete, and the rest of the search market (including AI) need not apply.
They have the money to compete and jumpstart Bing with default placements and reap the ad dollars and build Bing into a serious competitor.
If they don't want to compete because they think investing money in Xbox will have a higher return, that's their decision (and maybe their mistake). It's not Google's fault.
What is missing in this discussion is the fact -- explicitly called out by the court in the opinion, see e.g. https://news.ycombinator.com/item?id=45109999 -- that Google Search is so good because it gets so much search traffic and, critically, user interaction data such as clicks, dwell time and even hovers on search results, which it mines to figure out better rankings.
Unless competitors get that kind of traffic AND user behavior insights, their results will always be worse.
And as long as their results are worse, 1) their revenues will always be worse, which will 2) make it prohibitively expensive to even try to bid for such placements, which in any case 3) would be shot down by Apple because their results are not "good enough".
It's a Catch-22 from which the only escape is making a risky 20-billion-per-year traffic acquisition bet (on top of the billions already being invested) that they can get all that traffic and user behavior data and improve their search engine quickly enough to make the results good enough to drive enough revenue, all the while fighting the tendency of people to use Google anyway simply out of habit.
I don't think it's much of a choice.
The proposed remedies do talk about sharing search and user interaction data though, so if that survives appeals, it might help level the field a bit.